Changelog

The Law Evaluation lens now shows a numerical confidence score (0–100%) for each potential violation, plus clearer visual badges and stricter filtering to reduce false positives.

• Confidence scores — every violation displays a percentage (e.g., 87% confidence) with color-coded badges: green for high confidence (≥85%), amber for medium (75–84%), and rose for borderline (below 75%).
• Stricter 75% threshold — violations below 75% confidence are automatically filtered out and noted in processing notes for transparency.
• Line citation enforcement — message examples must include a “Line N” citation to be clickable; examples without citations are filtered out.
• UI polish — confidence badges animate on appearance, violation cards have subtle hover effects, and dark mode styling is more consistent.

These changes make legal assessments more reliable and help you focus on higher-confidence concerns. As always, the Law Evaluation lens is for reflection and possible attorney referral—not legal advice.

Quoted message examples in the Schema lens and DSM-5-TR lens now start collapsed (showing the first 4 examples) so the report is easier to scan. Click Show all to expand and view the complete set of supporting evidence for each pattern.

The background panels behind quoted message examples in the Schema and DSM-5-TR lenses now use a slightly transparent fill (70% opacity) so they blend more naturally with the report surface while keeping the blue (What to Say) and red (What to Avoid) accent colors readable.

In the Messages sidebar, clicking Send invite in Practicefor a client without portal access now reliably opens the client's Overview tab (where the portal invite panel lives) and scrolls directly to the invite section.

Bible, Pastor's Lens, Law, DSM-5-TR (including supporting evidence quotes), and DSM surrounding-message context no longer show the small slate Line N badges above each excerpt. Detected Patterns example cards no longer repeat the line number in the small meta row above each quote (time and sender stay when known). Where line linking is available, use the quoted text (or, in surrounding context, the message body): the same scroll-to-line behavior is preserved, and screen readers still hear the line number on DSM context rows.

The report header sits in a softer framed band, a new At a glance strip surfaces severity, confidence, visible patterns, and immediate concern in compact tiles, and Quick summary is elevated as a narrative “next steps” card so the page reads top-down from outcomes to detail. With Patterns only toggled on, the strip is hidden so Detected Patterns stays visually primary, and the report hero compacts (shorter copy, smaller title, slimmer glossary row) so pattern cards come up sooner. Detected Patterns has a stable in-page anchor (#safety-detected-patterns) with scroll margin; the patterns-only blurb links there, and opening the report with that hash expands the section and scrolls it into view.

In the Schema lens and DSM-5-TR lens, quoted message examples now use the same blue left-border panel as What to Say and the same red panel as What to Avoidin Immediate Guidance—so highlighted transcript lines feel consistent across the report. Bible, Pastor's Lens, and Law message examples keep their existing highlight style.

Visiting reflectwithemily.com or www.reflectwithemily.com now loads the app on that hostname instead of permanently redirecting to another apex. That keeps sign-in, cookies, and the first load on one origin so pages are less likely to hang on a spinner after a domain hop. Production nginx now proxies this host to the same frontend and API as the main site (Next.js legacy host redirects are opt-in at build time only). If a redirect still appears in the browser, check Cloudflare for an extra bulk redirect rule. Deployments that still need a marketing-domain redirect can opt in with documented environment variables.

When analysis notes mention automated guardrails, omitted projected-impact text, or capped severities, the report now shows a short how to proceed panel: continue with the grounded report (scrolls to Quick summary), read the full processing notes, jump to Detected Patterns, or jump to Immediate guidance. After you continue, a slim reminder stays available with links to reopen the choices or open notes.

In Patterns only view, a compact pipeline notice still appears under the severity summary with shortcuts to reopen the full choice panel or jump straight to Analysis & processing notes.

If the server answers with a plain “Internal Server Error” page (common when a gateway times out or the app is still busy), the app now shows guidance that fits Safety Check and other heavy tasks—not only multi-model chat. Retrying after a short wait, sending less at once, or splitting a long Safety Check in the wizard is called out when errors keep happening.

When participant preview hits an HTML error page or an invalid response, the message now matches the rest of the app: retry shortly, avoid overlapping heavy analyses, and use the wizard to split very long threads or send a shorter excerpt when previews keep failing.

Generic thread labels (for example You, Me, Them, Other, Person 1, Person 2, and transcript styles like Person 1 (sent) / Person 2 (received) / This person) no longer show a misleading single-letter initial in avatar circles. Names still replace those placeholders when your import includes real participant labels, consistent with how the pipeline treats generic transcript names elsewhere.

Error text for long threads, empty parses, and preview failures now points to the same step-by-step wizard path (/dashboard/safety-check/wizard) for Split into parts. Expected issues (for example paste format or over the per-part character limit) no longer open automatic support tickets. Long streaming runs are less likely to hit a noisy “response already started” server error when the database closes an idle connection.

When you paste a long conversation, you now see real-time feedback: message count, character count, and estimated parts needed. If it exceeds the single-run limit (~12,000 characters), a warning appears with a direct link to open the step-by-step wizard for splitting. This helps avoid the timeout error and guides you to the right workflow before you submit.

The Upload File option now accepts audio files (MP3, M4A, WAV, etc.). Audio is transcribed automatically using Whisper and then analyzed like any other conversation. Useful for reviewing recorded phone calls, voice messages, or any spoken exchange you want to check for patterns.

New privacy controls when sharing Safety Check reports let you decide exactly what recipients can see.

• Hide victim messages — Remove all messages from the victim/survivor (Me/You) so only the other party's messages appear in the shared report.
• Disable printing — Prevent recipients from printing or downloading the report as a PDF.
• Selective line redaction — Remove specific lines entirely from the shared conversation, useful for removing sensitive details, names, or locations.

The Upload File option now accepts audio files (MP3, M4A, WAV, etc.). Audio is transcribed automatically using Whisper and then analyzed like any other conversation. Useful for reviewing recorded phone calls, voice messages, or any spoken exchange you want to check for patterns.

When the evidence guardrail removes ungrounded examples, you can now review and correct them. For each flagged example, choose: fix the line number, fix the speaker attribution, confirm it is accurate (paraphrase), or remove it. Corrections are saved and applied to update your report with verified examples. This helps improve report quality when transcript parsing or speaker labels cause grounding mismatches.

Lens Message examples(Bible, Pastor's Lens, Law, DSM-5-TR, Schema) and supporting DSM snippets now use the same pink highlight and soft glow as pattern-card quotes. Pattern cards also add a light outer glow on each cited line row for consistency.

The in-progress analysis screen no longer says that no conversation content is stored on our servers. Copy now states that analysis runs for your account and that your report and the thread used for it may be stored for reopen, history, export, or sharing. Multi-provider flows use similar wording about possible retention of submissions and results.

When a message is not a pattern-highlight row, common Markdown noise (headings, **bold**, simple list bullets) is cleaned up for display so verses and notes pasted from chat match how you read them elsewhere. Rows that use phrase highlighting still show the stored text so matches stay aligned. Self-labeled Me no longer shows a redundant M in the avatar circle (same as You).

Pastor's Lens message examples no longer repeat the same quoted line when several Bible categories cited it; the list explains that quotes match Bible evaluation sources and are shown once per distinct line.

When the report lists detected patterns, the Projected impact block is kept only if the trajectory text actually names at least one of those pattern titles (or the main part of a compound title before an arrow). Generic five-field essays that never say which pattern they refer to are dropped, with a short note in analysis confidence; analyst instructions spell out the same rule.

Legal concern rows no longer show model artifacts like HarassmentIowa when the jurisdiction belongs in its own field, and law Message examples drop the redundant Line N (timestamp): wrapper so the quote matches what you read in the thread.

Pattern example cards no longer repeat Line N when the timestamp row already shows it, and quoted snippets fix double-quote-for-apostrophe glitches so possessives (e.g. Emily's mom) read naturally.

On classic results and on shared report pages (token link and one-time verify link), Reading mode hides the large severity gauge in favor of a one-line severity and confidence summary, starts Risk assessment collapsed, and opens the DSM-5-TR lens in matched-criteria-only view. The toggle is saved on this device (local storage) and stays in sync across those views.

Names typed for Party 1 / Party 2 on the classic form are now used when the participant modal opens (instead of resetting to the last preview-only labels). The "Name or label for this person" field also stays aligned when it still matched the other party's detected name and that name is corrected.

Analyst instructions now require each pattern's description to walk through a plain definition, which specific lines or behaviors met the bar for that pattern, and why the severity sits where it does. Risk assessment reasoning is guided the same way—so escalation and physical-risk settings are tied to what actually appears in the conversation. New runs benefit automatically; the judge also prefers reports that make this evidence-to-criteria link clear for non-experts.

DARVO, gaslighting, reactive abuse, and related terms now open in a dedicated screen glossary overlay (large type for projectors and shared screens). The dense link list was removed from inside the DSM lens card; official references stay in a collapsed section inside that overlay, and the Help Center glossary remains for full reading.

If all five trajectory blurbs read like the same generic essay (lots of “may / further / distress / escalate”) and never name your actual detected patterns, the report now omits that block and notes why in analysis confidence text—so you are not shown copy-paste filler. New checks get this automatically; the analyst prompt also calls out that shape as forbidden.

On the classic Safety Check results view, quote highlights use softer tints instead of full-saturation pink, pattern example panels and role chips use clearer light-mode contrast, the severity hero is a little less heavy typographically, and the page column is slightly wider with a gentler disclaimer callout so long reports are less tiring to read.

When emailing a public report link, you can choose a different confirmation email (the inbox that receives the link you must click before we mail the recipient). If you leave it blank, we still use your account email as before. The API also returns a clear validation error when no confirmation address is available.

GilApps-style and similar exports combine a date band with local times; those were previously stored as if they were UTC, so the full conversation panel showed the wrong calendar day or clock time. Parsed rows are now wall-clock values serialized without a misleading Z suffix, and the dashboard formats thread times so browsers interpret that shape as local. True UTC instants (suffix Z) still display correctly. The same thread timestamp rules apply to AI Chat safety-check upload responses and Bible Evaluation uploads so those flows stay consistent with the main pipeline. Exports, example line prefixes, and coach / fight-room context use the same local-friendly formatting for thread timestamps instead of raw ISO strings. Additional plain-text export date-time strings from the flexible parser are also treated as wall-clock (not forced into UTC) so they line up with the main transcript behavior. JSON and CSVexports that use non-ISO date strings now fall back to the same flexible parser instead of silently using “now” as the message time.

Mode-style hintsunder the schema lens no longer use the YSQ-R disclaimer ( “Based on this questionnaire…” / YSQ-R profile wording). That text applied only to real questionnaire scoring. Hints from detected patterns now say explicitly they are not YSQ-R responses and not a separately administered SMI.

The main report pipeline now forces pattern participant to match role_labeland the assigned victim/abuser sides when the model mixed them up—so abuser-tagged patterns no longer show as “Exhibited by: Me” while the analysis names the other party. The DSM-5-TR lens uses the same rule when resolving exhibitor. Reactive abusetext that incorrectly said the same person's “responses are reactive to” that same person's behavior is rewritten to name the reader reacting to the other party. Severity scores are capped when Evidence support is only low or medium (thin Line N spread), matching the badge logic. Analyst guidance excludes tagging Emotional Withholding for brief polite or spiritual closings alone (e.g. prayer wishes without a punitive withdrawal pattern). The ensemble judge is instructed to reject reactive-assessment copy-paste errors. The Analyst transcripts—the main generate_report thread and multi-perspective specialists—use the same shared speaker resolution as the Law and Bible lenses. Reactive-abuse feature extraction buckets messages with the same rules (e.g. Human / other roles) so classifier inputs match the transcript.

When a thread includes an embedded portrait for only one speaker, the other column no longer shows an amber “Initial fallback used” pill (it read like an error). A green Portrait from thread badge appears only for sides that actually have an inline image; initial-letter avatars stay as the default with no extra warning state.

When operators set DSM5_TR_THEME_EXPLORATION=true, the DSM-5-TR lens can add a short Optional: broader DSM chapter themes (AI) panel: reflection bullets aimed at mood, anxiety, trauma/stressor, substance, or related non–personality-disorder chapters—written as clinician prompts, not diagnoses or criterion codes. Default remains off (no extra latency/cost). The structured personality-disorder crosswalk is unchanged. When that panel is present, text, HTML, and CSV Safety Check exports include the same disclaimer, bullets, and optional clinical redirect.

The single-conversation DSM-5-TR lens now crosswalks to official criterion rows for all ten DSM-5-TR personality disorders in Clusters A, B, and C: paranoid, schizoid, schizotypal, narcissistic, borderline, antisocial, histrionic, avoidant, dependent, and obsessive-compulsive—with disorder-specific numeric thresholds in the UI. Help, export text, and lens picker copy describe this scope; mood, anxiety, and other chapters outside personality disorders are not crosswalked. Many Cluster A/C criteria use conservative pattern mapping so the lens stays excerpt-grounded.

The Law evaluation lens now uses tighter guidance so ordinary dispute, church-process, or accountability language is not treated as criminal harassment by default. After the model responds, the server drops any example whose Line N does not point to a message actually sent by the evaluated party—so citations cannot silently attach to the wrong speaker. If nothing valid remains, the concern row is removed and summaries reflect that attribution check. When no violation rows remain, overall, victim impact, and liability summaries are normalized so they cannot overstate concerns while the list is empty. Speaker resolution for each line also accepts common alternate roles (e.g. Human/other) and matches explicit per-message labels when role is missing.

Bible evaluation now uses the same shared thread speaker logic as the Law lens (alternate roles and label fallback), formats the transcript consistently, and drops examples that do not cite a verifiable Line N. When no violation rows remain, the overall summary is normalized so it cannot claim findings while the list is empty.

Pattern Message examples that include a line number now show Surrounding messages from your imported thread (several lines before and after the cited line, with the cited line highlighted). That makes cycles and escalation easier to read than a single excerpt alone. Line badges jump to the full conversation when that control is available.

Bible, Pastor's Lens, Law, DSM-5-TR, and Schema Message examples now sit on a warm highlighted panel with stronger text and borders so quoted lines are easier to read in light and dark themes—including when lens sections appear below the main report.

In the Full conversation panel, message text now sits in the same column as the sender name instead of starting at the far left of the card. That removes the look of a single odd character sitting below and left of the name (often the first highlighted character). Sender labels also strip invisible Unicode from exports, and initial avatars use a full grapheme so emoji names do not split into a broken glyph.

For long conversations analyzed in multiple parts, each part now keeps at least one message from you and one from the other person whenever the full thread has both—even when a size limit would have cut a chunk in the middle of a long run from one side, or left a short one-speaker tail at the end. That avoids a false “could not map first and second party” stop on part 2 when you cannot control the export format.

After you send a message, the share panel (copy link, Send in Messages, email link to partner, and share to an external email) no longer disappears automatically. Use Dismiss when you are done. For any message you sent, open it in the list and expand Share this message in the reading pane to use the same options later.

The Everyday AI Assistant conversation view no longer shows Copy, Regenerate, thumbs up/down, or follow-up suggestion chips (for example “Summarize that in one sentence”). You can still edit your own messages to resend, use search and export, and adjust options as before.

After you send a message, the success notice and share tools stay visible until you dismiss them (they no longer disappear on a timer). On any message you sent, open Share this message to copy the wizard link, send it in Messages, email the link to your partner, or use external email share. External share uses the saved message id for the message you sent, so recipients see the correct content.

Introducing Emily Chat — a dedicated AI assistant experience inspired by Claude Sonnet 4.6 Extended Thinking. Emily provides thoughtful, nuanced responses with deep reasoning while remaining conversational and warm.

• Extended thinking mode — Emily explores multiple angles before concluding and acknowledges complexity in responses
• Smart model selection — Automatically picks the best AI model based on your question complexity (vision, fast, capable, or premium models)
• Instant model switching — Type commands like /claude, /gpt4, /deep, /fast to switch AI models on the fly
• Voice conversation mode — Enhanced two-way voice chat for hands-free interaction
• Fully responsive design — Optimized for desktop, tablet, and mobile with sidebar navigation and touch-friendly controls

Access Emily Chat from the dashboard sidebar. Type /help in the chat for all available model commands. The existing Everyday AI Assistant remains available at /dashboard/ask-emily.

The AI-Powered Suggestions section on the dashboard now shows a unified Self-Reflection column that combines both Reflection prompts and Check-in ideas. This creates a more cohesive experience for introspective activities, showing alternating reflection and check-in suggestions in a single, streamlined view.

You can now upload conversations in more formats beyond SMS exports: PDF files, Word documents (.docx), and images (JPG, PNG, GIF, TIFF, WebP, BMP).

• PDFs — text is extracted page-by-page and parsed as a conversation
• Word documents — paragraph text is extracted and processed
• Images — text is extracted using AI vision (Gemini) to read screenshots, scanned documents, or photos of conversations

All file uploads have a 50 MB size limit. After text extraction, the conversation is analyzed just like pasted text or SMS exports.

When the server cannot map two distinct speakers, the dashboard shows the guidance in a banner above the paste area (as well as the toast), with Dismiss. Editing pasted text or choosing a new file clears it.

The Full conversation panel now shows file attachments (images, PDFs, documents) as styled chips with file type icons instead of plain text. Attachment-only messages show a clearer indicator, and rows have improved hover states and visual hierarchy.

The Safety Check now uses an ensemble judge that analyzes reports from multiple AI models, quantifies confidence for each detected pattern, and flags when models disagree.

New in results:

• Pattern confidence scores — see how much the models agreed on each pattern (with confidence intervals)
• Temporal dynamics — escalation velocity and cycle detection for patterns that change over the conversation
• Alternative interpretations — when confidence is mixed, see other possible explanations (e.g., situational conflict vs systematic abuse)
• Dissent flags — highlighted warnings when models disagree on severity or interpretation

This provides more transparency into how the analysis was derived and helps you understand which patterns are high-confidence vs. those that warrant additional consideration.

Two powerful new features for premium Safety Check users: specialized analysis and evidence kit export.

Multi-perspective specialized analysis:

• Forensic Linguist — analyzes control tactics, manipulation structures, and linguistic escalation patterns (DARVO, gaslighting, love-bombing cycles)
• Trauma Specialist — trauma-informed impact assessment, coping mechanisms, and evidence-based safety planning; non-pathologizing and stabilization-focused
• Legal Evidence Reviewer — evaluates evidence quality, admissibility, and protective order alignment

Evidence kit export:

• Tamper-evident PDF with complete analysis and chain of custody
• JSON export for machine-readable documentation
• Cryptographic SHA-256 hashes for integrity verification
• Verification statement for legal proceedings

These features provide professional-grade analysis for documentation, therapy preparation, and potential legal use. Consult a legal professional for admissibility guidance.

The results view now puts Risk assessment right under the severity hero, keeps Immediate guidance easy to find, and tucks longer sections (projected impact, reactive details, coping strategies) behind collapsed panels by default so the page is easier to scan.

History contextgroups past-check notes, comparison bullets (including stable baselines, recurring patterns, and role consistency when data is available), and optional timing detail in one card. Import “Send to Safety Check” can use in-depth analysis when you have premium credits.

Behind the scenes: multi-analyst guidance is merged after the judge, the judge sees both ends of long transcripts, longitudinal summaries are clearer for the models, and prompts emphasize specific next steps tied to your thread.

Treatment-plan and note-based ICD-10 suggestions are cross-checked against a curated set of common mental/behavioral F-codes and Z-codes so descriptions stay aligned with standard terminology for charts, care coordination, reporting, and billing when the model is vague. Always verify codes against the current CMS ICD-10-CM tabular list, your institution, and payer rules.

DSM-5-TR & clinical documentation now summarizes the six-step differential framework, symptom count/duration/impairment and specifiers, manual-associated assessment aids (with APA links), and why the manual is not for self-diagnosis—with Find a Psychiatrist and SAMHSA helpline. Maintainer doc docs/reference/DSM5_TR_ACCESS.md matches.

The results notice and Universal Chat handoff no longer frame the report as “not a real assessment.” Copy now states what the tool can and cannot see from the pasted thread, affirms your judgment when findings match your experience, and still points to crisis lines and in-person help when risk or decisions need a full context.

Clicks from the notifications list and dashboard widget now normalize links (including full site URLs and legacy /api/bug-reports/… paths) so they navigate correctly in the app. Opening a report via ?bugId= shows your submission with status, optional resolution notes, and read-only fields, plus shortcuts back to notifications or a fresh report form. The API also treats you as the owner when your email matches older rows that had no user_id.

When the model or an import left bold markers in plain text (for example * *The takeaway instead of a real name), those tokens are removed for display so pattern cards, evidence lines, and the conversation thread show readable speaker labels.

The streaming Safety Check card no longer flashes through fake “analyst 1/3” steps; it stays on a single AI safety review step while the real report runs (with copy that long threads are normal). Participant detection from pasted exports no longer treats the last name before the time as the second person when the label includes parentheses (e.g. Dad (Allen) Ramberg (6:03 AM)). Auto-select of “which name is you?” no longer matches your email local part as a substring inside the other person’s name (e.g. allen inside Allen in their display label).

Generic thread labels like You, Them, or Person 1 are replaced with your saved participant names when the report has them, so the thread matches the rest of the report. Timestamps sit on their own line under each sender (easier to scan). The intro explains that some imports split long messages or attachments into multiple numbered rows that still align with Detected Patterns.

In the full conversation panel, pink highlights now expand from each matched phrase to the whole sentence (using line breaks and . ! ? as boundaries, without splitting on decimals like 3.14). If a bubble has no sentence punctuation, the entire message is highlighted when a pattern quote matches inside it.

Text, HTML, and CSV exports now include a short note when formal Primary victim / Primary abusertracking labels were not set on that run, so “Not determined” does not contradict pattern attribution or narrative reasoning elsewhere in the report.

Long streamingSafety Check runs no longer keep an idle database connection open for the whole request, and teardown handles dropped Postgres/SSL links gracefully. That avoids rare “response already started” crashes when the analysis finishes after a long wait.

The streaming progress card shows one clear status line (and an optional short subline) instead of internal step names next to the message. The quick Safety Check dialog title and footer are tightened up for a calmer in-progress state.

Analyst instructions now require weighing every line in the transcript, not just the start or end. When the evidence does not support naming one person as the primary abuser, the report can leave those roles unset and explains that in role determination. The results screen shows that section (with a short intro) even when no abuser/victim badges apply. Dashboard and streaming checks use the same role logic as the main pipeline after the AI report is built.

If full AI analysis times out, fails, or is unavailable but the check still returns a minimal report, a clear Simplified report alert now appears above the results (with Try again when retry is possible). Previously that case looked like a full report because the old notice only showed when no report object was returned at all.

Evidence support now reflects how many different line numbersyour examples cite—not whether the pattern is “proven.” Repeating the same line no longer reads as “high” just because there are several snippets. Example lines with messy quotes (for example Name"s: before the real message) show the longest sensible quoted text so the conversation excerpt is readable.

The Projected impact section now shows each trajectory (victim, abuser, relationship, with/without intervention) as its own card with a color accent, icon, and easier-to-scan typography in both light and dark report themes. The section opens by default when the report includes trajectory text, so you do not have to expand it first. Print / Save as PDFuses the same card layout (no longer buried inside “Analysis details” as one plain paragraph).

Reports now require line-attributedexamples tied to the evaluated person's messages, and the model is instructed not to claim “no violations” for that person when concerns are listed. The server drops examples whose cited line is from the wrong speaker and fixes summaries that contradict the violation list.

The Projected impact section is guided to stay tied to this conversation: each trajectory field should reference concrete patterns or tactics from the report, avoid copy-pasting the same paragraph across fields, and keep physical-risk language aligned with what the text actually shows.

In the full conversation panel, message text uses higher contrast in light and dark themes, and timestamps are easier to read on dark backgrounds. If a saved line has no stored message body but your original paste still matches the same number of lines, the UI fills that text from the paste when possible. New saved checks (paste path) also store that paste on the report when needed so shared linksand reopening a saved run can recover the same lines. Otherwise you'll see a short explanation instead of a blank bubble.

In Detected Patterns, each pattern card now has its own emoji tile (for example gaslighting, DARVO, isolation, love bombing) so behaviors are easier to tell apart at a glance. The numeric severity badge still uses the existing severity icons.

Saved Conversation Safety Check results now show a clear Report # in the header (same id as the database run). Downloads and print/PDF exports include a matching Report reference number. In Admin → Safety Check Runs Analysis, use Look up report or ?runId= in the URL to open that run for review.

Voice conversation uses a ChatGPT-style round voice control, prefers a more natural browser voice when available, and says AI coach in read-aloud instead of legacy name mentions. After the coach finishes speaking, the microphone can reopen automatically so you can reply by voice; assistant replies also animate with a slightly slower, easier-to-follow typed reveal.

Analysis prompts now require quoted or clearly referencedthreats or harm before predicting physical violence or filling “without intervention” with a physical outcome. Emotional and psychological abuse can still be flagged seriously; escalation language distinguishes relational/emotional risk from physical risk when the text does not support the latter. The risk section includes a short note explaining this for readers.

In the Schema lens section, themes grouped by participant now explain that activation scores reflect patterns attributed to that person. When the model omits per-message attribution but labels victim/abuser roles, those patterns are grouped under the correct side. Unattributed patterns use clearer wording and guidance to check the main pattern list.

On the main Safety Check page, use Simple wizard (popup) for a short two-step flow in an on-screen dialog (paste or upload, then relationship and optional lenses). Results appear in the same window with a shortcut to the full original report. Add ?popup=1 to the Safety Check URL to open it automatically.

The wizard is available again at /dashboard/safety-check/wizard, including optional generic labels (“You” / “Other person”) and custom names for each side in the report. Opening a saved run with runId still uses the original streaming flow.

In the floating Full conversation panel, messages that match detected patterns now show the entire message text (with phrases still highlighted), instead of a shortened sentence window around the match.

The quick Simple wizard (popup)now fills the other person's name from detected speakers when there are two, tries to match your account name or email to Which name is you?, and adds optional JPEG/PNG/WebP photos for each side (same behavior as the full wizard), including on the streaming request.

The pipeline strips trailing parenthetical analyst labels models sometimes add after the closing quote (such as initial support and validation), and the analyst prompt forbids that format. Pattern cards show verbatim dialogue only for the quoted part; interpretation stays in the pattern description.

The Risk Assessment intro now states that escalation likelihood covers intensifying harmful dynamics (including emotional distress and coercion), separate from physical violence risk. It notes that high escalation with low physical violence risk is a common pattern when the thread shows serious emotional harm without described threats or bodily harm—so the three fields read as complementary, not contradictory.

The pattern-card Evidence support badge now matches the helper text: high requires at least three cited examples on three different transcript line numbers. Three examples that only span two lines rate medium. Two examples on the same line rate low (previously medium).

When the server drops individual pattern quotes that don't match the transcript (but keeps every pattern card), Analysis & processing notes no longer shows (0 pattern(s), N example(s)), which read like a contradiction. The text now states that only examples were removed and that all listed patterns remain.

When the report keeps a written role summary but does not set the formal victim/abuser labels for you vs. the other person, the intro no longer says no primary abuser was identified—so it cannot contradict a summary that still describes who showed harmful patterns. The note now explains that missing labels are about tracking slots, not pattern attribution.

If the server removed or tightened examples or guidance so they match your paste, a short Report adjusted to match your transcript notice appears under the severity hero. Open Analysis & processing notes expands that section and scrolls to it; opening the report with URL hash #safety-analysis-processing-notes does the same.

After transcript grounding, high-severity pattern cards no longer keep examples whose only cited text is a trivial acknowledgment (e.g. thanks, ok, u2) unless other substantive quotes remain. Prompts and the judge also warn against tagging DARVOon the wrong speaker when quotes are only the other person's confrontational labels without a clear deny-attack-reverse sequence from the attributed party.

Report generation now instructs models not to treat isolated pleasantries or bare acknowledgments as the main proof for high-severity control, cycle, or withholding patterns, and the judge step penalizes thin evidence. When a transcript line and pattern attribution disagree, the pattern card shows a clearer Check attribution callout above the line metadata.

When a pattern is tied to you vs the other person, each example is checked against the threaded messages: the cited line must contain the quoted or matching text, and the message must be on the correct side (your rows vs theirs). Examples that only looked like a line number, or that quoted the wrong speaker for that pattern, are dropped so cards rely on grounded evidence. If something was removed, Analysis notes may mention the evidence guardrail.

In Projected impact, the five trajectory blocks (victim, abuser, relationship, without intervention, with intervention) now share the same sage accent, border, and background tint so the section reads as one calm panel instead of five different hues.

Updated the shared site branding defaults to Toxic Partner, including the primary logo asset, default app title, default support/contact addresses, and SEO metadata copy used across public pages.

reflectwithemily.com and www.reflectwithemily.com are now forced to redirect at nginx for every path, so the legacy host cannot serve app pages or API routes and always forwards to toxicpartner.com.

Visitors who open reflectwithemily.com (including www) are now sent to toxicpartner.com with the same path when possible, using a permanent redirect so bookmarks and links update over time.

Results now sit in a clearer report shell with semantic surfaces, a tighter title and disclaimer block, and a calmer glossary strip. Longitudinal and timing callouts under the report use the same dashboard tokens for borders and text. The crisis-resources strip and the “Ready to take action?” hub use sage-accented cards and tiles instead of a heavy purple gradient, with clearer typography and focus rings. Confidence bar, pattern filter chips, summary-strip CTAs, and conversation jump controls now use sage, blue accent, and semantic focus rings instead of purple. The severity hero, pattern cards, role determination, people block, processing log, and analysis notes use semantic surfaces and text tokens; analysis accordion uses a sage icon theme and shared focus rings on headers.

Under Export report, you can show your side as You (without your display name), set an export-only label for the other party, and open Choose sections and quoted text to include or omit crisis resources, patterns, lenses, examples, trends, role lines, metadata, and next steps across text, PDF, and CSV. CSV pattern rows now include a dedicated Examples column when you allow pattern quotes.

When the model skips per-pattern participant tags, the pipeline now infers reader vs other from reactive assessment, victim/abuser roles, and line headers in quotes, so the DSM-5-TR per-person summary and criterion crosswalk attribute evidence correctly. Lens pattern rows use the same red (other) and blue (you) accents as the main report; schema lens examples use the same quote coloring when the speaker can be read from the line. If the full criterion table is missing for a run, an on-screen notice explains it.

After sending in Communications Wizard, the share flow no longer auto-targets your partner by default. Send in Messages now stays disabled until you explicitly choose who should receive the link.

The Safety Check DSM-5-TR lensnow makes each criterion easier to review: matched criteria are visually highlighted, each row includes a plain-language “why this matched” explanation, and supporting message excerpts are shown in clearer quote blocks with line, speaker, and timestamp context. Disorder cards now also show full diagnostic code context and the diagnostic criteria statement.

Bible evaluation, law evaluation, Pastor's Lens, and schema lens sections now spell out which participant quoted lines come from (using your names when the run includes them). Law rows use the same “Message examples” pattern as Bible rows, and the schema lens can surface short quoted lines from detected patterns next to each theme.

Safety Check export options now default DSM lens labels to plain-language wording and add an opt-in checkbox to include technical DSM domain names when needed for advanced review.

Safety Check results now offer the same Discuss in Universal AI Chat follow-up from classic, wizard, cards, minimal, and guided modes. Reports now include a structured coaching follow-up block with de-escalation steps and confidence-aligned guidance wording, plus clearer top-of-report action shortcuts.

Safety Check now uses one entry flow with internal mode switching (Quick, Guided, Cards) plus shared scoring guidance. Longitudinal Evaluation now supports standard and technical views, and report/export surfaces show friendlier domain labels while preserving technical labels for advanced users.

The Longitudinal Evaluation relationship filter now uses a dropdown built from relationships that already have longitudinal data, instead of free text. InAll relationships mode, the summary now shows a split by relationship context instead of one combined aggregate, and adds a per-relationship arc (trend + sparkline) with one-click filtering into that specific relationship.

Safety Check results now add mobile-safe bottom spacing so fixed actions do not cover report content, a new quick summary block near the top for faster scanning, and cleaner strategy cards with clearer labels. Duplicate AI coach prompting in the results actions area was also reduced to keep the primary next step simpler.

Admin Safety Check quality now includes a reviewer rubric and adjudication workflow, weekly drift alerts with report persistence, per-run prompt/guardrail version metadata, and a structured remediation queue linked to root-cause buckets.

Support Assist now supports a simpler flow: helpers enter a session code and request access, then the customer approves in one click. Under the hood, Support Assist now supports shared signaling via Redis for multi-instance reliability and backend-managed TURN/STUN ICE configuration (including ephemeral TURN credentials).

One-time Safety Check verify links now wait for an explicit Open report click on the verify page before redeeming the token, reducing failures caused by automatic email link previews consuming the link before you use it.

In Save & share, when a report is already shared with your Couples Connect partner, you can use Revoke partner access to remove it from their Shared with you list. Public timed share links are unchanged—use Revoke share link for those.

New Import ChatGPT history at /dashboard/import/chatgpt: consent, preview, analyze, and import official exports (ZIP or conversations.json). Also linked from the main Import page and site search.

When pasted text cannot be split into two speakers, the error message now explains concrete line prefixes (for example Me vs a name, or You sent vs their name) and reminds you to try the other option under Which person are you? when that step appears. Parsers also treat labels like Partner and Spouse as the other side when they match exactly.

Added Support Assist Screen Share at /dashboard/support-assist: end users can start a session, explicitly grant one internal helper by account email, stream their screen, and receive live laser-pointer guidance. Owners can revoke helper access or end the session at any time; v1 does not allow remote keyboard or mouse control.

Everyday AI Assistant now uses the same system context for streamed and non-streamed replies (including Reflect ecosystem context when enabled), a shorter default persona prompt, and backend guidance so answers stay conversational instead of echoing rigid bullet templates or bracketed context labels. The empty state and user message bubbles are tuned to feel closer to a standard chat assistant.

/dashboard/settings and /dashboard/send-documents no longer redirect immediately: each shows a short hub with explanation, primary button to Account settings or Send documents (admin), quick links, and the standard footer. Opening Settings with the subscription tab query jumps to the plan section on Account. Messages adds a dashboard home control in the top bar. Everyday AI Assistantuses the same "Dashboard home" breadcrumb label, skips duplicate dashboard breadcrumbs, and hides the bottom nav on mobile like other chat surfaces.

The internal Project inventory page (password-gated when configured) now uses the shared dashboard hub: back to home, the same summary line as before, quick links to Find your tool and Admin, plus the standard browse footer. The stone-toned page background is preserved.

Safe Chat Storage now uses the shared hub: back to the dashboard home, a concise description of what the tool does, quick links to Import, AI Chat, and the Frankenstein analyzer, plus the standard browse footer. Loading shows the same shell with a short status line.

The Preferences page now uses the shared dashboard hub: back to home, a short summary of what you can change here, quick links to Settings, Account, and the notifications inbox, and the standard browse footer. Layout matches the rest of the dashboard for consistent navigation and accessibility.

The Account settings page now matches the shared hub pattern: back to the dashboard home, a short purpose line, View pricing beside the title, quick links to Settings, Preferences, and Report an issue, and the standard footer. Loading uses the same shell with a clearer status line instead of duplicating title skeletons.

The dashboard bottom bar on phones now includes Safety and Therapy next to Find, so Safety Check and the therapeutic tools hub are one tap away without scrolling the full catalog. The same shortcuts appear on the practice mobile bar (after Find) for quick personal-tool access; the row still scrolls horizontally when needed.

Header search now recognizes more plain-language intents across the app — for example stress and overwhelm for check-ins, DMs and social threads for Safety Check, paste and backup wording for import, billing and copay language for practice, boundary and apology drafts for the communications wizard, custody and family-court phrases for case prep, plus richer matches for journal, AI chat, couples, schema questionnaire, hard conversations, and longitudinal reports.

Typing things like how to find things, lost on dashboard, or orientation in the header search now surfaces the Help center more reliably, alongside existing FAQ and navigation wording.

Messages (bottom strip), Account, and Preferences now include the same footer as other dashboard areas: all tools, Find your tool, and Help. Notifications already used the hub layout with that footer.

Accountability, Saved Insights, Memory & context, NPD Watch, and My Sessions now share the same hub shell: back to home, task shortcuts where it helps, and the standard footer. NPD Watch keeps its live refresh; sessions and memory content are unchanged aside from chrome.

Weekly Review and Present for the Next Generation now match other dashboard hubs: back to home, quick task shortcuts at the top, and the standard footer to all tools, Find your tool, and Help. The help icon and guided cards on Next Generation are unchanged.

Repair (from dashboard home and check-ins) now uses the shared hub layout: back to home, three quick shortcuts to the in-app worksheet, Cycle Breaker, and conversation prep, plus the usual footer links to all tools and Help. The same four repair paths as before are still on the page.

Long Home > … > page trails no longer force the whole dashboard to overflow horizontally on phones: crumbs sit in a single scrollable row with ellipsis on very long names, and the layout wrapper passes width through correctly so flex layouts stay stable.

My Goals and Self-Reflection Chat now include the same footer as other hubs (all dashboard tools, Find your tool, Help). The browse footer on hub pages uses the dashboard sage link color and semantic borders so it matches the rest of the member dashboard.

Find your tool now uses the same hub shell as other dashboard areas: clear back link to home, three quick task shortcuts (Safety Check, daily check-in, AI Chat), the standard footer to all tools and Help, and a reminder about Cmd/Ctrl+K search. The step-by-step wizard is unchanged.

Tapping or focusing the dashboard search field (including on mobile after you open it) now shows the same grouped shortcut list as Cmd+K / Ctrl+K, and clearing the field brings that list back. Under your account menu, Support includes Find your tool and How to find things (Help) with a jump to the Help page section on discovery.

The same footer used on major dashboard hubs — links to all dashboard tools, Find your tool, and Help — now appears at the bottom of Find your tool, Longitudinal Evaluation, and Case prep so you can leave those flows without hunting for the dashboard.

The Start here card adds shortcuts to Safety Check, the schema questionnaire, Couples, and (for practice access) billing. All tools now links to Help for finding your way. Site search includes the therapeutic tools hub, activity history, and broader everyday keywords; empty-query search surfaces Therapeutic tools, Hard conversations, and Self-Reflection Chat earlier in the list.

When you open header search with Cmd+K / Ctrl+K (or /outside a text field), the suggestion list is now organized into small sections — find your way, therapeutic & conversations, daily & chat, practice, and account & help — so it reads more like a light command palette than one long list.

The Today section leads with a clear path to Find your tool, plus Search dashboard (jumps to header search with your Cmd+K / Ctrl+K hint) and Browse all tools (scrolls to the full catalog). The All tools block has a stable in-page anchor for that jump. Header search matches more everyday words — e.g. text conversation for Safety Check, therapy chat for Therapeutic chat, feelings and mood language for check-ins, venting terms for AI chat, practice billing and payments, schema program hub and YSQ, hard conversations, drafting a message, importing or pasting a chat (including WhatsApp-style wording), worry and safety around texts, Case prep for legal preparation, and casual phrases like “not sure” for wayfinding. Search also lists Schema tools for couples (e.g. mode cycle, connection dialogue) and extra YSQ-R wording on the schema program hub.

Major hub pages (Therapeutic tools, Schema program, Schema tools for couples, Safety Check design picker, Wizards index, Hard conversations guide, Kids Connect, Practice Management, Admin, and Couples—including invite and error states on Couples) now share a consistent pattern: three tappable task shortcuts where it fits, plus a footer to All dashboard tools, Find your tool, and Help so you can jump out without memorizing names.

The All tools area on the dashboard home page no longer hides whole categories behind accordions. A sticky row of chips jumps to Therapeutic, Growth, Safety, Connection, and the rest; Stay connected and Wizards subsections stay open so partner, family, and wizard tools are visible without extra taps.

The bottom navigation bar now includes Find (opens Find your tool) so task-based wayfinding is one tap away, including on the practice mobile bar. On the dashboard home header, search appears at the top on phones so you can jump to any tool without scrolling past the welcome area first.

On most dashboard pages and on Devotion Connect, Ctrl+K or Cmd+K opens the header site search and shows quick picks (arrow keys and Enter work like before). Ctrl+/ or Cmd+/ does the same except on Universal AI Chat, where that shortcut still focuses the composer. Press / when focus is not in a field (and not in the communications wizard) to open search. Universal AI Chat and legacy dashboard chat keep their own Cmd/Ctrl+K behavior. In full Practice Management, Cmd/Ctrl+K still opens the practice command palette; on the practice upgrade screen only, the shortcut uses site search. The search field shows a small key hint on wider screens. Quick picks under Therapeutic & conversations now include Schema program and Schema tools for couples next to the main therapeutic hub, Safety Check, and Hard conversations. Additional sections cover Relationship & family (Couples, Kids Connect), Wizards, Draft & import (Communications Wizard and Import conversation), Insights & reports (Longitudinal report, Case prep, and My Sessions / activity history), and the usual Practice and account links. Daily & chat includes Journal (reflections) after goals; the palette shows more rows so nothing is cut off.

Quick Access now lists Journal (same place as before: prompts, responses, and insights) and opens /dashboard/journal, which redirects to your reflections. The dashboard home uses shared design tokens for primary buttons, card text, and focus rings so light and dark mode stay consistent. Product docs note that a separate “saved insights” library waits on a future API; insights still live inside each reflection.

From Dashboard → Couples, Shared with you now opens the shared report in the original Safety Check flow. Links to /dashboard/safety-check?partnerSharedRunId=… redirect there as well, so bookmarks still work.

While the shared report loads, you see a clear loading state (not the empty paste form). If it cannot open, you get a short explanation and links to Couples or Safety Check. After buying in-depth credits, return still lands on the classic flow without dropping a partner link. Wizard multi-part results now include working Share with partner and public link controls for the selected part—same as single-report results.

Starting a new check, running analysis, opening a saved run, or clearing the partner URL resets partner view correctly so Save & share shows again for your own reports. Couples → Shared with you shows a retry if the list request fails and hints when the list is empty.

Dashboard Chat, Therapeutic Chat, Bible Research, and Fight Room now keep the composer draft separate from the transcript, so the message list no longer re-renders on every keystroke. Universal AI Chat already used an isolated composer; its input focus scroll was tuned for less jank.

If a multi-part analysis stops partway (including when live progress falls back to sequential API calls), the form can save the queue. Use Resume to restart from the failed part or an earlier part, or Dismiss to clear the saved queue and start fresh.

Available on the original single-page flow at /dashboard/safety-check/classic (linked from Safety Check designs) and on the Wizard when you choose separate reports per split part.

The full-screen progress view now includes a pipeline status banner that follows each step (parsing, pattern scan, specialist review, report build), plus rotating process insights about timing, splits, and what to expect. Multi-part and fallback runs show extra context on the page; the queue card explains how parts connect to your timeline.

Auto-detect and the participant step now infer Me, You, and similar export labels so Party 1 / Party 2 and "which party are you" pre-fill more often (you can still change them).

When a long conversation is split for multi-part analysis, the dialog explains why parts are suggested, typical messages per part, and timeline span when timestamps are available.

In Cards/Wizard split options, max messages per chunk is a text field with clear validation instead of only a number stepper.

If the server could not send a reset email (for example Gmail or delegation misconfigured), the app no longer shows success while invalidating your previous reset link. You get a clear error and can try again; when email works, older unused links are retired only after the new message is sent.

If the live analysis stream stopped early (for example network or proxy timeout on very large splits), the app could return to the analyze step with no report and no message. The client now treats an incomplete stream as an error with clear guidance, only advances to the next part after a successful report, and allows longer waits (10 minutes) to better match the backend. Production nginx sample config includes a dedicated long-timeout path for the analyze-stream endpoint.

Analyst and judge guidance now forbid self-referentialwording (e.g. one person "dismissing" their own feelings under the same label). Empathy failure and other dyadic patterns must name reader vs other correctly so descriptions match a two-person thread.

Further analyst, judge, and synthesis instructions: a brief question about predictability (e.g. "Why the structure?") is not Entitlement by itself. If one person proposes a regular time or repair deadline, do nottag the other for "expecting" that schedule when their lines only question it. Descriptions must match the same speaker as the quoted evidence; otherwise omit Entitlement and note it in confidence notes.

On narrow screens, the floating Full conversation control (expand/collapse the thread on results) now sits above the dashboard bottom navigation and uses the same stacking layer as other fixed dashboard actions, so it stays tappable instead of being covered.

The DSM-5-TR lens per-person summary previously showed "No criterion items were linked" even when pattern indicators were clearly attributed to a participant. This happened because many LLM-generated pattern names (e.g. "Blame-Shifting & DARVO", "Reality Distortion") did not match the criterion crosswalk configuration.

The fix expands the criterion crosswalk to cover commonly generated pattern names, adds compound-name resolution (splitting "Emotional Withholding & Stonewalling" into its components), and provides a meaningful fallback summary from pattern indicators when formal criterion matches are absent.

A new Case Prep tool helps you build and organize defense evidence for your attorney. Upload text messages, court testimony (including PDFs), legal documents, and your own notes — the system guides you step by step on what to submit next.

Behind the scenes, text messages are analyzed using the same engine as Safety Check (abuse pattern detection, reactive abuse classification, DARVO detection). Testimony claims are extracted and cross-referenced against text evidence to find contradictions, gaps, and narrative divergences.

When ready, generate an Attorney Brief with contradiction reports, timeline reconstruction, pattern analysis, defense considerations, and evidence gaps — exportable as text or HTML for your lawyer.

The analyst instructions now exclude common false positives: mutual boundaries, proportionate accountability or apology requests, repair timelines, and structured contact as a condition for sharing—unless there is clear evidence of double standards, unilateral privilege, or classic entitlement phrasing. Ambiguous cases should move to confidence notes instead of a high-severity tag. The judge step is told to prefer reports that do not over-apply this pattern.

The dashboard results Next steps list also links to Case Prep (evidence organizer for counsel), Help: Safety Check, and the Longitudinal report, in addition to Fight Room, Communications Wizard, and Discuss with AI Coach.

The guided wizard and designs Ready to Take Action? panel includes the same three shortcuts as tiles alongside Fight Room, Communications Wizard, and Discuss with AI Coach, with a responsive grid (two columns on medium screens, three on large).

The optional pattern library for the analyst (when hosts enable learning RAG) now ships with curated fictional dialogue excerpts and guidance—covering themes such as denial versus evidence, accountability pivots, reactive escalation, stonewalling, coercion masked as humor, invalidation, punitive responses to boundaries, triangulation, economic control language, hollow apologies, repair-shortcutting, and uneven labor. Enable with SAFETY_LEARNING_RAG_ENABLED; maintainers can extend data/safety_learning_corpus.json after review.

For users who opted in to anonymized improvement data, aggregate feedback now feeds the analyst with a 365-day window and the lowest default statistical floors (a single qualifying signal can surface a pattern tag; up to 24 tags). No conversation text is used. Hosts can raise thresholds via environment variables if calibration notes become too frequent.

When the pipeline infers a shared UTC offset for message times, the report header now uses a plain label (for example UTC−06:00 or India Standard Time) instead of raw fixed-offset zone identifiers.

The same label is used in exported text, printable HTML, and CSV so downloads match what you see on screen.

Safety Check's Cards and Wizard flows now use a client watchdog that cancels stalled analyze requests after a long wait and shows a clearer recovery message, so transient proxy hangs do not leave the interface stuck in loading.

In Cards and Wizard split options, Max messages per chunk no longer stops at 200—you can enter any size at or above 10 messages (still with 5-message overlap between chunks for context).

Safety Check reports now surface clearer role framing (victim/abuser context where available), expanded reactive-abuse detail, and a more transparent processing log so users can better understand how each report was produced.

Long-run analysis now includes an explicit ignore limit and run as one path in the queue chooser, plus improved queue status behavior during multi-part runs.

Lens output is also richer: DSM-5-TR now includes per-person summary context, and the law lens supports dedicated victim-impact and abuser-liability summaries when provided by the backend.

Results now include a daily calibration chart for consented feedback (Helpful, Too certain, Misleading, and related signals) so movement is easier to see at a glance.

Operators can tune alert thresholds for when aggregate calibration should surface hints to analysts, and admins get a dedicated trend dashboard for Safety Check feedback over time, with CSV export for review.

Operators can set reactive-vs-primary classification to use a trained backend (HTTP JSON API or a local Hugging Face model) instead of the default zero-shot pass, using environment variables documented in env.example. The results screen renames the notes section to Analysis & processing notes so it matches processing limitations shown in the report.

Hosts can enable a reviewed pattern library that retrieval-matches anonymized reference examples to each run and adds them as educational context for the safety analyst (optional; off by default). Aggregate feedback calibration runs automatically for users who opted in to anonymized improvement data: analysts receive consent-gated hints from aggregated pattern tags when thresholds are met (no conversation text). That calibration path cannot be turned off by configuration. Neither feature trains models on private chats.

Safety Check now explains why reactive-vs-primary classifications were made using plain-language reasoning steps, confidence percentages, and clearer evidence support labels on detected patterns.

The pipeline also adds deterministic guardrails that downgrade overconfident reactive/primary labels to unclear when sequence evidence is thin, and reports now surface processing limitations (for example truncation, missing timestamps, and large time gaps) as Processing notes in both UI and exports.

The feedback section also adds Too certain and Misleading framing options to improve calibration over time, plus a rolling 6-month feedback summary so users can see how often those calibration signals are being reported.

Safety Check now also shows daily feedback trend snapshots, so calibration movement can be tracked over time rather than only as a single point-in-time summary.

Those trend snapshots can now be downloaded as CSV from the results page to support external review and auditing.

Safety Check exports now match the DSM-5-TR lens UI: text, PDF/print, and CSV include the exact criterion matched for each pattern (disorder + criterion ID + criterion text). Generic fallback labels such as Concerning communication (unspecified) are filtered out from this matched criteria view.

In Safety Check's DSM-5-TR lens, each pattern indicator now lists the specific criterion entries it matched (criterion ID + criterion text + disorder), so the report shows what criterion was matched each time. Generic fallback labels likeConcerning communication (unspecified) are no longer shown as matched criteria.

Pastor's Lens summaries now avoid invented local details (for example specific church names) and use generic wording like your pastor unless those details are clearly present in the source data. No-violation summaries are also phrased more carefully as no clear violations identified in the provided messages.

Bible evaluation no longer assumes your side automatically after parsing a conversation. You now choose You are explicitly before running, which prevents accidental role inversion and keeps the report focused on the intended other person.

When someone pastes a shared Safety Check link, apps that show a preview (for example Slack or email) now get a clear title and description instead of generic "shared link" text or a loading placeholder. The page stays private (no indexing) and does not put report details into the preview.

Under My practice site (SEO and help assistant), you can turn off the floating help button on your public /p/… page and in the patient portal, or enable training mode so visitors see the button with a setup message instead of live chat. Saving updates what customers see right away after publish.

Under Optional lenses, you can include the Schema lens on the same run. It maps detected conversation patterns to Young-schema-style themes and optional mode hints for reflection only—not a questionnaire, SMI, or clinical assessment. Exports and shared reports include this section when the lens was selected.

When you include the DSM-5-TR lens on a Safety Check, the criterion list now shows the diagnostic criteriafrom DSM-5-TR for Narcissistic Personality Disorder (301.81) and Borderline Personality Disorder (301.83), plus each disorder's Criterion A introduction and matching ICD-10-CM codes (F60.81 / F60.3). Detected conversation patterns still map to criterion items only as excerpt-level themes; this is not a diagnosis and does not apply the manual's threshold rules (for example, five or more criteria). Shared Safety Check reports use the same lens section when the lens was included in the run.

From a saved report, use Re-run analysis to run again with your current options (standard or in-depth, optional lenses). Choose Update this saved reportto replace that check's snapshot while keeping the same history entry and share links, or Save as new report to add another row in past checks.

When a pattern example references Line Nin your transcript, the report shows that line's date, time, and speaker and flags mismatches between attribution and the labeled speaker when possible. Download as text, Save as PDF, and Download as CSV add the same transcript metadata to those citations when the thread is available, including when you use name redaction for professional sharing.

On the full safety report, use Show or hide behaviors to toggle which detected pattern cards appear. Patterns-focused view hides notes, guidance, coping, risk detail, and the conversation panel so you can concentrate on patterns; you can optionally include dynamics summaries (flow and balance) in that mode. When only the shorter pattern list is available, the same style filters apply to the detected pattern rows.

Turning on Voice chator using the microphone with Voice chat on now opens a short explanation first, then your browser or device can request microphone access. The empty chat view no longer shows preset suggestion buttons—type or dictate your own message. The header "suggested starters" shortcut was removed.

The /marketing/beverages experience now has a tighter premium layout with stronger hero hierarchy, clearer action buttons, and an upgraded visual portfolio section.

Brand presentation was polished for consistency: official logo tiles, improved spacing and readability, and cleaner section transitions from services through contact.

The contact area now provides clearer status guidance and reassurance while keeping the production submission path and static pre-generated imagery model.

Everyday AI Assistant and dashboard Universal AI Chat include a microphone for browser speech-to-text with live interim text in the input. Optional Voice chat sends when you stop the mic and reads assistant replies aloud when your browser supports speech synthesis (any text-capable model). Each area remembers its own voice-chat preference in this browser.

The /marketing/beverages showcase now uses official brand logo sources (instead of Clearbit placeholders), improved color/contrast treatment, and tighter visual hierarchy—including more brands and filters (for example energy, juice, and sparkling).

Section imagery now comes from an offline image set selected by multi-agent consensus voting, then served as pre-generated static assets for predictable performance and no on-page image generation.

The beverages contact form is now production-ready: real submission handling, stronger validation, error feedback, and configurable routing to the configured practice inbox or lead capture fallback.

New public page at /marketing/beverages: full-bleed beverage distribution showcase with AI-generated imagery (Pollinations), brand logo grid, service sections, and a demo contact form. Uses Barlow Condensed display type; platform footer is hidden on this route for an immersive layout. Third-party logos and generated images are for presentation only.

On Couples Connect, creating or editing a relationship contract now uses one block per part—Heading and What you agree to—with Add section, reorder, and remove. The API still stores the same structured data; you no longer edit a JSON array by hand.

When analysis fails or times out, messages now suggest practical steps—such as Split into parts, a shorter paste, or Standard instead of In-depth—and distinguish rate limits from other errors. This applies to the main Safety Check flow and the wizard.

In the wizard, you can add an optional photo for each side; they show on the report header and are saved with the run. A new Dynamics summaries section breaks down conversation flow, relationship balance (from this excerpt), and messaging style from thread metrics. The DSM-5-TR lens now lists paraphrased criterion lines for NPD and BPD and marks where this excerpt shows conversation evidence aligned with a criterion—this is educational and not a diagnosis.

When a paste or export used the same display name for both people, reports could read as if one person was speaking about themselves. The pipeline now treats your messages as Meand the other side using your relationship line (or a generic "The other person"). In the wizard flow you can also set custom labels for each side or choose generic You / Other person for either side before running the check.

The shared task list adds Text file (for example .txt, .md, or .csv) and Paste list, alongside photo import. One line per task (up to 25); markdown checkboxes and bullets are stripped; then use the same review step to add tasks or send some to personal goals.

Invoice statuses show clearer labels (for example Sent, Overdue, Paid). Auth failures on refresh, payments, PDFs, and waiting room actions clear the portal session consistently and prompt for a new link from the provider.

Intake forms that define sections now render with section titles and descriptions instead of a single flat field list.

Token-backed read-only share pages for vulnerability cards and timeout agreements are now live at /share/vulnerability-card/[token] and /share/timeout-agreement/[token], including expiry and max-view behavior.

Analytics summary chips now include focusable info buttons that open explanatory text panels for keyboard and touch users, replacing the previous hover-only explanation behavior.

The new Safety Check analytics summary chips now include color-coded states (green/amber/red/neutral) and hover tooltips that explain each metric in plain language.

Safety Check results now include an at-a-glance analytics summary section with compact chips for reliability, acute change flags, dominant phase, asymmetry direction, new patterns, and top schema signal before the detailed lens sections.

Safety Check responses now include advanced analytics: pattern trajectory, acute change flags, conversation phase map, role asymmetry, and reliability score.

The YSQ-R Schema profile now includes a structured self-help path card that turns current schema and mode signals into practical daily steps: mode check, short regulation, need framing, and one Healthy Adult repair action. You can now save a weekly checklist (including a short intention) on each assessment, see last-saved state, and view a completion trend across assessments. It is educational reflection support and not diagnosis or therapy.

In Admin Settings, the Safety Check NPD model card now includes live status for the effective HF model source and path, admin auto-mode state, and override validity so admins can verify actual runtime behavior immediately after saving.

The Admin Datasets page now shows the effective Safety Check Hugging Face model resolution details (source, resolved path, admin auto-mode state, and override validity) so admins can quickly confirm what model is actually active on that host.

Safety Check now automatically uses a valid local Hugging Face NPD model when available. Admins can configure this directly in Admin → Settings by toggling auto-use on/off and setting an optional model-path override, without editing environment variables.

The YSQ-R v4.3 take flow now loads the exact 116 question statements from the official v4.3 scorer workbook source used in production, instead of placeholder-masked prompts. This restores full item wording in the questionnaire while keeping scoring behavior unchanged.

In Practice Management → Site → Visual style, the template picker now explains that the default Sage Editorial preset matches the Reflect design stakeholder page at /design/reflect-design-system-showcase.html and the same theme tokens apply to public /p/…, /book/…, and authenticated /portal/…. Template swatches use a shorter screen-reader label; full descriptions appear as hover text.

Cycle Breaker in Partner Connect now includes a structured trigger capture step with optional context, a dedicated repair conversation flow, and a new pattern insights panel that highlights recurring trigger patterns.

For reassurance-heavy moments, the Reassurance Plan module now lets you save concrete calming actions during a session so partners can use consistent, predictable repair behaviors.

Each downloaded chart page in Full or Charts-only YSQ-R PDF exports now includes a short accessibility note: where to find the same numbers in the document’s tables, or when to pick Summary/Full instead of charts-only. (Raster charts in PDFs still have no image alt tag—this text bridges that gap for screen-reader and read-aloud users.)

The help panel now picks up shared partner YSQ comparison and shared schema profile URLs with tailored questions and tips (including hidden chart tables and mode-map keyboard notes). The main schema program surface mentions the skip link and screen-reader tables in its quick tip.

The public schema patterns (side by side) page for linked partners now labels each bar chart for assistive technologies and adds hidden tables with the same subscale means, domains, and elevation flags as the schema profile bar chart (top twelve subscales per person). Sections use heading ids for clearer landmarks.

YSQ-R profile and clinical report tables now show workbook-style elevation (Format 1), mean-based flagging (Format 2), and higher-endorsed item counts (Format 3) side by side, with safe fallback for older saved runs. When no subscale crosses the workbook-style threshold, profile/report and PDF now include the same brief pattern note for discussion context (self-report only, not diagnosis).

On public practice pages, the floating Susan assistant is now more compact on phones. The launcher button, panel width/height, header spacing, and in-panel padding were reduced for mobile so it no longer dominates small screens; tablet/desktop layout stays unchanged. The assistant Voice/Text controls also now inherit theme tokens so button colors stay consistent across all practice templates, and the header presence indicator now uses success-theme tokens instead of a hardcoded green.

The mode map now uses a custom default node (same connection handles as before) with an explicit label for each node, including saved trigger notes when present. Tab / Shift+Tab move between nodes and edges, with React Flow keyboard support left enabled. A skip link inside the map jumps to the trigger-notes panel (focusable target). Visible keyboard instructions sit above the diagram; a full focus trap is intentionally not used so users can Tab out of the canvas.

The schema program mode map now includes a hidden summary before the diagram: tables of node labels (with trigger notes when present) and connections (from, relationship label, to). The diagram sits in a named region that points screen readers to that summary. Applies to both editing and read-only shared views.

When you have more than one YSQ-R completion, the longitudinal domain line chart now has a descriptive label for assistive technologies and a hidden table of domain means by assessment date. When RCI reference bands are shown, the label explains their meaning in plain language.

The all 20 subscales circular network (inventory order) on the printable report and PDF chart capture now includes a hidden table with order, name, domain, mean, and workbook / mean elevation flags for every slot, plus a clearer chart description. Matches the accessibility pattern used for other schema profile charts.

The domain summary radar on the schema profile now includes a clearer chart description for assistive technologies and a hidden table with domain means and workbook-elevated subscale counts, aligned with the bar and domain bubble charts. Captions reflect standard vs Yalcin-style domain grouping when that toggle is on.

The domain concentration scatter chart on the schema profile now has a descriptive label for assistive technologies and a hidden data table with the same domain means and workbook-elevated subscale counts, matching the pattern used for the main subscale bar chart.

From Schema profile, connected couples can create a time-limited compare link (two completed Rasch YSQ-R runs). The public page summarizes side-by-side patterns and research-informed discussion prompts—educational only. Profile also adds optional Yalcin-style domain grouping (with literature caveats), a domain bubble view, plain-language mode hints for reflection, and an optional streaming reflection paragraph when the environment is configured.

When hosting configures an inbound Stedi webhook (documented for operators), new 277 and 835transactions can trigger automatic ERA posting and claim status updates in the background—alongside existing cron and "Check status" flows. For denied claims, Suggest appeal now feeds the assistant a structured excerpt of the stored payer/clearinghouse response when available, not only a short denial label.

HTML notifications (session summaries, password reset, practice messages, and more) now use the same cream, slate, and sage tones as the public practice site—forest green action buttons for readability. The signed-in client portal Help and resources panel titles use tighter editorial letter-spacing to match the Reflect design language.

/help now includes Pricing & plans, Kids Connect, and for authorized admins a full Admin Dashboard section with linked routes (users, approvals, invites, plan changes, bugs, prompts, leads, analytics hub, API costs, global settings including session idle timeout, plus operations, deployments, email marketing, and more). Dedicated sections cover performance monitoring and analytics & reporting. Also: clearer Notifications (in-app bell and email) and How Emily uses AI (including models/providers). Contextual help for Activity History (Sessions), goals, pricing, admin, notifications, and related topics points to the same anchors; the docs/HELP.md mirror matches (including #performance-monitoring and #analytics--reporting).

On Practice → Claims, you can turn on also try 835 ERA before Check status. With Stedi configured, the app tries candidate remittance transaction IDs (your submit id plus hints from the live 276/277 response), fetches an 835 when the payer has posted one, and runs the same ledger posting as Process ERA. Requires claims write permission; no change to the separate manual ERA process by transaction ID.

Under Practice → Practice Site → Page layout, you can add Frequently asked questions between Services and Contact (default layout includes it for new setups). The public /p/… page shows the same accordion content as /book/… (privacy, rescheduling, telehealth vs in-person, empty calendar)—no ratings or invented stats. Nav gains an FAQ link when the block is enabled; existing sites keep their saved order until you add the block.

The Conversation Safety Check wizard (designs path) now includes Share with partner above the public link panel, and uses the same timed link, email, and settings behavior as the dashboard Safety Check—including when you analyze by multiple parts (sharing targets the report you have selected).

The in-app Help Center adds Public practice site, booking, and client portal (/help#practice-public-booking-portal) with how /p/…, /book/…, and /portal/login fit together for patients and visitors. Contact a practice is spelled out in docs/HELP.md with the same anchor pattern. Contextual help for Practice and Contact a practice includes booking and portal pointers. Schema profile sharing text now matches product behavior: Email link goes straight to the recipient; Safety Check and relationship practice profile sharing still confirm to your inbox first. Maintainer documentation adds a Documentation and help map in docs/DOCUMENTATION_INDEX.md (who reads which doc).

Contextual help (help_system.py) now ties more topics to Help Center anchors, including Find your tool, AI Chat, Longitudinal Evaluation (with a pointer to the DSM-5-TR longitudinal report when available), Coach, Roleplay, Notifications (in-app bell plus email), and Journal vs Reflections. docs/HELP.md adds a matching Guided discovery and AI tools section. See docs/development/HELP_CONTENT_AUDIT.md for the updated matrix.

/book/… now uses the same CSS variables, critical token CSS, and Google Fonts as /p/… for your site template, so headings and components that rely on var(--font-family-heading) and palette variables match the public site. Loading, error, and booking-confirmation states use template colors instead of generic slate, indigo, or emerald. Public practice layout shares the same theme root component for consistency.

For the default Sage Editorial template, the header uses a frosted cream bar; nav links use muted ink with a sage hover state; the desktop Book online / Schedule pill is slate with a sage hover nudge like the design showcase. The hero adds a soft sage radial accent, a pill eyebrow on sage-tinted fill, a larger display title scale, body copy in muted ink, and when you set a tagline it shows as an italic sage line under the title on both the full site and the booking page (using var(--color-brand) for that line). Sage Editorial scheduling buttons use the shared label Schedule a session instead of “View availability,” and the /book hero eyebrow follows the same rules as the public site (online scheduling vs get in touch, etc.). Primary CTAs stay a darker green for accessibility (not pure #7d9b76 on white text). See docs/practice/REFLECT_DESIGN_SYSTEM_PORTAL_SCOPE.md §10–12 for the manual QA matrix and scope notes (portal and email HTML remain on their own track).

On /book/…, About, Services, FAQ, and Contact use the same editorial section-heading treatment as /p/… (accent rule under the title), and prose blocks use the same width and hyphenation helpers. The booking footer matches the public marketing footer rhythm: trust pills under Practice, a copyright rule using var(--color-border), UTC copyright year, and aligned grid spacing. The Why book here card uses the same clinical left-rail accent as the public site spotlight, and the booking panel form (fields, errors, Confirm booking / Back) uses your template colors and CTA styles instead of generic dashboard form chrome.

Every site template now ships a compact Schedule / Book onlineheader pill matched to that preset's primary palette (not only Sage Editorial). The booking sidebar title Schedule an appointment uses the same editorial section-heading accent as other blocks.

For Safety Check, YSQ-R schema profile, and relationship practice profile shares, you can use Change link settings to adjust expiry, max opens, and the viewer email step while keeping the same URL. Email link flows now start with Step 1 — Configure link access, then the recipient step, so limits are explicit before sending.

On Practice → Clients → Treatment plan, saved drafts now appear under Saved plans (with preview and ICD-10 summary), the list reloads after Save draft, unsigned plans can be reopened with Edit, and New plan (manual) opens the editor without running AI draft generation.

Under Practice → ePrescribe, when operations enables the integration flag, you can record prescriptions (drug, quantity, refills, optional sig, NDC, pharmacy NPI) for a client opened from the chart, see them in Recent prescriptions, and run a PDMP demo response. Entries are stored for your practice; live pharmacy routing and real PDMP still require a certified vendor connection.

On /who-this-is-for, the Open dashboard control no longer uses a light background with white text in dark mode (Reflect tokens flip --reflect-slate to a light color). Pricing and Explore the platform now use the same sage outline style as other marketing secondaries instead of low-contrast violet text links.

On Schema program → Your schema profile and the member Clinical report share area, you can keep several active timed links at once (each with its own expiry, max opens, optional viewer email step, and how long the emailed one-time open link lasts). Copy, email, or revoke per link without affecting the others (up to 15 active links).

Emails to recipients now summarize the actual rules for that link, and the viewer one-time email uses the real minutes you set—not a fixed “1 hour” message when you chose something else.

On Schema program → Take questionnaire, Auto-advance is now checked by default for new sessions (you can still turn it off anytime). If you previously disabled it, your choice stays saved in this browser.

On Schema program → Clinical report, you can now create, copy, email, and revoke the same timed public link as on Your schema profile. Recipients open your interactive schema profile (merged history); use Print or Download PDF on the report for this printable layout.

Schema program → Your schema profile share panel now shows expiry, email activity, and opens in your device's local timezone (API sends UTC with Z). The optional viewer email step (they enter their email on the public page for a one-time link) is explained in the panel and in Create share link. Email link on YSQ timed shares sends the profile link straight to the recipient; Safety Check and relationship practice profile still use Send confirmation (your inbox first, then recipient).

For anytimed share using this panel (including Safety Check and relationship profile), once a link exists you'll see Link access (this share): expiry, how many opens are allowed vs used, unlimited vs capped, viewer email step on/off, and a note that changing rules requires Revoke then Create share link.

On Schema program → Your schema profile and the Clinical report share area, Email link sends the timed view link directly to the address you enter—no separate confirmation email to your own inbox. Safety Check and relationship practice profile sharing still use confirm-to-self first. Old YSQ confirmation links in email continue to work until they expire.

From Roleplay → Practice profiles, you can email a timed public profile link the same way as Safety Check and the schema profile: we send a confirmation link to your inbox first, then the view link to your recipient. The share panel shows email sends and each time someone opened the link.

From Schema program → Your schema profile, send the public profile link by email: we email you a confirmation link first (like Safety Check), then send the view link to your recipient. The share panel shows who was emailed and each time someone opened the link. The domain radar chart uses clearer labels on small screens.

On Schema program → Your schema profile, create a timed link so someone can open the same interactive charts and mode map without logging in—optional expiry, max opens, and email verification for a one-time view link, like Safety Check and relationship practice profile sharing. Revoke from the same page anytime.

Help Center and crisis pages use semantic text tokens (text-primary, text-secondary, text-muted) for light/dark consistency. Across marketing, home, compliance, and many dashboard shells, borders and backgrounds that used bg-[var(--…)]/xx now use color-mix so transparency works with brand CSS variables (same idea as semantic Tailwind colors from var()). HELP.md and the in-app help service match on Safety Check timing patterns and exports. Docs: color audit Phases A–B notes.

The main tagline, crisis/support line, and copyright strip in the site footer now use full white with a light shadow so they stay readable on the slate-to-sage gradient (including dark mode). Section headings and support links are slightly stronger for contrast.

The YSQ-R clinical report (/dashboard/therapeutic/schema-program/report and practice …/ysq/report) now stacks PDF/print controls and footer actions on narrow viewports, full-width history chips, clearer ISST schema card headers, improved legend and narrative text wrapping, stacked top nav links, full-width error-state actions, and smoother horizontal scroll for the full-inventory chart. The schema program hub (/dashboard/therapeutic/schema-program) uses the same min-w-0 shell pattern as the profile page to avoid sideways overflow.

The YSQ-R profile page (/dashboard/therapeutic/schema-program/profile) now stacks primary actions, uses horizontal scroll for wide charts on phones, tightens chart margins and tick labels under the sm breakpoint, and improves text wrapping for summaries and domain lists.

On Relationship Profiles (Roleplay), each card can create a share link (copy or revoke from the dashboard). The API supports optional expiry, max opens, and email verification before viewing. Recipients use /share/relationship-profile/… (and a one-time verify URL when required) for a read-only view—no login—aligned with Safety Check sharing.

On Take the questionnaire, the strip with links to Schema profile and Clinical report refreshes from server history right when you complete a run (during the brief processing step before redirect), so the count and latest date stay accurate without leaving the page first.

Signed-in clients see Full [app name] in the portal header (opens your main site URL in a new tab). The booking-scoped portal overview also includes a short card pointing to goals, journal, Bible study, and other tools on the primary app host.

Visiting /portal/… without an active portal session now sends you to /portal/login with a sanitized return path (magic links with token or reminder_token still load first). The edge proxy shares one rule set with unit tests so behavior stays consistent. The login screen uses a two-column layout on desktop: themed introduction on the left and instructions on the right; small screens stay a single column.

The public /p/… hero shows short trust labels (secure portal, HIPAA-aligned workflows). When online scheduling is published and portal booking is not turned off, a subtle pulse badge notes that online scheduling is available. Reduced motion preferences disable the animation.

Under Visual style, AI site setup opens a dialog that runs the same wizard AI as quick setup: it saves your title, contact email, and assistant name, then fills tagline, theme, About, Services, and assistant label into the form for you to review and save. A booking slug on Provider profile is required.

In Admin → User Manager, the user table shows each account's plan, and Edit user includes a subscription plan control with every tier (Starter through Coach / Practice). Applying a plan updates feature access in the app; it does not change Stripe billing—use billing tools or support processes when payment must match.

The floating care guide on /p/… now reflects when your site lists public intake previews: scripted answers for Patient portal and Intake forms, and the short capabilities line, mention read-only overviews and where to find them. When live AI is on, the same detail is included in the assistant context.

When at least one intake template has Public preview turned on, My practice site → Page content lists those preview URLs (copy and open) next to your connected intake forms, with reminders if portal intake or a public slug is still needed.

On My practice → Intake forms, each form can opt in to Public preview. When enabled, visitors can open /p/…/intake/… (path shown next to the toggle) to see the template overview; submission still happens in the patient portal after your workflow assigns the form. Preview stays off by default so form slugs are not discoverable unless you enable them.

On the public /p/… page, Care tools for patients lists those previews under Intake & forms when portal intake is on, and the top nav adds Form previews to jump there.

In My practice site About/Services AI assist, Services adds Draft from intake notes (primary source) (uses optional notes as the main input; avoids inventing offerings when notes are empty). Quick edits adds Wizard-style full pass (this section) for About or Services to match setup wizard tone and guardrails. Same rate limit (practice_site_editor_ai).

On a client's page, Send portal invite still emails the magic link; after it succeeds, use Copy invite link to paste the same URL for secure handoff. The link expires in the same window as before and is single-use to sign in.

On /p/… pages, choosing Dark now applies a full dark palette for each visual template (CSS variables and ps-* utilities), not only Tailwind dark accents. With no saved choice, the site follows prefers-color-scheme; picking Light or Dark still saves per practice slug in the browser.

Invite and reminder links that include ?token= or ?reminder_token= on /portal routes are no longer blocked by the portal session check before the app can exchange them. Optional /portal/accept-invite?token=… redirects to the standard portal entry with the same token (for docs or templates that use that URL shape).

Unauthenticated visitors to most /portal/… routes are redirected to /portal/login with a safe next= return path; signing in still uses the secure link from your provider (with an optional Public scheduling page link). A Book flow at /portal/book walks through reason, intake reminder, open times, and confirmation using the same public booking API as /book/…. Practice site hero Book can send patients through portal login first. Stub pages cover register / password questions, and shortcuts like /portal/appointments land on the right dashboard tab. Mobile gets a bottom quick nav; the API still enforces portal tokens.

Public practice pages link to /portal/login?practice=… (your booking slug) for patient sign-in when that slug exists, with practice logo and theme when the site loads. After sign-in, patients return to their usual portal home for that practice. The dashboard My practice site editor combines page layout with About and Services in one section, raises the sticky bar above overlapping UI, and keeps only Discard changes at the bottom (save stays in the top bar).

In Practice setup step 3 (practice site), you can add optional notes, pick a visual template, then run Generate starter content with AI to draft a tagline, About copy with a Frequently asked questions block, Services HTML, a suggested help assistant first name, and a short template rationale. Review and edit the fields, then Save & continue. Rate limits apply (practice_site_wizard_ai).

Practice → My practice site is grouped into clear sections (identity, logo, look with a compact live preview strip, layout, content, contact, web address, SEO) with a sticky bar for published status, preview, and save. You can set an optional subdomain on the platform host (shown next to your apex domain); when NEXT_PUBLIC_PRACTICE_PORTAL_PARENT_DOMAIN is unset, that host serves your /p/… page. When it is set, the booking subdomain still owns / for the patient portal; other paths can map under /p/label/….

Public /p/… pages inject critical theme CSS earlier to reduce flash, add a Patient portal highlight card (feature pills and sign-in), use theme-colored header, footer, and layout (plus a 988 crisis line), and style /p/…/login with the same template tokens. Theme thumbnails use a responsive grid (2 / 3 / 5 columns).

Emails for appointments, reminders, intake, portal invites, and related practice flows now use a teal and slate layout that matches your public /p/… practice site. Subjects include the practice name, with links to the practice website and patient portal in the footer. The public site also has a dedicated /p/…/login page that explains magic-link sign-in and links to the portal; header Portal and patient CTAs go there first.

The floating help assistant on /p/… practice pages uses a taller, wider chat panel on phones and desktops, a slightly larger launch button, bigger topic chips, and a three-line message box so visitors can read and type more comfortably.

On your public /p/… practice site and blog, the header now includes Light and Dark buttons so visitors can switch appearance. The choice is remembered for that site in the browser.

The client portal is now built into every public template: the /p/… header includes a Portal link to your real portal URL (path or subdomain), the hero promotes Patient portal as a full button next to Book and Contact, the /book/… page header adds Patient portal, and each Public site look theme thumbnail shows Book, More, and Portal chips so the picker matches the full experience.

On Practice → My practice site, under Public site look, each theme (Clinical Indigo, Calm Sage, and the rest) now includes a mini preview—header bar, hero area, and sample buttons—using the same styles as your live public page and booking flow, so choices match what visitors see.

On /p/… template sites, the floating care guide now includes a Your message field: pick a topic, type a follow-up, and tap Send (or press Enter). You can also send a short question on its own (uses the Contact topic). Text is passed to the AI layer when available, with stronger text and chip contrast for readability.

The first time you save your public booking slug in the setup wizard (or provider profile), the portal now creates your practice site and fills site title (from your name or slug) and contact email (from your account). Step 3 is for review and edits; you can still customize everything under My practice site.

On Practice → My practice site, Show preview loads your public page at /p/your-slug in the frame so it matches the dashboard origin. If you use a custom domain, the public link still opens that hostname in a new tab; the frame cannot show it cross-origin because of standard browser framing rules. A note appears when a custom domain is set, and a timeout clears a stuck loading overlay if the frame never finishes loading.

On /p/… practice sites, when both address or phone and the contact form are shown, visitors first see location details, then tap Send a message to open the secure form. The header Contact link scrolls to the form area and opens it. #contact and #contact-form in the URL do the same.

Under Practice → Practice blog, create posts with title, URL slug, optional excerpt, and a mini HTML editor for the body. Check Published when a post should appear on your public site at /p/your-site-slug/blog and on each post page. The main practice header shows a Blog link when at least one post is published. List and post views are logged in the practice audit log. Run the create_practice_site_blog_posts_table migration on existing databases.

On Practice → My practice site, the About and Services mini editors include an AI action menu with many drafting and rewrite choices (first drafts, bullets, plain language, tone, length, soft calls-to-action, and more). Generated HTML is inserted as a draft you must review before saving. Requires OPENAI_API_KEY; model defaults to gpt-4o-mini (override with PRACTICE_SITE_EDITOR_MODEL). Requests are rate-limited per provider and logged as practice_site_editor_ai.

In Practice → Setup wizard (practice site step) and My practice site, you can set a short help assistant first name for the floating help bubble on your public page. Leave it blank to use the simple default Alex. The live AI reply uses the same name in its instructions.

The floating help widget on public /p/…practice sites calls the backend with the visitor's topic choice. When OPENAI_API_KEY is set, answers are generated with gpt-4o-mini (override with PRACTICE_PUBLIC_HELP_MODEL) using only public page metadata—no patient records. Requests are rate-limited and logged in the practice audit log as public_practice_help_assistant. If AI is unavailable, the same topics still show scripted tips. The assistant's display name is configurable (default Alex).

When a practice has an online booking slug, portal links from invites, reminders, telehealth join links, Stripe return URLs, and in-app notifications prefer that slug. With PRACTICE_PORTAL_SUBDOMAIN_PARENT_DOMAIN and NEXT_PUBLIC_PRACTICE_PORTAL_PARENT_DOMAIN set to your apex domain (and DNS + wildcard TLS for *.yourdomain), links use https://your-slug.yourdomain. Otherwise they use /portal/your-slug on the main site. See configs/nginx/portal-subdomain-wildcard.example.conf for an nginx sketch. /portal still works; subpages like /portal/messages work on both apex and slug hosts.

Each load of the public practice site JSON (/api/practice/site/public/… and public-by-domain) is recorded in the practice HIPAA audit log as public_practice_site_view (provider id, site id, IP, user agent, access path). Public contact form submissions were already logged as public_contact_form. The /p/… footer and booking footer now state that the experience is HIPAA-aligned, with encryption and audit logging, and link to the platform privacy policy—without claiming your practice is automatically compliant.

Public /p/… sites and the booking page footer now use patient / patient portallanguage (therapy and clinical care) instead of "clients" in headings, nav, hero CTA, Marlowe help topics, and autofill services. In-page anchor for the care section is #patients (update any old #clients links).

Loading the Practice section (including My practice site) no longer waits as long when /api/me, subscription, or intake-forms calls stall: stricter timeouts and clearer subscription loading avoid an endless spinner in common slow-network cases.

/portal is labeled explicitly as the provider client portal (appointments, telehealth, messages, intake, invoices from your therapist or coach). The page title and description no longer mirror the generic site tagline. If you landed here without a provider link, you will see short copy plus Sign in, Create account, and See all features so you can reach the full member experience. The portal header adds quick links to the main app when you are not signed in with a portal token.

On Practice → Configuration, set a default intake form for new clients who book through your public link (requires portal intake). Optionally turn on Booking email: portal payment notice so confirmation emails mention the client portal for invoices and online pay when practice invoice Pay (Stripe) is enabled. Run the provider profile migration if those columns are not on your database yet.

The floating help on /p/… introduces Marlowe, a care guide that shows a short typing state, then streams plain-text answers before revealing the full formatted reply. Topics use your saved tagline, about, services, booking link, and portal flags when explaining what visitors can do. The public site and booking info APIs now include contact email when set on the practice site (for display alongside other contact options).

Public provider pages at /p/… and the public booking experience at /book/…no longer show the platform marketing footer, so visitors only see the practice's own layout and footer where applicable.

On Practice → My practice site, reorder main public sections (hero, client tools, about, services, contact) with drag and drop; add optional sections back with one click. About and Services use a simple in-page editor (bold, italic, lists, links). Contact can include phone and a multi-line address shown on /p/…. Run the new database migration so practice_sites has page_layout_json, contact_phone, and contact_address_text.

The public microsite at /p/… now has a full hero, navigation, and a Care tools for clients section that explains the client portal (appointments, telehealth, invoices, Pay, onboarding), and highlights secure messaging and intake when those features are enabled for the practice. The public site API returns those feature flags. A floating Help button opens quick, fixed answers about booking, the portal, billing, and privacy—clearly labeled as guidance (not medical advice and not a live AI on your health data). Booking, email, and portal entry points stay one tap away.

The /book/… scheduling link now uses a fuller landing layout (hero, trust, FAQ, contact) and pulls optional title, tagline, logo, about, services, and contact email from your saved practice site when present. Booking info includes a site template id (stored on the practice site; run the new migration on Postgres). Ten preset themes are defined in code for future themed landings. Success and error states match the new chrome. Public routes /book, /p/…, and root static files (service worker, manifest, robots, sitemap) stay reachable without a login redirect. The HIPAA BAA page uses server metadata and a client document view to avoid build-time errors on that route.

Practice → Configuration, Billing, and the setup wizard show a short message for clinicians: Pay waits on server-side Stripe setup; you can still manage invoices; use Admin → Stripe setup if you are an administrator, or ask whoever hosts the portal. Env vars, verify_practice_stripe_env.sh, and the deployment runbook are listed on Admin → Stripe setup for operators. Billing error toasts and API responses still describe the technical requirements when Pay is unavailable.

On Practice → Setup wizard, identity, booking slug, practice site basics, client portal booking, Stedi, and Google Calendar can be completed in-page with Save & continue and smart suggestions. Full profile and configuration pages stay available when you need every option. Step 3 shows an example site title derived from your account name or booking slug.

On Practice → My practice site, empty fields are filled once from your provider profile (credentials, billing city/state, portal options) and your account (name, email). The platform logo URL is suggested only when you do not have a saved site yet.

A progress effect was re-running continuously and locking up the page. Practice → Setup wizard now uses stable completion tracking so you can open steps, use Go to setup, and navigate like normal.

Help for the main Dashboard uses static guidance instead of routing through generic help that triggered a slow per-request enhancement. Audit-related logs also store client IPs in a consistent dotted-IPv4 form where possible.

On Couples → Shared memories, you can attach up to 20 images when creating a memory or add more when editing. Remove individual pictures before saving, or clear a legacy single photo with the existing checkbox. Deploy the new shared_memory_photos migration on Postgres so existing single-photo memories stay linked.

On Messages and Devotion Connect, the extra breadcrumb line under the dashboard bar is removed. Use the bar's Homelink for dashboard home; the page title stays in the screen's own header so labels are not repeated.

On Couples → Shared Calendar, tap a day number in month view to open the add-event form with the same repeat choices as date night scheduling (every week, every other week, monthly spacing, or every N weeks). Use View on a cell to jump to that day; week and day views include Add event as well. Recurring saves now send interval data consistently for the server.

The Practice setup wizard now follows the same checklist as the Practice hub and Calendar: provider identity (full US address and credentials), public booking slug, practice site title and contact email, client portal online booking, Stedi, invoice payments when your server enables them, and Google Calendar sync. Saved step index migrates from the older four-step wizard; completion is step 8. Practice → Configuration uses the same seven gates in its readiness grid (with the compact setup banner when anything is still incomplete).

On Couples → Daily Prayer & Bible Study, an AI reading block lets you choose a voice and listen to today's focus (passage plus the tailored explanation) using the same browser read-aloud as Bible Study. Read here still opens the full chapter with verse listening.

Pages that use your browser's speech synthesisfor read-aloud now share one voice list update channel, so voices load correctly when many controls are on screen. Unmounting a small read button no longer cancels speech for the whole page. Leaving Devotion Connect, the couples devotional, or Michael & Emily mail stops playback. If your browser cannot speak, you'll see a short note instead of a missing button.

The Help panel (header and pages) now shows a two-column layout on large screens: overview and FAQs from the server, optional feature screenshot, quick tip, and guide links. Suggested questions are always visible as chips. Routes like Communications Wizard, AI Chat sessions, Practice, and Find your tool pick the right topic automatically. The Help Center anchors Communications Wizard for deep links. Backend help resolution normalizes hyphens, prefers longer topic keys, and avoids treating AI Chat as reflection Chat.

On Practice hub and Calendar, the setup reminder now checks the full gate used elsewhere: provider identity (including 10-digit NPI and billing address format), booking slug, practice site title and contact email, client portal online booking, Stedi, invoice payments when enabled on the server, and Google Calendar OAuth. Practice site branding has a stable anchor for deep links, and Configuration scrolls to the client portal section when the URL hash is #practice-config-client-portal.

Where invoice Pay is explained — Practice → Configuration, Billing, and the setup wizard — the copy now points to Admin → Stripe setup for key entry, the server script verify_practice_stripe_env.sh, and the deployment doc path docs/deployment/PRACTICE_STRIPE_PAYMENTS.md for operators.

Under Practice → Reports, the summary now includes encounter volume, billed and telehealth encounter counts, paid and denied claim counts, and A/R aging buckets for open pipeline claims (submitted or ack received) by service-date age. CSV export includes the same fields, and a Related dashboards section links to calendar, billing, prior auths, waitlist, compliance audit logs, and tasks.

After a release, browsers sometimes request an old split chunk that no longer exists, which showed as Loading chunk … failed on Couples. Lazy-loaded widgets now trigger a cache-busting reload when that happens (before the error screen), chunk detection includes the Next.js error prefix, reload limits are slightly more forgiving, and static asset Cache-Control headers apply in the intended order for /_next/static.

The Communications Wizard dashboard uses an updated header (quick Help link, partner chip, short description), a visible private thread notice with lock icon, and a two-step progress row where Compose returns to editing when you are on preview. On phones, compose appears above the inbox so writing is not buried below the message list.

On Couples → Connection, the shared calendar uses month, week, and day views with navigation, plus a panel for each partner to set recurring weekly blocks next to date nights and other events.

Recurring blocks are framed as a general weekly schedule (work, childcare, focus time, sleep, and more), not work-only. The API adds GET/PUT /api/couples/schedule-blocks as an alias of /api/couples/work-schedules; when a block is saved without a label, it defaults to "Schedule".

On Couples, each upcoming or past shared calendar row has an Edit action that opens the date-night form with existing details; saving updates the event for both partners. The API supports PATCH /api/couples/events/{id} for partial updates.

Each memory now shows a labeled Memory line (when the moment happened—date, or date and time if the API stores a full timestamp) and a Saved line with the exact local date and time the entry was created.

The Partner activity card refreshes when you come back to the tab or reconnect, and the API avoids stale caching. Each listed tool is a link to that area of the app; More detail (exact times) shows local timestamps. What this shows opens Help with a full explanation of partner visibility and the 30-day window.

If you have an active relationship agreement that is not fully signed, a daily reminder can notify the partner(s) who still need to sign (in-app notification and email according to your reminder preferences). Notification links open Couples → Tools and scroll to the agreement. Server operators can schedule scripts/run_contract_sign_reminder_notifications.py once per day (see install_contract_sign_reminder_cron.sh).

GET /api/couples/dashboard could return 500 on a second requestwhen the in-memory dashboard cache held a database object for the partner's last mood check-in. After the request ended, that object was no longer valid for serialization. The API now stores a normal response shape for partner mood before caching, so repeat loads and the couples dashboard stay reliable.

When a browser still has old HTML but new JavaScript build output, a script under /_next/static/chunks/ can fail to load with little or no error text. The early layout script and ChunkErrorHandler now treat those script URLs as chunk failures and run the same cache-busting reload path, which reduces broken login and other routes right after releases.

The duplicate Admin → Calendar Sync Setup card was removed from the main Practice overview. The Admin hub and admin side nav no longer list Calendar sync as a shortcut. Google Calendar OAuth and two-way sync are configured from Practice → Configuration (Two-way calendar sync), which links to the same setup wizard when an admin needs it. The wizard URL is unchanged for bookmarks and support.

Public pages now ship server-rendered JSON-LD for Organization, WebSite, and (on /marketing) FAQ and SoftwareApplication, so crawlers see markup in the first response. Root metadata includes default Open Graph / Twitter images, expanded keywords, and shared descriptions via lib/seo/siteDefaults. The sitemap lists additional important URLs (privacy, terms, cookies, NPD Watch); robots.txt explicitly allows key paths. Help, features, about, pricing, compliance, changelog, and marketing segment layouts use consistent getAppName() titles and preview images. Invalid SearchAction markup was removed where there is no public search page.

Admin → Operations and backups centralizes PostgreSQL backup (with a post-backup acknowledgment checkbox), read-only platform status (database type, whether an encryption master key is configured, MFA TOTP env), and short runbook reminders for migrations and Phase 2 encryption. The main Admin page now groups every admin tool (settings, Stripe, calendar setup, token limits, analytics hub, and more) into clear sections, with links to Operations from service controls. The admin side nav also includes a direct Operations link.

Admin → Calendar Sync Setup now shows the site base URL from server config, the exact Calendar API scopes providers approve, step-by-step OAuth consent screeninstructions (test users, External vs Internal, verification), a troubleshooting checklist, and a quick link to Google's consent console. The authorize URL uses proper query encoding. Practice → Configuration → Two-way calendar sync explains what to do if Google shows Access blocked and points admins to the same setup page.

The Clients table Intake column now shows a short status (for example all complete or how many are pending). Click it to open assignments: View shows submission JSON and the form template; PDF downloads a printable summary (requires WeasyPrint on the server). Edit client shows the same summary with links to manage intake.

Hover (or focus) hints are expanded across the app where they help most: Practice hub tiles use the shared tooltip component; the practice sidebar and command-palette page list use native titles with the same copy from one data source. Dashboard home link, bottom nav, and the header search field include clearer descriptions. The shared Tooltip supports wrapped text and sits above the top bar so it is not clipped.

While the Treatment plan draft modal is open, the form stays on screen during generation (inputs disable instead of disappearing). The dialog only closes via Close or X—not backdrop click or Escape. Section tabs show a reminder if you try to leave before closing the editor.

On Practice → Clients → Treatment plan, generating a draft now also runs AI-assisted ICD-10-CM and DSM-5-TR suggestions so the comma-separated code field and documentation block fill in automatically (everything stays editable). The draft model uses a consistent outline: Draft Treatment Plan, Goals, Interventions, and Plan Summary. Suggestions are aids only—verify ICD-10-CM and DSM-5-TR before billing.

The Practice → Clients table is a single compact row per client. New columns (wide screens) show self-pay totals from the ledger—billed, paid, and balance—excluding insurance ERA lines. Edit client now includes intake forms: open a form or edit due dates without leaving the dialog.

In week view, click anywhere in a day column (including empty cells) to open Add appointment with that day prefilled (default 9:00–10:00 local). In month view, use + Add on each date. In day view, use Add appointment this day for the date you are viewing.

On Practice → Telehealth → Recording library, you can Delete a cloud recording immediately (Daily API) or Schedule delete to remove it automatically after a date and time you choose. Scheduled removals run when the library loads (e.g. refresh). Run the create_practice_telehealth_recording_retentionmigration on Postgres so scheduling persists. Custom S3 buckets may still need manual object cleanup per Daily's documentation.

The Practice sidebar Search hint no longer uses the Unicode Command symbol (⌘), which often showed as a missing-character box on Windows and Linux. It now shows Ctrl+K or Cmd+K (after load) depending on your system. Keyboard shortcuts help uses the same plain-text style.

On Practice → My practice site, the logo field placeholder and Use platform logo URL use your deployment's canonical HTTPS site (from NEXT_PUBLIC_SITE_URL, defaulting to https://toxicpartner.com/logo-icon.svg) so it matches the live domain instead of a generic example link.

Assigned intake forms now have Edit due date on Practice → Clients (per assignment) and on the client Intake tab. Use PATCH /api/practice/intake-assignments/{id} with due_date (or null to clear).

On Practice → Claims, click the claim summary (Claim #…, encounter, plan, dates) or Edit to open an editor: update denial reason / office notes, save, or run Check status, Suggest appeal, and Resubmit from the same dialog. The URL can include claim_id=… (used by denied-claim links on the practice hub).

Practice → Clients now shows gender and city/state (on wide screens), whether the client portal is linked, the latest encounter service date (link to Encounters), the next scheduled appointment (link to the client chart), and a count of unsigned treatment plans (link to the Treatment plan tab). The list API supports include_clinical_summary=true for those clinical columns.

On a client's Treatment plan tab, the generated draft opens in a modal you can edit. Use Suggest ICD-10 & DSM-5-TR to add AI-assisted billing documentation from the plan text (review before claims), edit comma-separated ICD-10-CM codes, and Save draft to store the plan on the chart. Copy to clipboard includes the plan and any codes you entered. Run migration add_practice_treatment_plan_billing_codes.py on existing databases.

Practice → My practice sitecould stay on "Loading…" or show a generic service-unavailable message when the database was missing the logo_url column on practice_sites. Deploy migrations now add it automatically, and the API repairs the column when needed.

Some environments were missing the calendar_default_view column on provider_profiles, which caused Practice → Provider profile to show a migration error. Deploy now runs the matching migration automatically, and the API can add the column safely when it was omitted from an older migrate pass.

On Practice → Reports, summary tiles (billed, paid, pending, denied, rates, unbilled count, claims pending) and the Breakdown by payer table cells link to Encounters, Claims (with status or payer filter where applicable), or Billing. Encounters supports ?unbilled=1 to show draft and ready-to-bill rows. Claims supports ?payer_name=… (exact payer name, case-insensitive) with a clear banner on the page.

On Practice → Client waitlist, each row has an Edit action that opens a dialog to update notes, workflow status (waiting through removed), and an optional link to the appointment you offered from the calendar. Clearing the offered appointment removes the link. Status is shown read-only in the table; use Edit to change it.

Practice → Configurationnow loads your feature toggles as soon as your provider profile is ready. Calendar sync and invoice payment status still appear in the readiness sections when those checks finish, so a slow or stuck secondary request does not block the whole page on "Loading configuration…".

On Dashboard → Messages, if you use Practice Management as a provider, Clients is now its own tab beside Friends and Conversations, so the list is easier to scan. Linked clients open a DM; others show how to open their chart in Practice. Your therapists for patients stays in the compact strip above the tabs.

Tapping a row in Inbox / Sent / All still selects the message and now scrolls the reading pane into view on small screens. For messages you sent, a Partner access & audit panel lets you add time (+24h / +7 days), set a new expiry, remove expiry, raise or remove the max open count, and see every open with timestamps.

On Practice → Calendar, the whole appointment row or week cell card is clickable to edit time and type—not only the text block. The Video link still opens telehealth on its own without opening the editor.

On Dashboard → Messages, if you are a patient with a linked portal account, Your therapists appears in a compact strip above the tabs so you can open a direct message with your provider quickly.

On Practice Management → Activity, each Your checklist row is a single line: the task link and date sit beside the actions. On narrow screens buttons show Dismiss and 24h (full wording in screen reader labels); from sm up they read Dismiss for now and Snooze 24h. Long titles truncate instead of wrapping to a second row.

The Messages screen uses the URL for optional message prefill. That path is now wrapped in a Suspense boundary, matching other dashboard pages, so Next.js can render the route without an internal server error during prerender or navigation.

Footer Legal, Explore, Learn, and related columns use a shared header height so section titles and underline rules line up; the brand block top band matches that row on desktop. The copyright / company line uses clearer left–right alignment above the fold width.

On Practice Management, the AI assist section lists more one-tap links to real workflows: ICD-10 suggestions, bulk note generation, billing CPT from duration, claims and appeals, intake assignments, telehealth transcription and recordings, and AI feature toggles in practice config.

The welcome banner's Maybe latermenu was clipped so only the "Snooze or dismiss" label appeared. The full list (remind tomorrow, 3 days, 7 days, hide until next visit, don't show again) is visible again.

The public site footer uses the full content width more evenly: brand and intro on the left, five balanced link columns (Legal, Explore, Learn, Support & safety, Feedback), clearer section headings, and a cleaner copyright row on large screens.

On Practice Management, each Activity checklist row has Dismiss for now (hidden until you close the browser tab) and Snooze 24h. A toast offers Undo. If everything is hidden, use Show all again. Preferences are stored on this device only; links still open the real work (encounters, clients, claims, waitlist).

The hub and calendar setup callout uses Welcome! Let's get your practice set up copy with Start setup wizard as the main action. Maybe later opens a menu: Remind me tomorrow, in 3 days, in 7 days, hide until next visit, or don't show again (reopen from Configuration anytime).

In Practice → Calendar, list, day, week grid, and month views now open the Edit appointment dialog when you click the appointment row or card (time, type, and duration). The Video link still goes to telehealth without opening edit. Month view: tap the day number for day view, or tap an appointment line to edit.

Practice → Billing no longer points to site Admin for Stripe setup. The amber banner links to Practice → Configuration → Invoice payments (Stripe) and the setup wizard, which now includes an invoice-payments step. Readiness on Configuration includes whether Pay is available. The Admin Stripe wizard remains for operators who set server keys; step 1 notes where providers see the checklist.

The Practice → Clients table now includes an Intake forms column: each assigned form is listed with status, and the name links to Practice → Intake for that assignment. Clients with no assignments show None assigned. After bulk assign, the list refreshes so new assignments appear without a manual reload.

Public practice pages at /p/your-slugno longer inherit the main app's marketing title and description. The tab title and meta tags now use your Meta title / Meta description when set, otherwise Site title and Tagline. The practice site editor explains those fallbacks under the SEO fields.

Slug fields now follow the same pattern: a suggested default (from context) stays visible up front; expand Advanced to edit the slug and use Reset to suggested slug when needed. Applies to Practice → Provider profile (booking link), Practice → Intake (new form wizard), and Journal → Manage journals (new custom templates). Shared helpers live in lib/urlSlug.ts.

On Practice → Clients and the client profile, DOB from the API (YYYY-MM-DD) now displays as that same calendar day in your timezone. Previously, parsing treated the value like UTC midnight, which could show the previous day in the Americas.

Connect Google Calendar and Sync now for two-way calendar sync now live under Dashboard → Practice → Configuration (section: Two-way calendar sync). The Calendar page links there; after Google OAuth, you are returned to Configuration. Authorized redirect URIs in Google Cloud are unchanged (/dashboard/practice/calendar).

On Journal in dark mode, primary actions that use a light fill (e.g. Guided reflection, Start free write, Finish entry) no longer show white-on-white text. They use the same contrast approach as the rest of the page.

The ⋮ menu on Practice → Encounters (edit note, generate note, suggest codes, delete, etc.) now opens in a fixed overlay so it is not cut off by the wide table scroll area or hidden under the bulk-selection bar. The menu closes on scroll, resize, Escape, or click outside.

The My practice site settings page (Practice → My practice site) uses the same sync pattern as other practice APIs so GET /practice/site does not block the server event loop. The client applies a 30s request timeout and clears a stuck Loading… state after a short failsafe with a clear retry message.

The Encounters page loads the encounter list and the client list together. If one call fails (timeout, gateway HTML page, or API error), the message now starts with Encounters: or Clients: so you can match it to the right request in the browser Network tab. When one side succeeds, that data still loads.

The Practice Management hub no longer retries the same slow tasks request multiple times on client timeout, and it clears Loading activity… with guidance if the call still does not finish. The GET /practice/tasks endpoint skips expensive full-table counts when your overdue notes, unsigned plans, and denied claims lists are below the page limit, which speeds up large databases.

Practice clients now expose SSN through a normal Python property (backed by encrypted storage), which removes server errors that auto-logged as PracticeClient has no attribute ssn when listing clients. On Admin Bug Reports, Verify on a fixed ticket now sets status to closed after recording verification notes, and the page loading watchdog waits long enough for large closed-list fetches (up to 90s) so the UI does not falsely time out early.

The Bug Reports admin page uses the same admin rule as the rest of the dashboard: users with admin access (owner emails or is_admin granted by an owner) now receive all reports from GET /api/bug-reports and can use admin-only actions (delete, auto-fix, generate fix, verify, etc.). Previously only a fixed email list matched the API, so some admins saw an empty or partial list while the UI showed them as admin.

The backend documents and enforces PostgreSQL for development, staging, and production when the database module loads, with EMILY_ALLOW_SQLITE_RUNTIME for maintenance tooling only. pytest uses Postgres at import, a session metadata template, and fast per-test databases via CREATE DATABASE … WITH TEMPLATE. Legacy one-off migrations no longer branch on SQLite. See docs/architecture/DATABASE_RUNTIME_POLICY.md.

The Telehealth dashboard loads today's appointments and the waiting room in parallel. The waiting-room API now uses portable SQL ordering (no NULLS LAST), runs in the thread pool like other practice lists, and returns an empty lobby if the waiting-room table is not migrated yet. The practice clients list uses the same sync pattern so it does not block or confuse the event loop. See docs/practice/TELEHEALTH_PRACTICE_PAGE_503.md for operations notes.

The prior auths list endpoint now runs in the thread pool (same pattern as calendar and billing), skips individual rows that fail serialization instead of returning a 500 for the whole list, and uses portable SQL ordering. The dashboard page loads auths and clients with allSettledand shows an explicit "list unavailable" state with retry when the auths API fails, so it does not look like an empty roster.

The calendar page loads appointments and clients together. The appointments API used the same pattern as other practice routes that could block the Python event loop under load, so the UI could sit on "Loading appointments…" while lighter calls (like the 24h reminder count) still returned. The list handler now runs in the thread pool; the calendar page also uses a Suspense boundary for URL params and allSettled so a partial failure does not discard the other request.

The practice hub shows an explicit loading row while schedule and summary data load. Client chart routes use a proper suspense boundary for tab URLs. Telehealth caps the client roster fetch like other practice pages to avoid slow loads. Practice site shows an inline retry if the site settings request fails, and the compliancedashboard no longer shows a perpetual "Loading…" label when metrics are simply unavailable.

Production builds could emit an invalid <script src="…css"> tag alongside normal stylesheets, which triggered MIME-type warnings and noisy preload messages in DevTools. Webpack chunk settings were aligned with Next.js defaults, and the global error screen now loads heavy reporting code only when an error actually occurs.

The practice clients API could fail with an internal error because the database model was missing SSN and encrypted contact field accessors that the API already expected. Those fields are now wired on PracticeClient, so Billing and other pages that load clients should work again when your database has the encrypted columns from the usual migrations.

Updating an appointment's start or end time now triggers the same client notifications as a status change (when a client record exists). HIPAA audit PDF export date-range lines were also corrected for Python compatibility so the report generates reliably.

Opening a client you can't access used to send you to the main dashboard without explanation. The page now stays on the client URL and shows access denied with links to the Practice hub, Clients list, or main dashboard.

The practice HIPAA Audit Logs table now includes a User ID column (the same user id used for the provider/staff filter), so each entry shows which account performed the action without opening exports.

On the Practice Management home screen, the three summary tiles—Appointments today, Encounters to bill, and Clients—are now links. Click or tap a tile to go to the matching screen (calendar, encounters list, or client list). Keyboard users get a visible focus ring.

The practice compliance HIPAA Audit Logspage used to refetch on every filter change, so the list often stayed on "Loading audit logs…" while typing. Filters now apply when you click Apply Filters (and when you change pages or clear filters). CSV and PDF exports match the filters used for the table.

The prior auths page no longer shows a duplicate Loadingline next to "Loading prior authorizations…" (one clear status for screen readers). Clinical staff can read and manage prior authorizations so encounter linking matches backend permissions; the prior-auth list API enforces the same read permission.

The billing dashboard loads invoices, clients, and encounters together. Billing staff can now list encounters (read-only metadata for invoicing), which avoids failed loads for that role. Invoice list responses also coerce odd encounter_ids or due_date values so the API does not fail while serializing rows. If you still see an error, check server logs or auto-logged bug reports for the underlying exception.

The Journal page no longer ties deep link handling to the search-params object identity (which could change every render). Opening the journal from a bookmark or shared link should behave the same without locking up the tab.

On large screens, the Practice left navigation now stays pinned under the dashboard header while long pages (for example Configuration or profile) scroll. If the menu is taller than the viewport, it scrolls inside the sidebar instead of moving off screen.

Starting a telehealth session from the practice telehealth page without picking a specific appointment could trigger an internal server error after the video room was created. The API now handles that path correctly.

When a Skeleton is given an aria-label, it is no longer marked aria-hidden; it uses role="status" by default so assistive tech can announce labeled loading placeholders. Decorative skeletons stay hidden as before.

The compliance badge grid now includes Official program links and a section listing authoritative sources (HHS HIPAA, HITRUST, PCI SSC, AICPA SOC, ISO 27001, EU data protection). This supports transparency without embedding third-party certification marks, which require separate trademark permission.

Two-step TOTP sign-in and practice MFA enforcement are controlled by MFA_TOTP_ENABLED on the API (unset/false by default). The login UI uses NEXT_PUBLIC_MFA_TOTP_ENABLED at build time so it stays in sync. With MFA off, password login always completes a full session; users can still call POST /api/me/mfa/totp/disable with password only to clear stored MFA. See docs/compliance/MFA_TOTP_IMPLEMENTATION.md.

Download actions for legal documents now serve PDF files (generated from the same markdown source as the on-page view). Use /api/legal/*.pdf or the Download PDF buttons on legal pages and the BAA page. In-browser viewing still uses markdown for fast rendering.

The public /compliance page now includes the site header with the real app logo, shared footer, dedicated page metadata for search and sharing, framework badges with consistent artwork, and a Sign up call-to-action that goes to /register (instead of the dashboard). The footer Product section is split into two balanced columns on all screen sizes for easier scanning.

The site root (/) is now a home hub with clear links to the main marketing page, /marketing/practice (healthcare providers), /marketing/personal (individuals and couples), compliance, features, and pricing. The footer Product section includes the same entry points. Public-route logic used by proxy.ts now treats all /marketing/*, /legal/*, and /compliance/* paths as public (no login redirect).

Practice Management adds HITRUST-oriented controls: optional data classification on client records (create/update), APIs for portable media transfer logging, vendor risk assessments, a compliance risk register, and mobile device policy acknowledgment. Run the new database migration documented in docs/compliance/HITRUST/PHASE_11_ADDITIONAL_DOMAIN_CONTROLS.md before using these endpoints in production.

Practice Management now includes a persistent sidebar (with mobile menu), ⌘/Ctrl+K search for clients and pages, and a keyboard shortcuts panel (⌘/Ctrl+/ or ? outside fields). Encounters supports multi-select with a claims queue (submit across multiple encounters in order) and bulk AI note generation. Clients supports selecting multiple rows to assign the same intake form (also via /dashboard/practice/clients?bulk_intake=1). Claims, Clients, Intake, and the portal loading state use skeleton placeholders where appropriate.

Layouts wrap practice and client portal content in an error boundary for safer failures.

Client portal Messages now use the same portal magic-link session as the rest of the portal (not a full Reflect account session). The portal home response includes unread message count so one request can show the badge. Dashboard Notifications shows distinct icons for practice types (appointments, invoices, messages, intake, claims).

Under Practice → Compliance (admin): new Compliance officers and Workforce training pages list assignments and training records and support creating entries via the existing APIs. The compliance home page now loads live metrics (score, audit totals, pending BAAs, training due soon, open breach investigations) instead of placeholder stats.

Accounts can enable time-based one-time passwords (TOTP) after signing in: use POST /api/me/mfa/totp/begin and POST /api/me/mfa/totp/confirm with an app such as Google Authenticator or 1Password. With MFA on, email/password returns a short-lived step-up token; the login page asks for a 6-digit code, then POST /api/token/mfa-verify completes the session. Run python migrations/add_mfa_totp_enc_columns.py on existing databases. See docs/compliance/MFA_TOTP_IMPLEMENTATION.md.

For organizations using Practice Management: entitlement and SOC 2 report access now recognize active HIPAA security/privacy officers (in addition to admins). Compliance officers can also view other users' MFA status where appropriate. Electronic signatures verify your account password when you provide one (session sign-in otherwise). Reproductive health attestations are stored with client and disclosure metadata; run migration migrations/add_reproductive_health_attestation_client_columns.py on existing databases (or the full compliance migration script).

The partner message area now behaves more like an email inbox: each row shows From, a subject line from the first line of the message, date, and in Sent a compact Not read / Read / Expired pill. Use Search (including shortcuts unread and pending) and Refresh to reload messages. Messages you sent show explicit Not opened yet copy when your partner has not opened the wizard thread. Reply on a partner message quotes an excerpt into compose.

Stripe client refresh now reads payment flags from the current environment (works cleanly with Admin → API Keys). Admin Stripe setup step 3 points to API Keys for adding STRIPE_SECRET_KEY without SSH. Docs add a turn-on checklist and verify_practice_stripe_env.sh for the prod env file.

Journal: Suggest / Next / Finish toolbar on phones sits above the dashboard bottom nav (no more hidden behind it), with extra scroll padding. Devotion Connect prayer form: Partner and Friends labels and checkboxes use higher contrast. Practice → Billing: payment-disabled notice links to Admin → Stripe setup and explains Stripe keys; Pay still needs a valid server-side secret key and backend restart.

Prior auths now support full edit (including reassigning client and correcting sessions used), status (pending / approved / denied / cancelled), and authorized procedure codes (CPT/HCPCS). Denied or cancelled auths cannot be linked to new encounters; the encounter form hides them unless already attached. Run the DB migration migrations/add_practice_prior_auth_status_and_procedure_codes.py on existing databases.

Practice → Calendar now includes Day and Month alongside List and Week. Month view shows a full grid with a few appointments per day; tap a day to open that day. Navigation (previous / next / today or this month) follows the active view. The week grid still loads even when the week has no bookings.

On the client portal video call, the provider's tile now appears first on narrow screens (your camera below), and track auto-subscribe is set explicitly—so patients see their therapist without scrolling past a full-height self view.

On small screens, the prayer list and composer no longer split into two very narrow columns: sharing options stay full width until larger breakpoints, filter pills stack with shorter labels, and prayer text uses normal word breaking (no aggressive mid-word breaks). Partner thread bubbles use a clearer max width; wizard review and ACTS actions stack full width on phones.

Prayer Requests now uses a calmer page background, skip-to-content link, a structured composer (draft, type, partner + friends sharing, primary action), segmented filters, and richer list cards with type badges. The friends-sharing option appears in the form so it matches what is saved. Prayer Wizard gets the same shell on the main screen; the partner thread uses chat-style alignment (your messages on the right), a larger scroll area, and a stacked composer on small screens.

When you open a conversation with your own account, the header now reads Notes to yourself with a short privacy line, the empty state explains the thread, timestamps say You sent this at … instead of “Sent to [your name]”, and cycle-breaker nudges stay hidden in that context.

Practice and portal telehealth screens, recording library, session notes/transcription hints, and API error messages now use neutral language (your video service / telehealth video) instead of naming the underlying infrastructure vendor. Admin → API Keys labels use the same approach; environment variable names are unchanged for operators.

Under Practice → Telehealth, intro copy is toned down to read like clinical workflow notes instead of marketing: shorter overview, billing modifiers called out with a "confirm with your biller" line, recording/consent bullets, and clearer waiting room and quick-visit sections.

Pages no longer ask the browser to run stylesheet files as scripts (which could cause console errors). Canonical URLs, Open Graph, and sitemap defaults now use the live production domain when a build omits NEXT_PUBLIC_SITE_URL—override that variable for white-label or alternate hosts.

Loading Admin → Calendar sync (/dashboard/admin/calendar-sync-setup) could return a 500 if reading stored OAuth settings hit a database error. The API now falls back to environment variables for status, returns a clear 503 with guidance when saving credentials fails, and the page has an error boundary with recovery links. Calendar sync also appears in the admin tab bar.

Client waitlist ( /dashboard/practice/waitlist) was calling GET/POST /api/practice/waitlist but the waitlist router was not included on the practice aggregate router, which produced Not Found in production. The waitlist routes are now registered.

Waitlist status updates use PATCH to match the backend. The Practice Hub / Calendar / Add buttons use a single horizontal row with no awkward line breaks on narrow phones.

After joining, the header could stay on Connection: Checking…because Daily's networkState defaults to unknown until a network-quality-change event—and on some mobile browsers that event is delayed or absent. While quality is not yet measured, the line now reads Connection: In call. Good, Fair, and Poor still appear when Daily reports them; reconnect behavior is unchanged.

The call screen could show No active call.for a moment after device check because Daily's useMeetingState() can be null before the first state update. That gap now shows the same Loading… / Joining call… UI as the rest of the join flow, and the video provider is still initializing when the Daily object is not yet in context.

Starting a new call after a long load resets the connecting timer so the timeout message does not appear immediately from the previous attempt.

Under Practice → Telehealth, open Recording library to list Daily cloud recordings for video rooms you start from the dashboard, and open time-limited download links. Older sessions may not appear until you start new calls after this update (practice_telehealth_room_sessions migration on Postgres).

Some joins still failed with You are not allowed to join this meeting even when the link included ?t=…. The call UI now passes Daily the room URL and meeting token as separate fields (recommended by Daily), which avoids fragile query-string handling. Start a new session if the previous token already expired.

The in-app Help Center, contextual help API (telehealth topic), and docs/HELP.mdnow include telehealth for providers: starting sessions, patient vs host links, token expiry, and "not allowed to join" troubleshooting.

The in-call screen could stay on Loading… because Daily prejoin is a Prebuilt feature while this app uses a custom Daily call UI (device check happens before join). New video rooms now default enable_prejoin_ui: false, and the client kicks join() if the call object idles in new / loaded. Start a new telehealth session after deploy so rooms pick up the updated Daily room settings.

Private Daily.co rooms require a meeting token on the join URL. The backend was sending token expiry (exp) at the wrong JSON level, so Daily returned 400 unknown parameter 'exp'and the UI showed "Video room creation failed." Tokens now match Daily's API (expiry inside properties). If token creation still fails (key or permissions), messages are mapped more specifically than the generic creation error. See docs/features/TELEHEALTH_CONFIGURATION.md § Troubleshooting.

Starting HIPAA video for an appointment now uses a unique Daily.co room nameeach time (instead of reusing the same name), so "Video room creation failed" should not recur when you start another session for the same appointment. If Daily rejects a room, you get a clearer message to retry or check your API key and Daily domain.

The in-call header showed Connection: Checking… twice because a hidden screen-reader region repeated the same text as the visible status row. The call UI now uses a single bar with aria-live so layout stays clean and assistive tech still gets updates.

If Pipecat Cloud returned a Daily room without a meeting token, the app used a bare URL and Daily showed You are not allowed to join this meeting. Those responses are now rejected so a proper Daily-direct room (with ?t=…) can be created when DAILY_API_KEY is set. The practice and portal call pages also detect missing tokens before join and explain how to get a fresh link.

Starting Record on a provider telehealth call could immediately show Could not start recording. The Daily client now calls startRecording()with 720p size only (room already enables cloud recording). If something still fails, the toast includes Daily's error message. Use the provider join link (owner token), not the patient link, for recording controls.

When building a new intake template in Dashboard → Practice Management → Intake, progress is saved automatically in the browser (about every 700ms while the wizard is open). If the tab is closed or refreshed, use Resume draft on the Intake page to continue, or Discard draft to start fresh. Drafts expire after seven days and are cleared when the template is saved successfully or when opening edit/duplicate on an existing form so drafts do not mix. Data stays on this device only.

In Dashboard → Admin → Users, opening Edit on a user now includes a Couples connections table (read-only): connection id, status, both account emails, pending invite email, whether user2 can see all of user1's connections, and connected date. Matches GET /api/admin/users/{id}/couple-connections. Purging a user still logs removed links at INFO and clears couples dashboard cache for affected accounts; operators can use scripts/one-off/list_couple_connections_for_user.py on the server. See docs/features/COUPLES_CONNECTION_TROUBLESHOOTING.md.

The couples dashboard uses a short server-side cache. It is now cleared when a partner completes join with code and when an admin purge removes couple links, so you should not wait up to a minute to see connected or disconnected update in those cases.

The couples dashboard invite screen now explains that "not connected" usually means there is no active link in the system (not only a manual disconnect), and lists steps: correct subscription tier for both people, re-invite or join with a code, and signing in with the account that accepted the invite. See docs/features/COUPLES_CONNECTION_TROUBLESHOOTING.md for operators.

Encounters, Billing, Claims, and Intake use larger touch targets (about 44px) on key buttons, clearer focus rings for keyboard users, and improved spacing on small screens. Prior authorization lines on encounter and claim views are less likely to truncate awkwardly on narrow widths.

Dashboard → Admin → API Keys now includes Daily.co and Pipecat Cloud variables used for practice telehealth (DAILY_API_KEY, PIPECAT_API_KEY, PIPECAT_AGENT_NAME, and optional API base URLs). Values saved there apply to the running backend without editing server env files.

The practice area was waiting on dashboard stats as well as user and subscription data, so a slow or stuck /api/stats request could keep every practice page (including Provider profile) on "Loading…" forever. Practice routes now gate only on user and subscription. The profile page also loads calendar sync in the background so it cannot block the provider form.

Practice Telehealth and Telehealth call use safer viewport height, horizontal overflow protection, and control bars that wrap with home-indicator padding. In-call Session notes sits below mic/camera/leave so controls stay reachable when notes are open. The signed-in Client portal uses the same overflow and safe-area patterns for small screens.

On /portal, clients can open Your information, edit contact and demographics, and save (same fields as in Practice Management). On Practice → Clients → [client], use Preview client portal to open the portal as that client; a banner links back to the full client record for insurance, intake assignment, billing, and notes. Payments and intake submission stay disabled in preview.

Practice managers can enable Allow cloud recording on Practice → Telehealth or Provider profile. New Daily rooms then allow cloud recording; providers join with a host link so Record / Stop recording appear in the in-app call. Copy join link still copies the patient URL. PostgreSQL: run migration telehealth_allow_cloud_recording on provider_profiles (see deploy migrations). Details: docs/features/TELEHEALTH_RECORDING_PRACTICE_MANAGER.md.

Today's appointment actions (copy patient link, waiting room, start session) stack in a full-width column on small screens so controls are not clipped. When appointments and the waiting room fail with the same server message (for example after a deploy), a single alert with Try againreplaces duplicate toasts and cards. Background waiting-room refresh no longer flashes "Loading…" every few seconds while the list is already visible.

Sign-in now uses the dedicated /token endpoint first (the same path your server sends to the API), with a fallback to /api/token when needed. After a successful login, the app also sets the secure session cookie the dashboard gate expects — not only localStorage — so you are not sent back to the login page when opening the dashboard. A 401 on the initial profile load no longer clears your session by itself.

If you are both an Emily client and an admin (or practice owner), the dashboard now correctly shows the Admin Tools and Practice Management sections. Admin visibility is based on your account's admin flag instead of a hardcoded email list; Practice Management is shown when you have the Coach/Professional plan or admin access.

Practice Site settings and Intake no longer show "temporarily unavailable" when saving; fixes ensure tier checks and database commits behave correctly. Billing now shows a clear banner when payment processing is not enabled, so you know upfront instead of only after clicking Pay.

Refresh buttons were added on Journal (History), Goals, Notifications, Practice Clients, and Reflections. Telehealth hub and call page support copy join link and refresh; waiting-room order and join time are more consistent.

A new Memory & context page (Dashboard → Prepare & store) shows what the app uses to personalize your experience: journal themes and insights, shared partner memories, and relationship profiles (partners, friends, family). The app now uses this context across chat, journal, couples features, and AI suggestions so your history and relationships are reflected in one place.

• View and manage your journal memory (tags and patterns from entries)
• See shared couple memories count and link to add or edit them
• See people you've added (roleplay profiles) and open them to update

From Today's schedule, "Start telehealth" now creates a session and opens the call in one click when Daily.co is configured. Activity shows a recommended next action (e.g. "Add note for [client]" or "Sign treatment plan") to reduce cognitive load.

Final review step now shows a mini form preview: each section with its field count. The Telehealth page highlights in-app video and waiting room with no separate app.

Therapeutic chat and the shared chat input now use tighter spacing on small screens so the composer uses less vertical space while staying within safe areas. Send and action buttons keep at least 44px tap targets, and the input bar no longer overflows on narrow viewports.

Error screens and toasts now use a single, user-friendly message format so you see clear, safe text instead of raw technical details when something goes wrong (e.g. dashboard errors, analysis stream, love language quiz, and bug report flow).

Login, refresh, and logout now use Next API routes to keep the auth cookie aligned with the active session, fixing the broken login flow caused by the missing server-action module.

Denied claims now expose an optional appeal submitted date in the claims workflow, so providers can keep the appeal timeline visible alongside the denial reason and draft.

Practice users can now keep prior-auth context visible as they move from authorization to encounter to claim, see denial reasons directly in the claims workflow, and get clearer setup guidance for Stedi, calendar sync, and profile configuration.

The Practice hub is also reachable from the mobile bottom navigation, so the same workflows are easier to get to on smaller screens.

Telehealth sessions and waiting-room admits now return the correct provider metadata so the in-app call shell stays on the right adapter after a client is admitted.

New telehealth rooms created by the portal now start with the requested Daily Prebuilt room defaults enabled, including prejoin, People, text chat, emoji reactions, background blur controls, PiP, network stats, noise cancellation, live captions, and English as the default language.

Practice intake templates can now be archived and restored, and the wizard includes a custom field builder so providers can add their own sections and questions before saving.

Providers can now reopen an existing intake template, revise its schema-backed settings, and save the changes in place. Templates can also be duplicated to create a new version without starting from scratch.

Practice intake now starts from a guided template library with more than 10 presets, including couples, telehealth, insurance-ready, trauma-informed, and crisis workflows. Providers can customize which sections appear before saving the template.

Practice billing language now makes the provider role explicit. The hub and billing page now describe providers collecting client payments, which matches how the Practice portal actually works.

Telehealth sessions now carry provider metadata from the backend and open through a portal-owned video shell. That keeps the UI and session flow in the portal while the media layer stays swappable underneath.

Practice Management invoice payments can now be enabled independently from the rest of the site's payment features. That keeps subscriptions and other payment flows off while still allowing the practice billing portal to open Stripe Checkout when the new practice payment flag is turned on.

Therapeutic chat input areas now behave much better on phones. The composer stacks cleanly on narrow screens, the send button stays full-width and easy to tap, and the shared chat input spacing has been tightened so the keyboard leaves more room for the conversation.

Shared calendar setup now lives in the Practice portal's Admin section, so providers can find Calendar Sync Setup where they manage practice tools. The Calendar page now points there directly for Google Calendar credentials and two-way sync.

Devotion Connect now opens in NKJV (New King James Version) by default site-wide. You can still choose KJV or any other supported translation from the dropdown. Previously saved translation choices in your browser are unchanged.

The Journal is now 10x better: write entries directly on the page (no dialog), see “This week in your journal” with AI themes, get a quick takeaway after saving, and turn any entry into a goal in one click. Onboarding and dashboard now highlight the Journal as Option 1. Focus mode hides title and mood for distraction-free writing.

• Inline composer: create and edit entries in the main interface
• First-visit: suggested prompt and one-click start
• Weekly summary card with themes from your writing
• Post-save offer: “Get a quick takeaway?”
• “Turn into a goal” on each entry links to Goals with pre-filled title
• Focus mode toggle for a larger writing area

The “Copy draft to clipboard” button on step 5 (Draft an apology) of the repair worksheet now shows correctly for all users.

On the admin bug reports page, when you filter by Resolved or Closed, the list now fetches up to 10,000 closed bugs (fixed, closed, wont_fix) in one request. This supports verifying and fixing closed bugs at scale.

When an e-prescribe vendor is configured, provider enrollment status is now stored in the database. Practices can view and manage ePrescribe/EPCS enrollment per provider; enrollment state persists across sessions and is available via the practice ePrescribe APIs.

The “How to Repair Any Relationship” worksheet now includes the full 7-step tool directly in the app: pause and calm down, acknowledge what happened, take responsibility, consider their experience, draft an apology, one small next step, and reconnect. Progress is saved on your device. Use Print / save as PDF to create a PDF from the tool. The standalone PDF file has been removed.

From the Journal page you can now export your timeline (journal entries and guided reflections) for backup or printing. Use Export (.md) for a Markdown file, or Export (HTML) for an HTML file you can open in a browser and print or save as PDF.

• Export includes up to 200 recent items from your unified journal feed
• Markdown: plain text, easy to move into other apps
• HTML: print-friendly layout; use your browser's Print → Save as PDF

You can now create and manage your own journal templates. From the Journal page, use Manage templates to add templates with a name, slug, and optional prompt. Your custom templates appear as quick-action buttons next to Morning, Evening, and Gratitude so you can start an entry with your own prompts.

• Create, edit, and delete custom templates
• Each template has a name, slug, and optional prompt/placeholder
• Use a custom template from the Journal header to write with that prompt

A dedicated Journal experience is now available at Dashboard → Journal. Write freely with a blank page, or use quick prompts for Morning, Evening, or Gratitude. All entries and your guided reflections appear in one timeline. Add an optional mood to each entry and build a journal streak. Weekly digest emails now include your journal entry count alongside reflections and goals.

• Freeform journal entries with optional title and mood
• Morning, Evening, and Gratitude templates
• Unified feed of journal entries and guided reflections
• Journal streak (writing or reflecting counts)
• Link to Guided reflection for AI insights and Bible linking

Text-to-speech (Read aloud) and notification sounds are more reliable across the app:

• Read aloud— Buttons no longer get stuck in “speaking” when you start reading from another section or cancel elsewhere. First click is more likely to work (Chrome first-load retry). Voice selector shows “Loading voices…” until voices are ready instead of disappearing.
• Voice choice — Default voice prefers local/system voices for better compatibility; you can still pick any voice from the selector.
• Notification chime — Sound now plays correctly when the browser had suspended audio (e.g. before the first user interaction).

Frankenstein Language Analyzer (multi-model consensus) now allows much larger inputs on Premium and Professional plans: up to 2 million characters (2000k) per run instead of 200k. Free and other plans keep the 200k limit. This lets you attach or paste long documents (e.g. large PDFs) for analysis without hitting the limit.

An in-app worksheet is available from the Repair hub: work through all 7 steps (pause and calm down, acknowledge what happened, take responsibility, consider their experience, draft an apology, one small next step, reconnect) with progress saved on your device. Use Print / save as PDF from the tool to create a PDF. Open from Dashboard → Repair → How to Repair Any Relationship.

New AI-assisted features for providers using Practice Management (Coach/Professional plan):

• Intake summary — On a client’s intake assignment (with a submitted response), use “Generate summary” to create a short clinical summary stored on the assignment. Summary is shown on the client detail page.
• Suggest appeal — For denied claims, use “Suggest appeal” to get a draft appeal paragraph (from denial reason and encounter note). Copy and submit to the payer after review.
• Suggest ICD-10 from note — On Encounters, use “Suggest ICD-10” next to an encounter with a note to get suggested diagnosis codes (e.g. F33.1, F41.1) and prefill the encounter.

Everyday AI Assistant now feels closer to ChatGPT with clearer actions and more power:

• Copy / Regenerate / Edit — Copy any assistant reply, regenerate the last response, or edit a user message and send again (hover over a message to see actions).
• Follow-up chips— Short suggestions below the last reply (e.g. “Summarize that in one sentence”, “Give me an example”) that fill the input when clicked.
• Search conversations — Filter the sidebar list by title or date.
• Export — Download the current conversation as .txt or .md from the header when there are messages.
• Streaming — When using OpenAI and MCP tools are off, replies stream in as they’re generated; otherwise the same typing effect as before.

Login is more convenient and still secure:

• Keep me signed in — Check the box on the login page to stay signed in longer on this device (e.g. up to 30 days). We never store your password; only your session is extended. Idle timeout still applies for security.
• Remember my email — Your email is pre-filled next time you visit the login page (stored only on your device). Password is never saved by the site.
• Your browser can offer to save your password if you want; we don't block that.

A new God-centered tool under Present for the Next Generation: Father-Son Repair. It includes a short scripture-anchored assessment, guided reflection via Therapeutic Chat (Scripture, prayer, forgiveness, and concrete next steps), and progress milestones you can log (e.g. had a conversation, prayed together). Grounded in biblical wisdom; not therapy or diagnosis. Open from Dashboard → Present for the Next Generation → Father-Son Repair.

When something goes wrong—login, dashboard pages, Safety Check, Bible research, or notifications—you now see clearer, user-friendly error text instead of raw technical messages. Errors are handled consistently so you get a helpful message and can retry or get support when needed.

A search bar now appears at the very top of the site on every page. Type to find pages and features—dashboard, Safety Check, AI Chat, Goals, Bible, Preferences, and more—then select a result to go there. Works from marketing, login, and dashboard.

In Dashboard → Preferencesyou can now install Emily's Growth Portal on your device. A new “Download and install the app” section shows an Install button when your browser supports it, or a short note when the app is already installed or install isn't available. Installing adds a home screen icon and optional offline support.

Loading the Clients page with a large limit (e.g. 500) no longer triggers a server error. The API now applies sorting before pagination so the request completes successfully.

On the Encounters page you can now delete an encounter from the actions column. Deletion is only allowed when the encounter has no claims; otherwise you’ll see an error and must remove or void claims first. A confirmation prompt appears before the encounter is removed.

The dashboard header is more compact: logo and welcome message sit on one line, padding and spacing are reduced, the “How are you today?” button uses a smaller size, and nav links are slightly tighter. Same links and behavior with less vertical space.

Practice Management (Coach/Professional plan) now includes a live Reports page with total billed/paid, denial rate, collection rate, and unbilled counts; a Prior authorizations page to track auth numbers and session limits; Suggest CPT from session duration on Encounters; Generate note stub and Validate claim before submit; and a Telehealth “Start session” stub. Integration tests added for the new endpoints.

Placeholder pages for Practice Management are in place so Calendar, Clients, Billing, Claims, and Telehealth links no longer 404. Full features will roll out by phase (foundation, Stedi EDI, automations, telehealth and scale).

The dashboard top bar uses the app name with logo icon, one primary button “How are you today?”, and a More dropdown for quick links: Who this is for, Devotion Connect, Michael–Emily Mail (when available), and for admins a grouped Admin section. This keeps the header minimal while preserving access to all links.

The dashboard top bar is restored to the March 3 design: title “🌱 Emily's Growth Portal”, “Welcome back, [name]!”, and compact inline links (Who This Is For, Michael–Emily Mail, Devotion Connect, admin) with emoji icons. No logo image or primary button; simple text links only.

The dashboard top bar is restored to the previous design: app name with logo icon, and all quick links in a single horizontal row (Who This Is For, Michael–Emily Mail, Devotion Connect, and for admins the admin links), each with an icon. The row wraps on smaller screens and uses design tokens for consistent styling.

The dashboard top bar is restored to the prior design: title “Reflect with Emily” (no logo image), prior font and colors, and quick links (Who this is for, Devotion Connect, Michael–Emily Mail, admin) under a single Moredropdown next to the “How are you today?” button.

The Therapeutic Chat page now uses design tokens for backgrounds, borders, typography, and focus rings. Message bubbles, input area, and loading state follow the same semantic colors as the rest of the app for consistent look and better contrast in light and dark mode.

The All tools section on the dashboard now uses design tokens throughout: section bars (Longitudinal Evaluation, Faith & Bible, Stay connected, Wizards, AI & chat, Settings & Tools, Admin Tools) use a card-style bar with semantic typography and focus rings; feature cards (Safety Check, Bible Evaluation, Ask the Word, Everyday AI Assistant, Longitudinal Evaluation) use the same surface and typography tokens with colored accent borders. Full-page category view and nested expandable rows also use tokens for a consistent look and better contrast.

Links or bookmarks using /dashboard/weeklyreview or /dashboard/weekly_review now redirect to the canonical /dashboard/weekly-review. No more broken or inconsistent URLs for the Weekly Review page.

The dashboard home header now uses your app name (e.g. Emily's Growth Portal) with the logo icon, and all quick links appear in a single horizontal row instead of a “More” dropdown. Links include Who This Is For, Michael–Emily Mail (when available), Devotion Connect, and for admins: Admin Dashboard, Admin: API Costs, Admin: Bug Reports, Admin: AI Prompts, and Admin: User Manager, each with a small icon. The row wraps on smaller screens and keeps 44px touch targets for accessibility.

On Admin → Datasets, when the NPD dataset has not been built yet, you can click Run build to start the download script from the server. The build runs in the background; the page continues to auto-refresh every 12 seconds so when the dataset is ready it appears without a manual reload. You can still copy the command and run it in your terminal if you prefer.

A new brand report at /brand/practice-management-landscape covers SimplePractice, 15+ competitors, and the mental health practice management market: pricing, workflows, and where the industry is heading. Linked from the brand landing and from the Platform vision page. Brand and vision content only.

A new public brand area at /brandintroduces Toxic Partner's broader vision alongside the current product. The brand landing explains two arms: Toxic Partner (current product) and our platform vision (next-generation therapy and coaching platform with Stedi EDI). The full strategic report lives at /brand/platform-vision. Both pages are linked from the site footer under Product.

Family Communication (Kids Connect) now has a 2-step wizard: choose family meeting prompts, whole-family conversation prompts, or simple frameworks (including Rose–Thorn–Bud). Pick one or use “Surprise me”; the selected prompt or framework appears in a large, readable card. Copy to clipboard or start an optional 5–10 minute timer for check-ins. Supports connection and healthy communication; not therapy or conflict mediation.

The Partner activity card (when enabled in admin) now lists more of what your partner recently used. In addition to Safety Check and Communications Wizard, it can show: Chat, Goals, Reflections, Check-ins, Gratitude, Everyday AI Assistant, and Bible Study—each only if your partner used that tool in the last 30 days. Items are sorted by most recent use.

Everyday AI Assistant is a new ChatGPT-style chat for everyday questions: how-to, recipes, quick facts, describing photos, and recommendations. One input bar with optional image attach (paste, drag, or button); an expandable Options panel lets you choose model, temperature, max tokens, app context, AI context ecosystem, MCP tools, and an optional custom system prompt. Sessions are separate from Ask the Word and full AI Chat; find it in the dashboard and in Quick Access.

All therapeutic chats (When Triggered, 30-Day Curriculum, Next Generation Upbringing, and others) now use a modern, high-quality chat experience: rich markdown in responses (lists, headers, bold, code), a typing effect when the guide replies, an animated thinking indicator, dark mode support, auto-resizing input, and smooth scrolling. The same experience powers every prompt-slug and reflection-prompt chat.

AI Chat message loading is faster: we now fetch the last 300 messages instead of 2000, so opening a conversation is quicker. The Bible chapter page can take 10–15 seconds on first load (then it’s cached); we’ve adjusted admin performance alerts so these expected long-running requests are no longer flagged as “Needs attention.”

When the NPD training dataset has not been built yet, the Admin → Datasets page now guides you through the full flow: a clear step-by-step (build from repo root, set NPD_DATASET_DIR if needed, then Retry), a Copy command button for the download script, and the training wizard opens automatically so you can follow build → dedupe → train in one place. Retry re-checks immediately; the page also auto-retries every 12 seconds (up to 5 times) so it can update without clicking after you run the script.

The Present for the Next Generation section now has its own help content. Use the (?) button on that page for section-specific FAQs, and find a dedicated section under Help Center → Present for the Next Generation.

Couples can set a daily time (e.g. 9pm) and receive a tailored Bible chapter suggestion based on your relationship context—growth goals, recent conflicts, gratitude—so you can read and pray together and let the Holy Spirit guide you.

• Settings — Enable the feature, choose hour, minute, timezone, and Bible translation.
• Today's focus— One suggested chapter per day with a short reason; read in-app or open in the full Bible; mark when you've read and prayed together.
• Reminders — Optional in-app and email reminders at your scheduled time (requires Couples Connect).

Kids Connect is a new dashboard section for parent–child connection, co-parenting tools, biblical parenting resources, and family communication. Find it on the dashboard and open the Kids Connect card to see all areas.

• Parent–Child Activities — Age-appropriate activity ideas and conversation starters for quality time with your kids.
• Co-Parenting — Communication guidelines, schedule tips, and shared parenting principles when households are separate.
• Biblical Parenting — Scripture-grounded passages and practical guidance for discipline, instruction, and love.
• Family Communication — Prompts and simple frameworks for family talks and whole-family connection.

A new Personal Growth section helps you process your upbringing and strengthen presence with your kids. Find it on the dashboard under Personal Growth.

• Processing your upbringing — Guided reflection on the impact of your father or mother not being present; validate the experience, name what was missing, and connect to what you choose to do now, including playfulness and presence.
• Being present for your kids — Reflect on small, concrete ways to show up: 10–15 minutes of purposeful or playful time, rituals, and repair when you miss the mark. Stepfamily-aware for your kids and your partner’s kids.

Cycle Breaker helps you interrupt the anxious–avoidant cycle with a regulated 5-minute pause, message translation, and structured repair steps. Find it under Relationship Tools on the dashboard.

• 5-minute pause — Choose a trigger, start the timer, and follow the breathing guide (In 4s, Hold 4s, Out 4s, Hold 4s) while the timer counts down. Timer shifts from blue to amber in the last minute, then green when done.
• Translate a message — Optionally rephrase what you were about to send into safer, clearer wording before you continue to repair.
• Repair steps — Walk through five steps: acknowledge your part, validate your partner, state your need, invite them to share, and make a small commitment. Mark each done as you go.
• Proactive nudge — When recent partner conversations (Messages or Communications Wizard) show signs of the cycle, a supportive nudge can suggest taking a Cycle Breaker pause. Nudges appear on the Messages page, Communications Wizard, dashboard Relationship Tools, Couples hub, and Fight Room.

The Admin Datasets page (Admin → Datasets) now includes a step-by-step training wizard and clearer guidance for building, deduping, training, and verifying the NPD/abuse detection model.

• Open training wizard— Four steps: Build dataset (quick or custom options), Dedupe (recommended), Train model (DeBERTa/BERT, batch size, learning rate, epochs), then Run & verify with a single copyable or downloadable script.
• Training readiness — Page shows whether train/val/test splits exist and displays last training run metrics (eval loss, accuracy, F1) when available.
• Verify — Instructions to check test_results.json and run the interactive test command; app integration via NPD_HF_MODEL_PATH and ENABLE_NPD_LOCAL_MODEL.

A new guided wizard walks you through training the NPD/harmful-language classifier and other local AI models from the dashboard.

• Dashboard → Wizards → NPD Training Wizard — Step-by-step: build dataset, dedupe, train (DeBERTa or BERT), verify, and enable in app with copyable commands.
• Refresh status — Check whether your dataset and trained model exist (e.g. after running commands on the server).
• Phase 2 (NPI) — Copyable command and notes for training the NPI-40 model used by NPD Watch.

The GPS tracking option and admin GPS export have been removed from the product. Admin user management no longer includes GPS-related controls or location data.

The Communications Wizard now uses a familiar email-client layout for easier scanning and reading.

• Inbox, Sent, All tabs— Switch between messages from your partner (Inbox), messages you sent (Sent), or the full thread (All). Inbox shows an unread count when you have messages you haven't read yet.
• Message list and reading pane — A list of messages with sender, preview, and date; click one to read the full message in the reading pane. Unread messages are marked with a dot and bold sender.
• New message— A clear "New message" button opens the compose section to write and send.

Expiration options (Valid for, Max opens, Expires on) are now in a single compact row. After sending, the share-link card is clearer with a short description. Writing tools (AI suggestions, boundaries, Scriptures, structure tips, suggested for you) are grouped under one "Writing tools" section to reduce clutter.

On large screens, the conversation thread and the compose (or preview) area appear side by side for an email-style read-and-write layout.

The Communications Wizard page and partner notification emails are improved for usability and privacy.

• Layout— Conversation and Compose are now clearly separated with section headings; your messages align right and your partner's left; expiration options are grouped; and the share-link options after sending appear in a clear card.
• Emails — Notification emails to your partner (new message or link shared) now contain only a short line and the link to open the thread in the app; no message content or preview is ever included in the email for security.