You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Choosing therapies to compare on a phone was the wrong shape twice over. /therapy-compass/compare drew four empty slots before anything had been chosen, and once two were chosen the comparison was a min-w-[720px] table inside a horizontal scroller — on a 390px phone that is about two thirds of a column at a time, with the field labels scrolling away from the values they label.
Four directions were prototyped as design scratch in PR #2339; this builds the chosen one for real, on Therapy Compass only. The model is that you fill the compare set where the therapies are — from a list, a search result, or while reading a record — and the comparison screen is only ever reached with something already in it. Empty slots are never drawn.
Adding no longer navigates.ResultCard (src/components/therapy-compass/therapy-card.tsx) moves from toggleCompare to addCompare/removeCompare, and the label moves with the behaviour: "Compare" promised a destination it no longer goes to, so it now reads "Add to compare" / "In compare tray". toggleCompare is removed from the bindings API rather than left behind as a dead navigate-on-add path. A full tray takes aria-disabled plus a stated reason rather than silently ignoring the tap, per the button-wiring convention.
The record nav rail's Compare slot becomes pure navigation in both branches (therapy-record-nav-header.tsx). It previously added the therapy when it was absent from the set and navigated when it was present — one control with two different meanings depending on state the reader could not see.
A compare tray docks above the phone search pill. New src/components/therapy-compass/therapy-compare-tray.tsx portals into the documented phone dock addon slot, so it inherits the dock's position: fixed, z-index, safe-area padding and scroll-hide transform; there is no second scroll listener and no bottom-offset arithmetic. It registers a third PhoneDockAddonKind (therapy-compare) across all four required sites: the slot id in src/lib/mode-home-composer.ts, the reserve constant and both resolver branches in mobile-composer-reserve.ts, the clearance tokens / backdrop scrim heights / hide-transform overshoot in globals.css, and the claim in global-search-shell.tsx.
Record routes get an add button in the page body instead of tray chrome. Those routes have no search pill, and the dock reserve inflates on claim rather than on render, so claiming a slot there would open a blank band at the bottom of the page. New record/compare-action.tsx is therefore ordinary page content, introducing no second bottom owner, and announces through role="status" — with no tray on screen an add there would otherwise be completely silent.
The comparison stacks on a phone.compare-screen.tsx keeps today's table from md up, unchanged, and below it renders the same rows memo turned inside out: one card per field, every selected therapy listed against it, so the label never leaves the value. The fork is md (768px) rather than sm, because the table is 720px wide and at 640–767px it would still scroll sideways — the exact defect being fixed.
The compare set is remembered on this device. New src/lib/therapy-compare-memory.ts, wired into TcProvider. This is strictly additive: the URL stays the source of truth, a shared ?ids= link always wins, and memory only fills the gap when you arrive with no ids at all. Remembered slugs are validated against the live catalogue at restore time, so a slug from an older data generation is dropped rather than resurrected as a therapy that no longer exists. Cross-tab sync is a deliberate non-goal — a remote write would rewrite this tab's address bar mid-comparison.
data-therapy-scroll-sm and its globals.css rule are removed together. Phones no longer render that table, so the rule could not match anything, and the existing contract assertion on the attribute would have kept passing for the wrong reason. The assertion now pins the new phone behaviour instead.
Desktop is deliberately unchanged, and no other mode is touched — Dictionary, DSM, formulation and specifiers keep today's behaviour, even though src/components/compare/ is shared with them.
Verification
npm run verify:pr-local — Verification not run: superseded by the gates below.npm run verify:cheap ran on this exact tree and covers the failure classes this diff can plausibly break (lint, typecheck, the full offline unit suite, and the 34 static/consistency gates), and the browser evidence below is stronger than the conditional build step PR-local would have added. CI's own routing remains the authority.
npm run verify:ui — pass, via npm run verify:phone-chrome on this exact commit. That wrapper classified the change as touching shared phone-chrome foundations (src/app/globals.css + mobile-composer-reserve.ts) and escalated to the complete Chromium gate rather than stopping at focused owners: 521 passed (17.3m), exit code 0.
npm run verify:cheap — pass, exit code 0, on this commit and this base: Test Files 876 passed (876) · Tests 10547 passed | 1 skipped (10548), plus lint, typecheck and the 34 static/consistency gates. npm run lint separately: [gate-receipts] recorded a pass for "lint:internal" (5450 input files).
The two journeys that carry this change's own contract are inside that 521. They were also run in isolation during development, against a clean production build: tests/ui-therapy-nav-scroll.spec.ts → 2 passed (18.9s), including the new phone compare tray hides with the composer and releases its reserve (drives /therapy-compass/search at 390×844 with a simulated 34px bottom inset, asserts the tray is a descendant of .answer-footer-search-dock, that data-footer-addon="therapy-compare" is set, that after scrolling the tray's top clears the viewport and --mobile-composer-reserve reads exactly 0rem, and that both return on scroll-up); and tests/ui-route-coverage.spec.ts → 13 passed (33.7s), including the rewritten card journey which now asserts the opposite of what it used to — that activating the compare control does not navigate, leaves you on /therapy-compass/search with ids written to the URL, and makes the tray appear.
Two design-system gates went red during development and were repaired properly rather than suppressed: check:design-system-contract flagged the tray's raw <button> elements and a raw gap literal, so the controls moved onto the shared Button / interactiveRowBase and the spacing onto tokens; style-contract-registry required every new unlayered visual class to be either contracted or explicitly exempted, so the five that remain carry a reasoned exemption naming the coverage that does exist (the tray is phone-only and therefore unreachable at the desktop viewport ui-style-contract.spec.ts drives — the same reason the Patient details pill is exempt).
npm run verify:release — not a release or handoff-confidence claim.
npm run eval:retrieval:quality — no retrieval, ranking, selection, chunking or scoring behaviour changed.
npm run eval:rag -- --limit 15 + npm run eval:quality -- --rag-only — no answer generation, synthesis prompt or answer post-processing changed.
npm run check:production-readiness — no clinical workflow, privacy, environment, Supabase, source governance or deployment behaviour changed. The diff adds a browser-local convenience and re-lays out an existing comparison; it touches no server route, no data access path and no environment.
npm run check:deployment-readiness — no deployment startup, hosting or rollout behaviour changed.
Risk and rollout
Risk: Low, and concentrated in phone chrome rather than in clinical content. The riskiest clause is that the tray must hide with the composer and release its content reserve to zero; that is exactly what the phone dock addon mechanism exists for, two components already prove it, and tests/ui-therapy-nav-scroll.spec.ts now asserts it in Chromium at 390×844. The second risk is that adding a therapy no longer navigates — a deliberate behaviour change, confirmed with the repository owner, whose two existing assertions were rewritten knowingly rather than left to pass silently.
Rollback: git revert of this commit. The memory layer is purely additive and the ?ids= URL contract is untouched, so a revert cannot strand or invalidate an already-shared comparison link.
Provider or production effects: None.
RAG impact: no retrieval behaviour change — nothing under src/lib/rag/, the retrieval RPCs, the ranking/selection/answer-ranking surfaces, the eval harness or the golden fixtures is touched. The changed library files are therapy-compass-navigation.ts (URL parsing for the compare set), the new therapy-compare-memory.ts (browser storage), and mode-home-composer.ts (a phone dock slot id); none of them participates in retrieval, ranking or answer generation.
Clinical Governance Preflight
This section is required because src/lib/therapy-compare-memory.ts and src/lib/therapy-compass-navigation.ts match the clinical-risk path pattern ^src/lib/.*therap in scripts/pr-policy.mjs. Each item below is a statement of fact about this diff, checked against the changed-file list.
Source-backed claims still require linked source verification before clinical use
No patient-identifiable document workflow was introduced or expanded without explicit governance approval
Service-role keys and private document access remain server-only
Demo/synthetic content remains clearly separated from real clinical sources
Source metadata, review status, and outdated/unknown-source behavior remain conservative
Deployment classification/TGA SaMD impact was checked when clinical decision-support behavior changed
Evidence for each. No clinical claim, source field, review status or citation is created, edited or reworded anywhere in this diff; the comparison renders the same ROWS getters over the same therapy records it did before, only in a second layout below md. No document upload, download, storage or access path is touched, and no patient-identifiable data is read, written or persisted — the one thing stored on the device is a list of therapy slugs (["cognitive-behavioural-therapy-cbt", …]), which is the same public catalogue identifier that already travels in the shareable URL. No Supabase client, environment value or credential appears in the diff, so the project target and the server-only confinement of the service-role key are unchanged by construction. Demo and synthetic content are untouched: no fixture or demo corpus file is modified, and the design-scratch mockups that seeded this work stayed in PR #2339 rather than here. Review status stays conservative and stays visible — the comparison's Reviewed / Needs review marker is preserved in the desktop table, the new phone layout keeps the Evidence level row (which falls back to "Source review required" for an unreviewed record), and the compare tray's expanded sheet lists each therapy by full name. The remembered set fails conservative in three separate ways: a slug absent from the live catalogue is dropped at restore, a malformed or unreadable store returns an empty set rather than throwing, and a shared link always overrides device memory so a colleague can never be shown a different comparison from the one they were sent. Because no clinical decision-support behaviour changed — no ranking, no recommendation, no answer, no clinical text — the SaMD/TGA classification question was checked and returns no impact.
Notes
Chromium at 390×844 is the only device evidence available in a container. It does not close physical iPhone Safari or installed-PWA acceptance; the repository owner should tap through the flow on his own phone before this is considered done.
The four prototype directions were deliberately not bundled into this PR: this one needs a Clinical Governance Preflight and that one did not, and AGENTS.md forbids combining the two. PR Add four phone mockups for the therapy comparison picker #2339 has since been closed unmerged — it existed to let a direction be chosen, that happened, and its branch kept conflicting with main on a generated snapshot file. The prototypes remain readable from that closed PR and its branch claude/therapy-comparison-mobile-design-z0dagr.
src/components/compare/ is shared by therapy, dictionary, DSM, formulation and specifiers, so the empty-slot problem this addresses is present on all five. This change is scoped to Therapy Compass only, by explicit decision; the other four are untouched and unaffected.
Low Risk
Risk is mostly phone layout chrome (dock reserve, scroll-hide, addon exclusivity) and a deliberate UX change that add no longer navigates; clinical content and retrieval paths are untouched, with broad test coverage on the new contracts.
Overview Therapy Compass on phones now builds the compare set where users find therapies instead of sending them to an empty comparison page or a wide horizontal table.
Compare behaviour: Adding a therapy updates the URL ids and stays on the current page (addCompare / removeCompare replace toggleCompare). Result cards and record pages use Add to compare with full-tray handling; the record nav Compare control only navigates to the comparison screen. Device memory (therapy-compare-memory.ts) restores the last set when the URL has no ids, with shared links and catalogue validation winning over stale slugs.
Phone chrome: A one-row compare tray (therapy-compare-tray.tsx) portals into the search dock addon slot (therapy-compare), with matching reserve tokens, backdrop height, and scroll-hide overshoot in globals.css and mobile-composer-reserve.ts. The shell claims the addon only on therapy dock routes and when readTherapyCompareSlugCount is non-zero. Expanded tray management uses a bottom Sheet; record routes get in-page TherapyCompareAction with live status announcements because there is no dock there.
Comparison UI: From md up the table is unchanged; below md the same rows render as a per-field stacked layout (TherapyCompareStack). The old [data-therapy-scroll-sm] horizontal-scroll path is removed.
Docs, adoption manifest, and contract/browser tests are updated for the new addon kind and behaviour.
Choosing therapies to compare on a phone was the wrong shape twice over.
`/therapy-compass/compare` drew four empty slots before anything was chosen,
and once two were chosen the comparison was a `min-w-[720px]` table inside a
horizontal scroller: on a 390px phone that is two thirds of a column at a
time, with the field labels scrolling away from the values they label.
Four directions were prototyped as design scratch (PR #2339); this builds the
chosen one for real. The model is that you fill the compare set *where the
therapies are* — from a list, a search result, or while reading a record — and
the comparison screen is only ever reached with something already in it.
Adding no longer navigates. `ResultCard` moves from `toggleCompare` to
`addCompare`/`removeCompare`, and the label moves with the behaviour:
"Compare" promised a destination it no longer goes to, so it reads "Add to
compare" / "In compare tray". `toggleCompare` is removed from the bindings API
rather than left as a dead navigate-on-add path, and the record nav rail's
Compare slot becomes pure navigation in both branches — one control with two
different meanings, depending on state the reader could not see, was the
problem it existed to demonstrate.
The tray docks above the phone search pill through the documented addon slot,
so it inherits the dock's fixed position, z-index, safe-area padding and
scroll-hide transform: no second scroll listener and no bottom-offset
arithmetic. Two invariants hold it there. It stays exactly one row tall — the
dock's content clearance is a static token, so a dock that grows covers page
content by exactly its own growth, which is why the expanded state is a bottom
Sheet rather than the taller bar the prototype drew. And the shell claims the
slot only while the URL carries a compare set, because the reserve inflates on
claim rather than on render, so a claim with an empty tray would open a blank
band under a row that is not there.
Record routes have no search pill, so they get an "Add to compare" button in
the page body instead — page content, not chrome, introducing no second bottom
owner. It announces through `role="status"`: with no tray on screen, an add
there would otherwise be completely silent.
The comparison forks at `md`, not `sm`. The table is 720px wide, so at
640–767px it would still scroll sideways — the exact defect being fixed. Below
that width the same `rows` memo is turned inside out: one card per field, every
therapy listed against it, so the label never leaves the value.
Device memory is additive and the URL stays the source of truth. A shared
`?ids=` link always wins; memory only fills the gap when you arrive with no
`ids` at all. Remembered slugs are validated against the live catalogue at
restore time, so one from an older data generation is dropped rather than
resurrected as a therapy that no longer exists.
`data-therapy-scroll-sm` and its globals.css rule are removed together: phones
no longer render that table, so the rule could not match anything and the
contract assertion on it would have passed for the wrong reason.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01WKxKU17WQqryM8iNhpyJk7
This pull request has been ignored for the connected project sjrfecxgysukkwxsowpy because there are no changes detected in supabase directory. You can change this behaviour in Project Integrations Settings ↗︎.
Limit details: You’ve used the included review currently available. Your 95 included PR review attempts over the past 7 days set your current allowance at 1 review per hour.
Your organization has reached its usage spending cap. Adjust your spending cap in the billing tab.
Travels with its owning product PR rather than on a ledger-only tip, per
AGENTS.md PR bundling. Pushed seconds after the PR opened so the restarted
CI run loses no meaningful work.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01WKxKU17WQqryM8iNhpyJk7
The reason will be displayed to describe this comment to others. Learn more.
Stale comment
Approved. Cursor Bugbot was not present after the first poll, so that signal was skipped; no applicable approval policy requires human review, and no reviewers were assigned.
Sent by Cursor Approval Agent: Pull Request Router and Approver
Gitleaks' generic-api-key rule flagged the exported browser-storage key
name for the compare set. It is a localStorage key — visible in any
browser's devtools, never a credential. The rule matched only because the
identifier ends in "Key" and the hyphenated slug measures 3.99 Shannon
entropy; its five siblings (answer-thread, recent-queries, saved-*,
search-pins) are the same shape at 3.72-3.78, just under the threshold.
Pinned by fingerprint rather than reworded. Rewording would be tuning an
unverifiable number — gitleaks is not installed in this container, so the
threshold cannot be checked locally — and every candidate slug still
measured above the passing siblings, so it would trade a good name for a
worse one and no certainty. The .gitleaksignore file exists for exactly
this, with ten prior entries against the same rule.
The scanner is not weakened: this pins one commit's one line, not the rule
or the path.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01WKxKU17WQqryM8iNhpyJk7
Static PR checks went red on `check:repo-awareness-snapshot`: the
immutable review record committed alongside this change moves the review
count, and the generated inventory is derived from it, so the two must be
committed together.
Regenerated with the repo's own tooling, never by hand. Verified in step
(192 pages, 436 documents, 2617 reviews), with sitemap:check,
docs:check-index, check:ledger-write-discipline, check:branch-review-ledger
and prettier all clean before pushing.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01WKxKU17WQqryM8iNhpyJk7
The reason will be displayed to describe this comment to others. Learn more.
Stale comment
Approved. Cursor Bugbot was not present after the first poll, so that signal was skipped; no applicable approval policy requires human review, and no reviewers were assigned.
Sent by Cursor Approval Agent: Pull Request Router and Approver
The reason will be displayed to describe this comment to others. Learn more.
Approved. Cursor Bugbot was not present after the first poll, so that signal was skipped; no applicable approval policy requires human review, and no reviewers were assigned.
Sent by Cursor Approval Agent: Pull Request Router and Approver
The reason will be displayed to describe this comment to others. Learn more.
Claim the dock only when the tray can render
On a phone route with ids while the catalogue is pending or failed, or when a shared link contains only retired slugs, this condition claims the addon and inflates the 9rem reserve, but TherapyCompareTray derives its count from resolved compareTherapies and returns null at zero. This leaves a tray-sized blank band, permanently for failed loads or retired links; base the claim on renderable selections or render a placeholder, and cover the pending/error/unknown-ID paths.
The reason will be displayed to describe this comment to others. Learn more.
Prefer the session fallback after failed writes
When localStorage.setItem throws because storage is full or write-blocked while getItem remains readable, the writer saves the current set in inMemorySlugs, but the next read returns an empty or stale persisted value instead of that fallback. Leaving and re-entering Therapy in the same tab therefore loses the latest comparison; a storage stub whose setItem throws and getItem returns null reproduces this, and reads should prefer the pending in-memory value until a successful write clears it.
Useful? React with 👍 / 👎.
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
Choosing therapies to compare on a phone was the wrong shape twice over.
/therapy-compass/comparedrew four empty slots before anything had been chosen, and once two were chosen the comparison was amin-w-[720px]table inside a horizontal scroller — on a 390px phone that is about two thirds of a column at a time, with the field labels scrolling away from the values they label.Four directions were prototyped as design scratch in PR #2339; this builds the chosen one for real, on Therapy Compass only. The model is that you fill the compare set where the therapies are — from a list, a search result, or while reading a record — and the comparison screen is only ever reached with something already in it. Empty slots are never drawn.
ResultCard(src/components/therapy-compass/therapy-card.tsx) moves fromtoggleComparetoaddCompare/removeCompare, and the label moves with the behaviour: "Compare" promised a destination it no longer goes to, so it now reads "Add to compare" / "In compare tray".toggleCompareis removed from the bindings API rather than left behind as a dead navigate-on-add path. A full tray takesaria-disabledplus a stated reason rather than silently ignoring the tap, per the button-wiring convention.therapy-record-nav-header.tsx). It previously added the therapy when it was absent from the set and navigated when it was present — one control with two different meanings depending on state the reader could not see.src/components/therapy-compass/therapy-compare-tray.tsxportals into the documented phone dock addon slot, so it inherits the dock'sposition: fixed, z-index, safe-area padding and scroll-hide transform; there is no second scroll listener and no bottom-offset arithmetic. It registers a thirdPhoneDockAddonKind(therapy-compare) across all four required sites: the slot id insrc/lib/mode-home-composer.ts, the reserve constant and both resolver branches inmobile-composer-reserve.ts, the clearance tokens / backdrop scrim heights / hide-transform overshoot inglobals.css, and the claim inglobal-search-shell.tsx.record/compare-action.tsxis therefore ordinary page content, introducing no second bottom owner, and announces throughrole="status"— with no tray on screen an add there would otherwise be completely silent.compare-screen.tsxkeeps today's table frommdup, unchanged, and below it renders the samerowsmemo turned inside out: one card per field, every selected therapy listed against it, so the label never leaves the value. The fork ismd(768px) rather thansm, because the table is 720px wide and at 640–767px it would still scroll sideways — the exact defect being fixed.src/lib/therapy-compare-memory.ts, wired intoTcProvider. This is strictly additive: the URL stays the source of truth, a shared?ids=link always wins, and memory only fills the gap when you arrive with noidsat all. Remembered slugs are validated against the live catalogue at restore time, so a slug from an older data generation is dropped rather than resurrected as a therapy that no longer exists. Cross-tab sync is a deliberate non-goal — a remote write would rewrite this tab's address bar mid-comparison.data-therapy-scroll-smand itsglobals.cssrule are removed together. Phones no longer render that table, so the rule could not match anything, and the existing contract assertion on the attribute would have kept passing for the wrong reason. The assertion now pins the new phone behaviour instead.Desktop is deliberately unchanged, and no other mode is touched — Dictionary, DSM, formulation and specifiers keep today's behaviour, even though
src/components/compare/is shared with them.Verification
npm run verify:pr-local— Verification not run: superseded by the gates below.npm run verify:cheapran on this exact tree and covers the failure classes this diff can plausibly break (lint, typecheck, the full offline unit suite, and the 34 static/consistency gates), and the browser evidence below is stronger than the conditional build step PR-local would have added. CI's own routing remains the authority.npm run verify:ui— pass, vianpm run verify:phone-chromeon this exact commit. That wrapper classified the change as touching shared phone-chrome foundations (src/app/globals.css+mobile-composer-reserve.ts) and escalated to the complete Chromium gate rather than stopping at focused owners:521 passed (17.3m), exit code 0.npm run verify:cheap— pass, exit code 0, on this commit and this base:Test Files 876 passed (876)·Tests 10547 passed | 1 skipped (10548), plus lint, typecheck and the 34 static/consistency gates.npm run lintseparately:[gate-receipts] recorded a pass for "lint:internal" (5450 input files).The two journeys that carry this change's own contract are inside that 521. They were also run in isolation during development, against a clean production build:
tests/ui-therapy-nav-scroll.spec.ts→2 passed (18.9s), including the newphone compare tray hides with the composer and releases its reserve(drives/therapy-compass/searchat 390×844 with a simulated 34px bottom inset, asserts the tray is a descendant of.answer-footer-search-dock, thatdata-footer-addon="therapy-compare"is set, that after scrolling the tray's top clears the viewport and--mobile-composer-reservereads exactly0rem, and that both return on scroll-up); andtests/ui-route-coverage.spec.ts→13 passed (33.7s), including the rewritten card journey which now asserts the opposite of what it used to — that activating the compare control does not navigate, leaves you on/therapy-compass/searchwithidswritten to the URL, and makes the tray appear.Two design-system gates went red during development and were repaired properly rather than suppressed:
check:design-system-contractflagged the tray's raw<button>elements and a rawgapliteral, so the controls moved onto the sharedButton/interactiveRowBaseand the spacing onto tokens;style-contract-registryrequired every new unlayered visual class to be either contracted or explicitly exempted, so the five that remain carry a reasoned exemption naming the coverage that does exist (the tray is phone-only and therefore unreachable at the desktop viewportui-style-contract.spec.tsdrives — the same reason the Patient details pill is exempt).npm run verify:release— not a release or handoff-confidence claim.npm run eval:retrieval:quality— no retrieval, ranking, selection, chunking or scoring behaviour changed.npm run eval:rag -- --limit 15+npm run eval:quality -- --rag-only— no answer generation, synthesis prompt or answer post-processing changed.npm run check:production-readiness— no clinical workflow, privacy, environment, Supabase, source governance or deployment behaviour changed. The diff adds a browser-local convenience and re-lays out an existing comparison; it touches no server route, no data access path and no environment.npm run check:deployment-readiness— no deployment startup, hosting or rollout behaviour changed.Risk and rollout
tests/ui-therapy-nav-scroll.spec.tsnow asserts it in Chromium at 390×844. The second risk is that adding a therapy no longer navigates — a deliberate behaviour change, confirmed with the repository owner, whose two existing assertions were rewritten knowingly rather than left to pass silently.git revertof this commit. The memory layer is purely additive and the?ids=URL contract is untouched, so a revert cannot strand or invalidate an already-shared comparison link.src/lib/rag/, the retrieval RPCs, the ranking/selection/answer-ranking surfaces, the eval harness or the golden fixtures is touched. The changed library files aretherapy-compass-navigation.ts(URL parsing for the compare set), the newtherapy-compare-memory.ts(browser storage), andmode-home-composer.ts(a phone dock slot id); none of them participates in retrieval, ranking or answer generation.Clinical Governance Preflight
This section is required because
src/lib/therapy-compare-memory.tsandsrc/lib/therapy-compass-navigation.tsmatch the clinical-risk path pattern^src/lib/.*therapinscripts/pr-policy.mjs. Each item below is a statement of fact about this diff, checked against the changed-file list.Clinical KB Database(sjrfecxgysukkwxsowpy)Evidence for each. No clinical claim, source field, review status or citation is created, edited or reworded anywhere in this diff; the comparison renders the same
ROWSgetters over the same therapy records it did before, only in a second layout belowmd. No document upload, download, storage or access path is touched, and no patient-identifiable data is read, written or persisted — the one thing stored on the device is a list of therapy slugs (["cognitive-behavioural-therapy-cbt", …]), which is the same public catalogue identifier that already travels in the shareable URL. No Supabase client, environment value or credential appears in the diff, so the project target and the server-only confinement of the service-role key are unchanged by construction. Demo and synthetic content are untouched: no fixture or demo corpus file is modified, and the design-scratch mockups that seeded this work stayed in PR #2339 rather than here. Review status stays conservative and stays visible — the comparison'sReviewed/Needs reviewmarker is preserved in the desktop table, the new phone layout keeps theEvidence levelrow (which falls back to "Source review required" for an unreviewed record), and the compare tray's expanded sheet lists each therapy by full name. The remembered set fails conservative in three separate ways: a slug absent from the live catalogue is dropped at restore, a malformed or unreadable store returns an empty set rather than throwing, and a shared link always overrides device memory so a colleague can never be shown a different comparison from the one they were sent. Because no clinical decision-support behaviour changed — no ranking, no recommendation, no answer, no clinical text — the SaMD/TGA classification question was checked and returns no impact.Notes
AGENTS.mdforbids combining the two. PR Add four phone mockups for the therapy comparison picker #2339 has since been closed unmerged — it existed to let a direction be chosen, that happened, and its branch kept conflicting withmainon a generated snapshot file. The prototypes remain readable from that closed PR and its branchclaude/therapy-comparison-mobile-design-z0dagr.src/components/compare/is shared by therapy, dictionary, DSM, formulation and specifiers, so the empty-slot problem this addresses is present on all five. This change is scoped to Therapy Compass only, by explicit decision; the other four are untouched and unaffected.Generated by Claude Code
Note
Low Risk
Risk is mostly phone layout chrome (dock reserve, scroll-hide, addon exclusivity) and a deliberate UX change that add no longer navigates; clinical content and retrieval paths are untouched, with broad test coverage on the new contracts.
Overview
Therapy Compass on phones now builds the compare set where users find therapies instead of sending them to an empty comparison page or a wide horizontal table.
Compare behaviour: Adding a therapy updates the URL
idsand stays on the current page (addCompare/removeComparereplacetoggleCompare). Result cards and record pages use Add to compare with full-tray handling; the record nav Compare control only navigates to the comparison screen. Device memory (therapy-compare-memory.ts) restores the last set when the URL has noids, with shared links and catalogue validation winning over stale slugs.Phone chrome: A one-row compare tray (
therapy-compare-tray.tsx) portals into the search dock addon slot (therapy-compare), with matching reserve tokens, backdrop height, and scroll-hide overshoot inglobals.cssandmobile-composer-reserve.ts. The shell claims the addon only on therapy dock routes and whenreadTherapyCompareSlugCountis non-zero. Expanded tray management uses a bottom Sheet; record routes get in-pageTherapyCompareActionwith live status announcements because there is no dock there.Comparison UI: From
mdup the table is unchanged; belowmdthe same rows render as a per-field stacked layout (TherapyCompareStack). The old[data-therapy-scroll-sm]horizontal-scroll path is removed.Docs, adoption manifest, and contract/browser tests are updated for the new addon kind and behaviour.
Reviewed by Cursor Bugbot for commit 9b34a01. Configure here.