776 commits

This branch

This branch

All branches

Author	SHA1	Message	Date
Daniel	ff2b2bc9d3	chore: Kitten and Supertonic retired; Kokoro is the local voice ... Both were removed from the gateway in favour of Kokoro, so the app's curated voice lists for them went too. Kokoro's own check no longer has to exclude them — it names its own voices and the list is open, so it accepts anything that is not another family's voice. Voices stay curated per family rather than discovered. Models are discovered from /model/info; voices are not, because no TTS provider exposes its voice list consistently, and a list guessed from a model id is how a picker ends up offering a voice the provider rejects. Tests that pinned the retired voices now pin the families that remain. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01Dv6sqaY6Vq3ChZHMem3cnU	2026-09-12 16:45:05 +02:00
Daniel	fce05a2749	fix: the active library view is unmistakable, and a leading question cannot mislead the vision model ... Documents and Images were two buttons of the same weight, which reads as two buttons rather than as a current view and an alternative. The active one now carries the surface, the weight and a hairline — keyed off aria-selected, so the visible highlight and what a screen reader announces cannot disagree. The look_at_image question is written by a model that has not seen the image, from what the user said, so it can presume something that is not there — "describe this rash" about a photograph of a drug chart. Answering the presumption would send it back as fact. The seeing model is now told the question may assume something absent, and to say so first and describe what is actually there. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01Dv6sqaY6Vq3ChZHMem3cnU	2026-09-12 16:27:21 +02:00
Daniel	1f06a19007	feat: a text-only model can ask a model that can see; and the image regex is gone ... **The regex is gone.** The route ran a pattern over the user's message and enqueued an image from the answer text when the model had not called the tool. It was a compatibility path for models without tool calling and it did more harm than good: it decided in English only, it could not see the conversation, and "image summary" fell through it while reading as an obvious image request to the model itself — which was measured, not assumed. A second and worse decision-maker sitting behind the first. Whether a message deserves a picture is now the model's call, made from the tool description, which is the only place it ever belonged. **Lending eyes.** The same shape, for a different capability. When someone attaches a photograph and the chat model cannot accept image input, the attachment was either refused by the provider or silently dropped — an answer about a picture nobody had looked at, which is worse than a refusal. The chat model is now offered look_at_image beside the image tool and decides when to use it. The attachment goes to clinical_assistant.vision_model, whose description comes back as a tool result, and the chat model answers in its own voice with its own sources. Only the seeing is delegated; the clinical reasoning stays with the model an administrator chose. The seeing model is told to report and not to diagnose, because it has a picture and no context and an opinion from it would carry weight it has not earned. Delegation triggers only on an explicit supports_vision: false from the gateway. An unknown is left alone — most of a roster reports nothing, and treating silence as blindness would route good models through a detour. The capability lookup moved to its own module, is cached for five minutes because it runs on exactly the requests that are already slowest, and is never inferred from the model id. liteLLMBaseUrl moved from the admin route to litellm.js, where the other gateway helpers live. The new setting is guarded like the slide reviewer: a model the gateway calls text-only cannot be saved as the one that looks at images. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01Dv6sqaY6Vq3ChZHMem3cnU	2026-09-12 15:43:36 +02:00
Daniel	03621752e8	feat: image fallback chains for every workflow, and a library worth looking at ... **Fallbacks.** One image model meant a refusal, a rate limit or a model the gateway had since dropped ended as a missing picture. Every workflow now tries its model, then each fallback in order, stopping at the first that produces an image. Primary plus two, capped: each hop is a paid request, and a chain long enough to need a cap is long enough to surprise someone. My Resources previously had no fallback at all — only the Clinical Assistant did, and only one. That is backwards: a missing figure is most visible in a deck, where it leaves a hole in a slide. The retry rule is now a classifier that says *why*, rather than a boolean. Transient faults, a 404 for a model the gateway does not have, and a content refusal all move to the next model — a refusal because policy is a vendor decision, not a fact about the request. 401/403 stop immediately (one gateway, one set of credentials, the next model fails identically), as do 413 and any other 4xx, which are malformed everywhere. Refusals are recognised from the message: no provider sends a machine-readable reason and the status varies. Each hop re-leases the job, so a chain cannot outlive its claim and let a second worker repeat the same paid work, and the row records the model actually being paid for so a picture made by the third model is not attributed to the first. The old singular `fallback_image_model` is still read, so an existing configuration keeps working without anyone re-entering it. **Library.** Documents/Images tabs in My Resources, with a real grid: fixed aspect tiles so the rows line up whatever shape the pictures are, a source badge on the picture, two-line prompt, hover lift, shimmer skeletons while thumbnails land, and a lightbox that closes on Escape or the backdrop and restores focus. Actions are hidden on hover only behind `@media (hover:hover)` — hiding delete behind :hover would put it out of reach on touch and keyboard. Downloads go through privateImageBlob rather than a bare `<a download href>`: a mobile client's session is a bearer token an anchor cannot send, and these assets are served no-store on purpose. The gallery lives in My Resources only. Assistant images appear in it, which was the point; the assistant page does not grow a gallery of its own, and a test asserts no assistant module lists the endpoint. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01Dv6sqaY6Vq3ChZHMem3cnU	2026-09-12 14:53:04 +02:00
Daniel	259b4858be	feat: an image library — every picture this account has made ... A figure outlives the deck it was drawn for: the deck gets replaced, the diagram is still good. Until now a generated image could only be seen in the resource it was made for, and there was no way to find one again or to delete it. Library → Images is a grid of every finished image the account owns, across all three workflows, newest first. GET /api/generated-images is scoped by owner_id in the statement rather than filtered after, returns only finished jobs — an unfinished one is a broken frame in a gallery — and pages by keyset, because a gallery that grows while you scroll repeats or skips rows under OFFSET. Most of this already existed. Thumbnails were already rendered at 256 and 640 by sharp and already served by ?w=, with their own checksum so the client's tamper check passes on a derived copy; hydrateImage already handles auth, the account boundary and caching. The tiles ask for the 256px preview, so thirty of them cost a few kB each rather than thirty full-size downloads, and the prompt is decrypted for the caption because it is the only human-readable label an image has. Deleting needed new work. The storage interface had no remove at all, so a delete that dropped the row would have left the object and both previews in the bucket — paid for, and still readable by anything with credentials. Storage now removes all three keys, and the bytes go before the row: a row pointing at a missing object is a broken image in a gallery, while an object without its row is only wasted space, and unreachable storage refuses the delete outright rather than reporting a success that left the picture behind. THUMB_WIDTHS now has one definition, in generatedImageStorage. Two copies drift, and the drift that matters is a width that gets written and never deleted. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01Dv6sqaY6Vq3ChZHMem3cnU	2026-09-12 14:36:30 +02:00
Daniel	83206e907c	fix: a second deck reused the first deck's pictures ... The image job key was the request-body hash plus the figure's position — slide index for a deck, reply index for a document. Two generations from the same form produce the same body hash, so figure 4 of the second generation collided with figure 4 of the first on the unique (owner, workflow, idempotency_key). The constraint handed back the existing job, and the new deck displayed the old deck's artwork. The decks are not even the same length, so the reused picture could land on a slide about something else entirely. Keyed on what is being drawn now: the body hash stays, so submitting the identical request twice still dedupes rather than billing twice, and a hash of the prompt (plus layout and shape) is what makes two different pictures two different jobs. Same fix in deckBuild and resourceImages. Also split fileLog out of logger. logger requires the database at module load, so importing it to record a diagnostic pulls in a connection pool — wrong on its own terms, and it hung the whole test suite when imageTool started logging its refusals: a unit test that never touches a database inherited an open pool handle and never exited. logger.file now delegates to fileLog, so there is still one implementation of where a line goes and how it is redacted. With that in place, every image-tool refusal is recorded durably. There are five of them, they want five different fixes, and until now none of them left any trace once the container was replaced. Verified against a mutation: restoring the index-based key fails two of the four collision tests. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01Dv6sqaY6Vq3ChZHMem3cnU	2026-09-12 08:26:09 +02:00
Daniel	94f320f140	fix: an empty answer from the model was the one failure never caught ... The Clinical Assistant answered "image summary" with a zero-length string, saved that as the turn, and drew it as an empty bubble. No error was raised, nothing was logged, and the empty turn stayed in the conversation history where it degrades every answer after it. Confirmed by decrypting the saved chat: the final assistant turn is content length 0. The cause is one line. finalizeAssistantAnswer only regenerates an answer that shouldRegenerateTruncatedAnswer flags, and that function opens with `if (!answer) return false` — there is no dangling conjunction to detect in a zero-length string, so empty was classified as "not truncated" and returned as a result. Every other failure mode had a path; this one had none. An empty answer is now asked for once more — a model returning nothing is usually transient — and if it comes back empty again it raises 502 `empty_answer`, which both callers already turn into a visible error. Whitespace counts as empty, and so does an answer that strips to nothing. Without a callAI to retry with it still raises rather than returning empty. Verified against a mutation: removing the new branch fails five of the six tests. This is separate from the image question in the same request. No image job was created, because "image summary" does not match the text fallback pattern — which requires a verb (create/generate/draw/…) before the noun — and the model did not call the tool. That is left alone for now: the request is genuinely ambiguous, and guessing at it is how an assistant starts making pictures nobody asked for. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01Dv6sqaY6Vq3ChZHMem3cnU	2026-09-12 07:41:04 +02:00
Daniel	4dcd0b4fc7	docs: the generation comment still said "one pass, on generation only" ... Modifying now reviews its own result, so that line described the rule as it was a commit ago. Also says what the generation pass can actually see: the deck is rendered there without its figures, because they were enqueued a few lines earlier and are still being drawn, so it judges the space a figure will occupy rather than the artwork that lands in it. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01Dv6sqaY6Vq3ChZHMem3cnU	2026-09-12 07:29:01 +02:00
Daniel	e244ee5240	feat: modifying a deck can see it ... The model editing a deck could not see it, which made most of what people actually ask for unanswerable: "that slide is too crowded", "the diagram is in the wrong place", "this one looks empty" are facts about the rendered page, not about the JSON. When a vision model is configured, modifying now renders the current deck — with its figures, unlike the review pass, which runs while they are still being drawn — and hands the model one image per slide alongside the JSON. Same pipeline as review, reused rather than reimplemented: pptx, Gotenberg, PDF, pdftoppm, capped at MAX_SLIDES. The vision model then does the editing, which is a second and separately measured benefit. On a real 20-slide deck, ds-deepseek-v4-flash returned the deck unchanged for "make it better" — the echo reported yesterday — while openrouter-gemini-3.8-flash applied it. So the stronger model fixes the echo even without sight. A model the author picks explicitly still wins over both. The result is rendered and reviewed again. Generation-only was the old rule, on the reasoning that refining is a text edit; it is not. The edit is made against how the deck looked before it, so a slide that gains two bullets only overflows once it is rendered again. The reviewer may reposition but is held to the same words, so a verification pass cannot quietly undo what was just asked for. Whether an instruction achieved anything is judged on the model's edit, before the reviewer runs, or a reviewer nudging a slide into two columns would mask an instruction that did nothing. Sight is an upgrade, never a dependency: no vision model, Gotenberg down, a render that fails — each falls through to editing blind, and a test covers each of those paths. Verified against two mutations: keeping the text model when images are attached, and dropping the verification pass, each fail exactly one test. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01Dv6sqaY6Vq3ChZHMem3cnU	2026-09-12 06:55:16 +02:00
Daniel	973f9d554f	feat: a text-only model cannot be set as the slide reviewer ... The reviewer is shown rendered images of the deck, so a text-only model there fails on every generation — at request time, long after the one moment an administrator could have picked differently. Nothing checked it. Saving my_resources.review_model now asks the gateway what it reports for that model and refuses only an explicit supports_vision === false. Three answers, not two. Most of this roster carries no supports_vision at all (every openrouter-* entry here), and refusing unknowns would block the reviewer this deployment already runs on. An unreachable gateway is not evidence about a model either, so it never blocks the save. Empty means review is off and skips the lookup entirely. The capability is read from the gateway, never inferred from the model id. Verified against a mutation: relaxing `canSee === false` to `canSee !== true` fails the two tests that say unknown must stay allowed. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01Dv6sqaY6Vq3ChZHMem3cnU	2026-09-12 06:42:44 +02:00
Daniel	491a2b0811	fix: the My Resources diagnostics survive a deploy ... logRefine writes the one line that answers "did that modification change anything" — path, before and after size, CHANGED=yes/no, figures, model, instruction. It went to console, so it lived in the container's stdout and was destroyed the next time the container was recreated. That cost a diagnosis today: a modification came back unchanged, the user asked why, and the evidence had already been deleted by a deploy. The deck-fallback warnings and the deck-vocabulary gaps had the same problem, and those exist specifically to be read later — the vocabulary gaps are meant to show which shapes to build next, which is a question about weeks, not about one container. All of them now go through logger, which writes the dated file in the scribe-logs volume and ships to Loki when it is configured, and carries the event as structured data rather than only as a formatted string. console.error is left alone: those are failures, and logger.error already echoes to the console. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01Dv6sqaY6Vq3ChZHMem3cnU	2026-09-12 06:09:34 +02:00
Daniel	c9b6d6b3c9	docs: deploy.sh is the deploy path, and it was documented nowhere ... scripts/deploy.sh landed on 2026-09-11 as "a deploy you can repeat, and prove afterwards". It moves the PED_AI_IMAGE pin, waits for health, reads /api/build to find out which revision is actually serving, and rolls back when that is not the revision asked for. Its own comment says it: "Healthy is not the same as running what was asked for." No doc mentioned it. Both deployment.md and DEVELOPMENT.md instead showed `build-image.sh` then `docker compose up -d --no-build` then `curl /api/health` — a sequence that moves no pin and proves no revision. Following it deployed nothing four times in one session, eventually reverting the app by 31 commits and removing a feature, which was then reported as a bug. Both docs now lead with deploy.sh and say why `up` by hand is not a deploy. Also replaced yesterday's build-image.sh warning, which told the operator to `sed` the pin themselves. That reimplemented, badly, one of the three things deploy.sh already does — and it was written while auditing these very files without noticing the script was there. It now prints the deploy.sh line to run. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01Dv6sqaY6Vq3ChZHMem3cnU	2026-09-12 06:04:13 +02:00
Daniel	739286b53b	fix: warn when the built image is not the one a deploy will start ... `docker compose up` starts whatever PED_AI_IMAGE in .env names, and build-image.sh does not move that pin. A pin left from an earlier deploy therefore starts the older image while every signal reports success: the build completes, `up` says the container started, and /api/health returns {ok:true} from the wrong revision. This is not hypothetical. The pin here had been sitting on a revision from two hours before the Modify card was added, so a rebuild-and-restart rolled My Resources back 31 commits and removed the feature. The missing card was then reported as a new bug, and three deploys in this session had in fact deployed nothing. build-image.sh now compares the pin to the revision it just built and, when they differ, prints the pin, says that `up` will start it instead, and gives the command to move it. It does not correct the pin: naming a revision is also how a deliberate rollback is done, so this is said rather than silently overridden. Both deployment docs now check /api/build against `git rev-parse HEAD` after starting, because /api/health passing only proves a container is up, not that it is the one you built. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01Dv6sqaY6Vq3ChZHMem3cnU	2026-09-12 05:58:37 +02:00
Daniel	07d1c541a0	fix: a deck the model fumbles once is asked for again, not abandoned ... A presentation is generated as a typed deck — the model chooses each slide's layout — and that deck is what scripts/render_pptx.py renders. When the reply did not parse as a deck, the route fell straight back to asking for markdown, and slideSpec.build() then inferred slides from it. Both paths go through python-pptx, but the fallback's layout is guessed from heading and list structure rather than chosen, so everything lands as title-and-bullets. Measured on the stored library: since decks landed, 7 of 8 generations produced one and 1 did not. Models are stochastic, so one unlucky reply was costing the whole layout. It now asks a second time with the same prompt before giving up. The fallback was also invisible. It warned to the console, where the person who would simply have generated again could not see it, so they kept the plainer deck without knowing a better one was one click away. The response now carries deckFallback and the UI says it came out as plain slides, and why. Fixed the reason heuristic while adding tests for it: truncation was claimed for any reply not ending in "}", which is every prose refusal. It is now only claimed for a reply that began as JSON and stopped. The four generate tests run the handler. Verified against a mutation: removing the retry fails tests 6 and 7 and nothing else. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01Dv6sqaY6Vq3ChZHMem3cnU	2026-09-12 05:34:14 +02:00
Daniel	e69eb9a9f7	fix: modifying a presentation failed whenever illustration was ticked ... existingDeck was declared below the branch that reads it, so `var` hoisting made it undefined there. With illustration on, a deck modification therefore appended the *markdown* instruction — "Returning the markdown is still required; a tool call is not a substitute for it" — to a prompt whose body asked for deck JSON. The model was told to produce two different artifacts in one reply, the reply parsed as neither, and deckBuild.parse returned null, so the handler answered 502 "That change could not be applied." Moved the declaration above its first reader. Same class of fault as the savedFigureIds one, in the same file. Two things made it hard to see, both fixed: - The library row read created_at, so a modification that did apply left the visible timestamp on the generation time. That timestamp is what led to "modification doesn't work" — it was the only signal available, and it was reading the wrong column. Rows now show the modified time when there is one. - A model can also return the document back unchanged. That was logged server-side and answered "Applied. Download it to see the result", which sent people to download an identical file. The response now carries `unchanged` and the UI says so, keeping the instruction in the box so it can be reworded. Also surfaced has_deck on the library list: 28 of 38 stored presentations have no deck and go through the weaker flat-markdown path, and nothing in the UI distinguished them. They now read "plain text, no slide layout". test/my-resources-refine.test.js runs the handler rather than reading it, since all three faults were invisible to source reading. Verified against a mutation: putting the declaration back where it was fails test 1 and nothing else. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01Dv6sqaY6Vq3ChZHMem3cnU	2026-09-12 05:05:39 +02:00
Daniel	cfb8aab77d	fix: /api/health/detailed reported a text-to-speech provider that does not exist ... The endpoint answered from environment variables of its own rather than from the speech code, so ELEVENLABS_API_KEY being set made it report tts: 'elevenlabs'. There is no ElevenLabs path in ttsProvider.js — getTTSProvider() only ever returns 'litellm' or 'none' — so the one endpoint an operator checks to find out what speech is doing was reporting a provider this app cannot use. Likewise whisper: OPENAI_API_KEY, which stopped describing STT when transcription moved behind the gateway. Both now ask getSTTProvider() and getTTSProvider(), the same functions the routes ask, so the answer cannot drift from behaviour again. Removed the two other ElevenLabs leftovers: the FAQ told users their notes were read aloud by "Google, OpenAI, or ElevenLabs", and a TTS test used 'elevenlabs' as its sample value for the passthrough of TTS_PROVIDER, which implied the provider was supported. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01Dv6sqaY6Vq3ChZHMem3cnU	2026-09-12 05:05:39 +02:00
Daniel	67e416c6d9	docs: merge the duplicate pairs and correct them against the running app ... Three pairs of docs described the same thing twice, and the copies had drifted apart. Merged each into one file, keeping the unique content from both: - ARCHITECTURE.md -> architecture.md (its operational map: ownership, request flow, runtime boundaries, source of truth, deployment shape) - DEVELOPMENT.md -> developer-guide.md (change workflow, Clinical Assistant high-risk areas, frontend rendering rules, deployment checks) - transcription-options.md -> speech.md (the clinic setup table, and the list of browser-Whisper paths that must stay removed) Then audited what remained against the code and the live database rather than against the previous docs. Corrected: - Google Vertex was still documented as a provider across nine files. The SDK is gone; AI_PROVIDER=vertex now logs an advisory and falls back to OpenRouter, and Gemini is reached through LiteLLM. Fixed the provider selection order to match src/utils/ai.js, which starts from LITELLM_API_BASE. - promptSafe was documented on 8 routes; it is on 13. - Node 20 -> 24, "24 vanilla JS modules" -> no fixed count, and transcribe.js/tts.js -> sttProvider.js/ttsProvider.js, which is what exists. - STT/TTS are LiteLLM-only; README listed direct Google, AWS Transcribe and ElevenLabs paths that are not in the runtime. - Learning Hub PPTX export was documented as pptxgenjs, which is not a dependency. It is pandoc against a reference deck. - POST /api/admin/milestones/seed does not exist; it is /bulk-import. - NEXTCLOUD_URL and NTFY_TOPIC are not read anywhere. Nextcloud is per-user in the users table, and the ntfy topic is derived as pedscribe-{userId}. - A prose paragraph sat inside the Clinical Assistant settings table, so half the rows rendered as text. Filled the gaps the audit exposed: - database.md was missing 12 of 29 tables, including user_resources, personal_notes, login_codes, registration_invites and generated_image_jobs. - developer-guide.md was missing 11 routers and 10 frontend modules. - api-reference.md detailed 121 of 244 endpoints and said so, but whole features were absent. Added an endpoint index covering Clinical Assistant, My Resources, Notes, Diagrams, ED Encounters, invites and sign-in codes. - configuration.md was missing METRICS_TOKEN, REDIS_URL, API_RATE_LIMIT_MAX, the LITELLM_* model variables, the DB_* ones maintenance.js reads, and the per-purpose S3 resolution scheme. - clinical-assistant.md documented 2 of its 17 environment variables. - features-explained.md had no entry for My Resources or Clinical Assistant. Renamed the three remaining SHOUTING filenames to kebab-case, which is what the docs viewer's prettyName() was working around, and rewrote README's index, which listed architecture.md twice and omitted nine files. Noted but not changed: the Turnstile site key is hardcoded in index.html rather than read from TURNSTILE_SITE_KEY, and /api/health/detailed can report tts: 'elevenlabs' though no ElevenLabs path exists. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01Dv6sqaY6Vq3ChZHMem3cnU	2026-09-12 04:57:35 +02:00
Daniel	5614a436be	fix: a deck asked for a figure and never got one — three causes, one symptom ... "Include a diagram" produced decks with no picture. Three separate faults, each hiding the next, found by generating the same deck after fixing each one. First, deck generation ran on the default 4000-token budget. A deck's JSON is several times the size of the prose it holds, so a long deck came back truncated, failed to parse, and fell back to markdown — which has no way to request a figure, so the model described one instead and the slide rendered a literal "![Placeholder: Flow diagram ...]" as its first bullet. Deck generation now gets room, and the fallback says how the reply failed: empty, cut short at N characters, or not a deck. Second, the figure request sat inside the layout vocabulary, one line among forty, and the model passed over it. It goes last now, after the author's own instructions — the same placement lesson the image tool taught earlier. Third, and the one that actually mattered: image_prompt is only read on the figure and image types, so an image_prompt on a bullets slide was dropped in silence. The instruction said "add image_prompt to N slides" without saying which types carry one. It now names them, and a misplaced request is honoured rather than discarded — a slide with words becomes a figure, one without becomes a full-slide image. Image markup is also stripped wherever text enters a slide, on both paths: a described figure is not a figure, and a bullet of raw markdown is worse than no bullet. Verified end to end after: the same request produced a deck with one figure, the job completed, and the exported pptx carries one embedded image across 21 slides. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01Dv6sqaY6Vq3ChZHMem3cnU	2026-09-12 04:28:30 +02:00
Daniel	2c3fbbcf37	fix: decks were falling back to markdown, so no figure could ever be requested ... A deck's JSON is several times the size of the prose it holds, and generation used the default 4000-token budget — raised for refine and for slide review, but never here. A sixteen-slide deck ran past it, came back truncated, failed to parse, and fell back to markdown. Markdown has no way to ask for a figure, so the model described one instead and the slide rendered a literal "![Placeholder: Flow diagram — "Neonate with rash" → ...]" as its first bullet, above the steps it was meant to illustrate. That is why no generated deck was arriving with an image. Deck generation now gets room for a deck. The fallback also says how the reply failed — empty, cut short at N characters, or simply not a deck — because those want different fixes and "not usable" covered all three. Image markup is stripped wherever text enters a slide, on both the deck and markdown paths, since a described figure is not a figure and a bullet of raw markdown is worse than no bullet. The model is also told plainly: if a figure is wanted say so with image_prompt, and if that is not on offer, write the slide without one rather than describing the picture you would have drawn. Separately, the Documentation list showed ARCHITECTURE, CLINICAL_ASSISTANT, DEVELOPMENT, MODULE_CONVENTIONS and SCALING shouting in caps with underscores intact: the label builder replaced hyphens but not underscores, and uppercased the first letter of each word rather than normalising the case, so a SHOUTING_FILENAME stayed shouting. It now reads "Clinical Assistant", keeps acronyms as acronyms (AI, API, OpenID, LiteLLM) and leaves joining words lower. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01Dv6sqaY6Vq3ChZHMem3cnU	2026-09-12 04:21:20 +02:00
Daniel	9b2cb339a1	test(e2e): drive My Resources through the browser, and assert what it sends ... The unit tests for this feature read source files and assert patterns. They prove the code says the right thing, not that the screen does it, and nothing exercised the browser at all — so a mismatch between what the form sends and what the route reads passed all of them. Three real bugs shipped through that gap in one session: a modification that updated the markdown but not the deck, generation that failed whenever the slide reviewer was off, and a figure generated for a slide that never referenced it. Every one was found by driving the running server by hand. So these assert the request bodies, not only the rendering: that Generate sends topic, kind, slideCount, refinement, model and all four options as the strings the route compares against; that unticking the library sends 'false' rather than omitting the field, which the route would read as on; and that Modify posts to the right resource with every source option. Plus the screen's own behaviour — availability gating on both cards, the illustration hint switching on and staying off once overruled, the bounded searchable library, the two different empty states, an article never being offered as slides, a local refusal that spends no round trip, and a refused modification surfacing its reason. Fourteen tests, both viewports. The API is stubbed. This is the contract between the screen and the route, and stubbing keeps it fast, free and deterministic. Proven to catch regressions rather than merely pass: renaming useCorpus in the form failed two tests, breaking the availability gating failed one, and truncating the modify picker failed another. Two flakes of my own were fixed rather than retried. openTab slept 400ms for the library and picker instead of waiting for them, which made Modify report "nothing to modify yet" under load. And the console-error guard failed on net::ERR_ABORTED and net::ERR_NETWORK_CHANGED — a request in flight when the context closes, and the host network reconfiguring under a browser that runs on it. Both are the harness, not the page: anything genuinely failing carries a status code and is still caught. Five consecutive clean full runs after. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01Dv6sqaY6Vq3ChZHMem3cnU	2026-09-12 01:51:15 +02:00
Daniel	3ec65a91f6	fix: a figure asked for while modifying a deck now belongs to a slide ... Ticking "Add illustrations" on Modify offered the image tool regardless of what was being edited. The tool returns job ids and has no way to place them, which is fine for markdown — there is nowhere to put a figure in markdown anyway — and wrong for a deck, where figures are placed by a slide declaring them. So modifying a deck with illustrations on generated a figure, paid for it, recorded it against the resource, and referenced it from nothing. Measured: one figure recorded, zero referenced by a slide, and absent from the export. Deck mode now asks the revised deck to declare its figures and draws them with the same drawFigures() generation uses, so each one belongs to the slide that wanted it. Slides that already have a figure keep it. The tool path stays for markdown resources, where it is the only option. Verified: the same modification now records one figure, one slide references it, and the exported deck embeds one image. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01Dv6sqaY6Vq3ChZHMem3cnU	2026-09-12 01:24:40 +02:00
Daniel	012346528c	fix: generation stopped working whenever the slide reviewer was switched off ... savedFigureIds was declared inside the review branch, so with no reviewer configured — the default, and what everyone is running — it was undefined by the time the INSERT stringified it. JSON.stringify(undefined) is not a string, the column is NOT NULL, and every generation failed with "Generation failed". `var` is function-scoped, so nothing complained until the database did. This is the second bug of exactly this shape in this file, so the test asserts position rather than presence: the value must be declared before both the review and the insert read it. Found by the logging added in the same change, which is the other half of this commit. Every modification now says what it did: [my-resources] refine id=29 path=deck outcome=applied 13→14 slides changed=yes [my-resources] refine id=37 path=markdown outcome=applied 2635→3018 chars changed=yes CHANGED=no is warn-level and deliberately shouty, because that is the failure worth catching: the response says success either way, the row updates, and the download is identical — which is exactly how the deck bug went unnoticed. A refusal logs its reason. ped_ai_resource_refine_total{path,outcome} counts the same thing over time, so "did that modification do anything" is answerable without watching logs live. Verified across every path rather than the one that was broken: a deck presentation modified and exported to both pptx and docx carries the change; a legacy presentation with no stored deck still takes the markdown path and carries it; an article generates, modifies and exports; and a presentation generates with the reviewer off. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01Dv6sqaY6Vq3ChZHMem3cnU	2026-09-12 00:57:30 +02:00
Daniel	f66daf0c02	fix: modifying a presentation changes the presentation, not just its markdown ... Export renders a presentation from its stored deck. Refine edited the markdown beside it and never touched the deck — so a modification reported success, updated the title and the library row, and produced a byte-identical download. Nothing said otherwise. It looked like the model had ignored the instruction. Measured before the fix, with a marker that was definitely not in the deck: refine succeeded, the stored markdown gained the new slide, the stored deck did not, and the exported pptx did not. After: the export gains the slide and the marker, twelve slides where there were eleven. A presentation with a stored deck is now edited as a deck — the deck goes to the model, a revised deck comes back, and the markdown is serialised from it, which is the same direction generation runs in. Layouts, custom slides and image_job values survive a modification instead of being flattened away. A reply that is not a usable deck is refused rather than saved as markdown: saving it would drop every layout the deck held while looking like it worked, which is the failure this commit exists to remove. Articles have no deck and keep the markdown path unchanged. The reply restates the whole resource, so the token budget is raised to match — the old default was already close to truncating a long deck's markdown. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01Dv6sqaY6Vq3ChZHMem3cnU	2026-09-11 23:55:27 +02:00
Daniel	4f8e686907	feat: record what a deck wanted and could not have ... The shape vocabulary is deliberately small, which leaves the question of what to add next. Rather than guess, it now records demand. Two signals, because a model asks both ways. It can say so outright — {"kind":"unsupported","need":"a SmartArt cycle of four stages"}, which draws nothing and is told about in the same file that validates it — or it can reach for a kind, chart type or slide type that does not exist, which is the more common way of asking and just as much of a signal. Both produce a log line naming what was wanted and the topic it came up on, and increment ped_ai_deck_vocabulary_gap_total{wanted}, so it can be counted over time in Grafana rather than noticed once and forgotten. Deduplicated per generation and capped at twelve: a model that asks for a hundred things it cannot have should not write a hundred log lines. It can never fail a generation — it is a note to whoever decides what to build next. This is also the answer to whether to run model-authored code in a sandbox instead. The log will say whether the gap is real. Some of it is not closeable by any sandbox, being python-pptx's own ceiling — no SmartArt, no animations or transitions, limited chart types — and a sandbox would only let a model write code against the same library and hit the same wall. Documented in docs/my-resources.md, which the in-app Docs tab serves directly. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01Dv6sqaY6Vq3ChZHMem3cnU	2026-09-11 22:51:15 +02:00
Daniel	af2e09c1de	feat: a slide can be drawn from primitives when the named layouts have no word for it ... The nine layouts are a fixed vocabulary and a good default, but "lay the three severity levels out left to right with arrows between them" had no expression in them at all. A "custom" slide now carries a list of shapes: positioned text, eight autoshape families, lines, images, tables, and native PowerPoint charts — column, bar, line, pie, doughnut. Coordinates are percentages of the slide rather than EMU, because a model reasons about "the left half" and not about 12192000. Shapes draw in array order, so a later one sits on top. The model never emits Python. It names shapes and the renderer draws them: running model-authored code to lay out a slide would be an enormous amount of trust to buy a feature, on a server holding clinical data and secrets. Validation lives beside the text that teaches the vocabulary, in one file, so what the model is told about is exactly what is accepted. Kinds are an allowlist, colours must be six hex digits, coordinates are clamped inside the slide — a shape at x=95 w=30 is cut to the edge rather than drawn half off it — counts are capped, a pie is held to one series, and anything that cannot be understood is dropped. A custom slide that loses every shape becomes a plain one rather than a heading over an empty frame, and one bad shape is caught in the renderer so it cannot cost the slide it sits on. A figure on a custom slide is requested through an image shape, drawn by the same path as any other, and attached by job id. Word renders a custom slide as its words in reading order with its tables and figures — lossy, and better than dropping the slide. Verified live end to end: asked to "lay the three severity levels out left to right as coloured boxes with arrows between them", the model produced [rect arrow rect arrow rect], chose green/amber/red itself, and the rendered slide was looked at. A column chart beside its commentary renders with real axes and gridlines. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01Dv6sqaY6Vq3ChZHMem3cnU	2026-09-11 22:12:26 +02:00
Daniel	f7e0277552	docs: accounts in the README, and the sign-in code path is verified end to end ... The README's feature list never mentioned accounts at all. It now says how somebody signs in — password or an emailed code, side by side — that passwords are argon2id with bcrypt rows rehashed on next sign-in, and that registration can be open, closed or invite-only with codes that are revocable while live and deletable only once spent. The browser-STT note now says why the Web Speech gate exists rather than only that it exists: Chrome and Edge send that audio to Google, so it is off unless a user turns it on. And the one thing that could not be demonstrated before now can. With SMTP configured, a real send returns true, and driving the sign-in screen in a browser — request a code, type it — returns 200 from /login-code/verify and enters the app. docs/authentication.md records that. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01Dv6sqaY6Vq3ChZHMem3cnU	2026-09-11 22:02:25 +02:00
Daniel	154b896d5b	feat: Word is built by python-docx from the same typed source as the deck ... Pandoc reads markdown, so every Word export had to flatten the resource to markdown first — and a deck flattened to markdown stops being one. A comparison became two headings and two lists, a callout became bold text, and a figure became nothing at all, because markdown has nowhere to put it. src/utils/docSpec.js reduces either source to the same blocks: a stored deck where there is one, the markdown where there is not. scripts/render_docx.py draws them. A comparison comes out as a labelled two-column table, a callout as a shaded box, a table as a real table, a figure embedded at its own aspect ratio with its caption, and speaker notes as muted indented text. The deck wins over the markdown beside it, because that markdown is a serialisation of the deck and reading it instead would be reading a lossy copy of what is right there. Word now carries the figures too. The export route skipped fetching them for docx, which was correct when pandoc could not place them and wrong the moment this could. Pandoc stays installed and stays the fallback: a plainer document beats a failed download. Both renderers now share one spawn helper. Verified end to end: a deck with two figures exported as a six-page Word document with both images embedded (537KB, two files in word/media), rendered to PDF and looked at — the comparison is a labelled table, the figure sits at its true aspect ratio, and the notes read as notes. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01Dv6sqaY6Vq3ChZHMem3cnU	2026-09-11 21:58:53 +02:00
Daniel	9d307fd442	feat: a vision model looks at the rendered deck and fixes the layout ... The model that writes a deck never sees it. It cannot tell that slide four overflowed, that a nine-item list would read better in two columns, or that two labelled groups want to be a comparison — those are facts about the rendered page, not about the text. So each generated deck is now rendered to PDF through Gotenberg, rasterised to one image per slide with pdftoppm, and shown to a vision model. Off unless an administrator names a reviewer, in its own admin card because it is the one setting that spends money on every generation without a user having asked for anything. One pass, on generation only: a second pass costs as much as the first and fixes far less, and refining is a text edit. It returns a patch, not a deck. Asking for the corrected deck back put the reply in proportion to the deck rather than to the number of problems, and a fourteen-slide deck came back cut off mid-object at every output budget the provider would honour — measured twice before changing shape. The patch is better for a second reason. The reviewer names a slide and an action — two columns, one column, split after bullet N, compare with these two labels — and the server moves the text it already has. The words never pass through the model, so a review cannot reword, drop or invent a single bullet. That is a stronger guarantee than instructing it not to and checking afterwards. The check runs anyway, because a bug in applyChanges would be as bad as a model rewriting the words and worse for being trusted: body text must come out the same multiset, figures the same set, and a heading may only be reused or extended. A continuation heading is the reviewer's one piece of text and is replaced when it does not continue anything. Nothing here can fail a generation — no reviewer, an unreachable one, an unparseable reply, a deck too long to look at, or a patch that applies to nothing each return the deck that was written. Verified end to end against a deck with a deliberately overloaded slide: three slides rendered and sent, one change returned, ten bullets split into five and five under "Stepwise Management … (continued)", text intact. Left switched off; enable it under Admin. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01Dv6sqaY6Vq3ChZHMem3cnU	2026-09-11 21:38:18 +02:00
Daniel	f3c3f47d99	fix: voice mode reads the answer that just arrived, and reads what the page shows ... Two bugs, one cause each. It read the previous answer. setBusy(false) is what announces assistant-answer-done, and it ran before lastAnswer was assigned — so every listener was handed the answer before last. It now fires after the answer exists both in that variable and on the page. A test asserts the order, because the order is the whole bug. And it read the markdown. The better answer than unpicking the markup is not to have any: the rendered bubble is already the answer with its headings, emphasis and tables resolved, so voice mode reads that. It cannot drift from what the reader is looking at, and it needs no rules about what "##" sounds like. Read from a clone, with the parts that are not the answer removed — the action buttons, the sources list, the follow-up suggestions, code blocks and tables — so the page itself is untouched. A bubble still thinking is never read. speakableText() stays as the fallback for when the bubble cannot be found, since raw markdown read aloud is worse than silence. Separately: e2e/seed.js hashed with bcrypt directly, so seeded accounts did not exercise the argon2id path production writes. It uses the app's own hasher now. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01Dv6sqaY6Vq3ChZHMem3cnU	2026-09-11 21:10:50 +02:00
Daniel	0e17f553fc	fix: hanging up the voice call stops the voice, and the answer is read not its markup ... Ending the call stopped listening and aborted the request but never stopped playback, and the <audio> element was a local variable with no handle kept — so nothing could stop it and the answer talked on until the page was reloaded. conversationMode now holds whatever is speaking, whichever of the two players it is, and hanging up silences it before anything else. Starting a new turn does the same, because speaking over the previous answer is how an assistant talks past you, and a reply that arrives after the overlay has closed no longer starts talking into a closed call. The other half: the raw answer went straight to the speaker. It is markdown, so a browser voice reads "#" and "**" aloud or stumbles over them. speakableText() now reduces it to what a person would read out — headings and bullets become sentences, links keep their words, emphasis and code fences are dropped, tables are dropped entirely because a table read aloud is noise, and a removed citation marker leaves no gap before the punctuation it preceded. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01Dv6sqaY6Vq3ChZHMem3cnU	2026-09-11 20:52:35 +02:00
Daniel	fa2e7523d6	docs: My Resources, sign-in codes, invitations, and what the image carries ... Nothing documented My Resources, the slide renderer, PubMed or web search, and the authentication doc predated both sign-in codes and registration invitations. docs/my-resources.md is new and covers the feature end to end: what a resource is, where its material comes from, why both searches run in the route rather than as tools the model never called, why keyword engines get the topic while retrieval gets the instruction too, how a presentation is designed as a deck rather than written as markdown, the separate multi-image path, and what the export pipeline is made of. docs/authentication.md gains sign-in codes — storage, lifetime, reuse, supersession, guessing, and that two-factor still applies — and registration invitations, including the exact condition that decides when a code may be deleted and why it is written to match the status the list displays. Both new rate limits are in the table, with a note that Express matches app.use paths on segment boundaries, so a new sign-in endpoint needs its own limiter or it has none at all. docs/deployment.md now says what the runtime image carries and why — pandoc for Word, python3 with apk-installed lxml and pillow for the slide renderer, python-pptx pinned, and that PDF conversion is not in the image at all but goes to Gotenberg, so Word and PowerPoint still work when it is down. docs/configuration.md picks up LOGIN_RATE_LIMIT_MAX, LOGIN_CODE_RATE_LIMIT_MAX and GOTENBERG_URL, none of which were listed. README gains a My Resources section and indexes the two new docs. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01Dv6sqaY6Vq3ChZHMem3cnU	2026-09-11 20:49:10 +02:00
Daniel	b98ffc61c7	fix: expired invitations can be cleared too, revoked ones still cannot ... An expired code is as dead as a used one and just as accounted for, so it is now deletable. The rule the code enforces is the one that matters: a code that could still be redeemed is never deleted, because that takes it off the list without taking it out of anybody's inbox — the holder keeps something that looks valid, it quietly stops working, and nothing is left to say who had it. One condition, shared by the single delete and the bulk clear: (used_at IS NOT NULL OR (revoked_at IS NULL AND expires_at <= NOW())) Written that way rather than as "used OR past its date" because the second form also catches a revoked code whose date has since passed — a row the list still labels revoked and offers no delete on, so the button and the query would have disagreed about the same row. Revoked codes keep their rows. Revoking records a decision somebody took, and a handful of them is not the clutter a pile of expired codes is. Verified against the live database across every state: active refused, used deleted, expired deleted, revoked refused, and revoked-with-a-past-date refused rather than slipping through as expired. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01Dv6sqaY6Vq3ChZHMem3cnU	2026-09-11 20:41:18 +02:00
Daniel	7b084c7edf	fix: an invitation can only be deleted once it has been used ... The delete button was offered on every invitation regardless of state, and the query behind it deleted any row it was given. Deleting an unused code takes it off the list without taking it out of anybody's inbox: the person still holds something that looks like a valid invitation, it silently stops working, and there is no longer a record of who it went to or why. Revoke is what stops a live code — it leaves the row behind, marked. So the delete is now for spent codes only, in three places rather than one: the query carries AND used_at IS NOT NULL, the route answers 409 with the reason instead of pretending the row is missing, and the button is rendered only on a used row. A "Clear N used" control alongside, since the complaint was clutter and clearing them one at a time is not much of an answer. Same rule — nothing unused or revoked is touched — and it confirms first, because it is still a delete. The bulk route is declared before /invites/:id, or Express reads "used" as an id. Verified against the live database: deleting an unused invitation is refused and the row survives, deleting a used one works, the bulk clear removes only used ones, and the unused probe row was still there afterwards. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01Dv6sqaY6Vq3ChZHMem3cnU	2026-09-11 20:30:06 +02:00
Daniel	22683f3584	feat: sign in with a code emailed to you, offered beside the password ... The sign-in screen asks for an email first, then offers both ways in together: a six-digit code sent to that address, or the password. Beside rather than instead — a code depends on mail being delivered and a password does not, so neither may be the only route. "Use a different email" goes back a step, and creating an account stays where it was. What keeps it from being a second, weaker front door: - Only a bcrypt hash is stored, so a code read out of the database is not a working credential. - Ten minutes, single use, marked used before the session is issued so a replay cannot race it, and requesting a new one deletes the old. - Five wrong guesses burn it. Six digits is a million possibilities, which is plenty against a person and nothing against a script with unlimited tries. - Requesting a code answers identically whether or not the address exists, and every verify failure returns one message. A sign-in screen that says "no such account" is a way of finding out who has one. - Two-factor still applies: a code proves you can read the mailbox, which is one factor, and an account that asked for a second still wants it. - Its own rate limits, tighter for requesting than for attempting, because requesting sends mail to someone else's address. These had to be separate limiters: Express matches app.use paths on segment boundaries, so /api/auth/login does not cover /api/auth/login-code — checked against a real router rather than assumed. Two bugs found while building it, both mine: authFetch keeps an allowlist of endpoints callable with no verified owner and rejects everything else before it is sent. The new endpoints were not on it, so the request never left the browser and surfaced as "Connection error". reveal() hid elements by appending 'hidden' to className and showed them with a non-global replace, so hiding twice left two copies and showing stripped one. The "use a different email" link never reappeared. It uses classList now, which is idempotent. Verified against the running server: correct code signs in, the same code again is refused, a superseded code is refused, five wrong guesses burn it, an expired one is refused, and the stored value is a hash. In the browser: requesting a code advances the screen, a wrong code is refused without losing the screen, and the password route still signs in. Not yet demonstrated: a correct code typed into the browser. The harness keeps racing the one-live-code rule — the page's own request supersedes whatever code the test holds, and with SMTP off the delivered one cannot be read. The same request reaches the server on the wrong-code path, and the endpoint itself is verified, but that last step is untested end to end. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01Dv6sqaY6Vq3ChZHMem3cnU	2026-09-11 20:12:03 +02:00
Daniel	087f717f55	feat: the model designs the deck instead of writing markdown for a parser to guess at ... Markdown could express about five of the things the renderer can draw, so the model had no way to say "put this figure beside these three bullets" or "make this a comparison with two labelled columns" — my parser inferred a layout from the shape of a list, and inferring is what made every deck look the same. A presentation is now described as a deck: the model returns JSON naming a layout per slide and the prompt it wants each figure drawn from. Four layouts were added to the renderer for it — two tinted labelled columns for a comparison, a callout card for a red flag or a dose, a figure beside its bullets, and a full-slide figure. Articles stay markdown, which is what prose wants. Markdown is still produced, serialised from the deck, so Word export and text editing keep working and the stored artifact stays readable by a person. The deck is stored alongside it because that serialisation is lossy by design: round-tripping through markdown would throw away exactly the layout choices this was built to capture. A resource made before this, or an article forced into slides, still renders by inferring from its markdown. Nothing here can cost more than the thing that went wrong. A reply that is not a deck falls back to asking for markdown rather than saving the model's apology; a malformed slide degrades to bullets rather than throwing; a comparison with one column is not a comparison; a figure that cannot be queued leaves a slide of text rather than an empty frame; and JSON wrapped in fences or a covering sentence is read rather than refused. Verified live on "croup versus epiglottitis": the model chose section, bullets, table, compare, figure, callout and image layouts across thirteen slides, and the exported deck was rendered to PDF, rasterised and looked at — the comparison renders as two tinted cards, the red flag as a callout, and the figure sits beside its bullets. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01Dv6sqaY6Vq3ChZHMem3cnU	2026-09-11 19:49:55 +02:00
Daniel	73ce4049d4	feat: decks are built with python-pptx instead of pandoc, and carry their figures ... Pandoc's pptx writer was the ceiling on how good a generated deck could be, and the model on top made no difference to it. It maps markdown onto a handful of reference layouts with no per-slide layout, no positioning and no control over how large an image is drawn, which is why every deck came out as bullets on a template — and why autofit had to be injected into its emitted OOXML by hand afterwards, because LibreOffice ignores the autofit pandoc leaves off. scripts/render_pptx.py draws the deck and src/utils/slideSpec.js decides what each slide is. Markdown stays the stored artifact, so "change slide 4" is still a text edit and Word export is untouched — pandoc still writes docx, where its output is good. What that buys, all of it visible in a rendered deck rather than argued for: - 16:9, not pandoc's 4:3. - A pipe table becomes a real table with a header band and banded rows, not eight lines of text with pipes in them. - A list longer than seven items becomes two columns instead of a wall of text. - Text is measured and sized to fit before the file is written, so nothing depends on a renderer honouring autofit. - Wrapped lines hang under the text instead of running back to the margin, which is the clearest single tell that a deck was generated. - An image is drawn at its own aspect ratio, centred, with a caption. Figures now reach the deck at all, which they never did. They were queued and shown on the page, but nothing recorded that they belonged to the resource, so an export could not include them: user_resources.image_ids holds them, a modification adds to that list rather than replacing it, and export fetches the finished ones to a scratch directory. They are spread through the deck rather than appended, because ending on three unexplained pictures is worse than showing each near its material, and a References slide stays last. If the renderer fails for any reason, pandoc still produces a deck — a plainer deck beats a failed download. Verified end to end: a seven-slide request with three figures exported as a 13-page deck; the slides were rendered to PDF, rasterised and looked at. All three formats still download. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01Dv6sqaY6Vq3ChZHMem3cnU	2026-09-11 19:34:26 +02:00
Daniel	1ad72b134b	feat: a resource can have several illustrations, on its own path ... My Resources used imageTool.dispatch, which permits exactly one image per request — "Only one image tool invocation is permitted per request". That is the right rule for a chat reply and the wrong one for a twelve-slide deck where the author asked for three figures. Rather than relax a limit the clinical assistant and the Learning Hub also depend on, this adds a separate dispatcher for this feature. Same queue, same storage, same my_resources workflow, same asset endpoint — only the number of figures differs, bounded at six because each one is a paid request. "Use 3 diagrams" in the instructions is read as the number it is, and the illustration option now says several are possible rather than promising one. Three things had to be got right, each found by measuring rather than assuming: The illustration guidance has to be the last thing in the prompt. Placed before the output rules it lost — with the tool offered and the paragraph present, the model returned 3297 characters of markdown and zero tool calls, while the same tool and wording in a shorter prompt produced three calls. Even last, it loses to a prompt carrying thirty library excerpts: deterministically, with the library off "use 3 diagrams" made three calls and with the library on it made none and wrote a longer deck instead. So when the author names a number the call is required rather than merely offered. With no number named the choice stays the model's. And a model that has just made three tool calls tends to sign off instead of writing: "I'll create the presentation and the three teaching diagrams." was returned as the resource — 61 characters — because only a completely empty body counted as missing. A body with no title block and no heading is not a Pandoc document whatever its length, and if the continuation is no better than the first attempt, whichever actually reads like a resource is kept. Verified end to end with the library on: generate produced three figures and an eight-slide deck; modify added two more figures and a ninth slide. The figures were fetched and looked at — labelled airway anatomy, and a croup/epiglottitis/ bacterial tracheitis comparison. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01Dv6sqaY6Vq3ChZHMem3cnU	2026-09-11 19:19:22 +02:00
Daniel	055a86bbb1	feat: My Resources says what it is, offers its sources in one place, and Modify gets them too ... The screen had been trimmed to controls with no explanation of what it was for; "Private to you" said who could see it, not what it did. It now opens with a sentence that says what you get and how you get it out, and the header reads "Only you can see these". The four choices — clinical library, PubMed, web, illustration — are one "Draw on" group instead of four separate rows, with the library ticked by default. Each option hides itself when an administrator has not enabled it, so nothing appears that a person could tick and then be refused. Modify offers exactly the same choices. It had none, so "add what the 2024 trial showed" was answered from the model's memory rather than by looking anything up. Generate and Modify now go through one gatherSources(), so they cannot drift into offering different things or searching them differently. Writing "include a diagram of the airway" in the instructions now switches the illustration option on and says why, rather than the request being dropped in silence. Switching it off by hand sticks — the hint then reminds instead of fighting — and when no image model is configured it says so rather than pretending. Both the generate and modify boxes behave this way. Two things found by testing this rather than assuming it. PubMed ANDs every mapped term, so one unrecognised word takes the query to zero. "febrile seizures" returns six results and "febrile seizures in under-fives" returns none; "the anatomy of croup: subglottic narrowing and the steeple sign" returned none until it was narrowed to "anatomy croup", which returns six. A query that finds nothing is now retried against progressively shorter versions of itself, longest first, and the response says which query actually worked so the screen cannot report one that found nothing. Those extra calls tripped NCBI's three-a-second limit and produced a 429, so retries are spaced and the first attempt waits for nothing. Separately, the searches run on the topic while the library retrieval also gets the instruction: retrieval is semantic and benefits from the context, but a keyword engine handed a whole sentence returns nothing. And when a search was asked for and came back empty, the prompt now says not to invent a citation, a PMID or a URL to fill the gap. Without that the model supplies them from memory, and a fabricated PMID looks exactly like a real one. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01Dv6sqaY6Vq3ChZHMem3cnU	2026-09-11 18:54:17 +02:00
Daniel	05dcd1146d	test(e2e): repair the harness, taking the browser suite from 96 failures to 11 ... Three separate reasons tests were failing, none of them a defect in the app. The calculators. e2e-harness.html loaded calculators.js and drugs-loader.js with `defer` after they were split into ES modules; index.html was updated at the time and this page was not. A module parsed as a classic script throws "Cannot use import statement outside a module" before a line runs, so no click handler was ever attached: the pills rendered from static HTML and did nothing. Only the first calculator appeared to pass, because it carries `active` in the markup and needs no click. That was 52 failures. Settings and FAQ. Both moved from the tab rail into the account-card menu; the helper still clicked button.tab-btn[data-tab=…] and timed out. Ten more. The AI mocks, which had stopped intercepting for two independent reasons and so were calling the real model on every run — spending credits and comparing genuine output against strings like "MOCK HPI from dictation". A '**/api/x' glob matches no URL on Playwright 1.50, and page.route fails silently when nothing matches; measured against a real URL, that glob and '*/**/api/x' both matched zero times where a regex matched. Fixing that alone was not enough: the app registers a service worker that answers every /api/ request with its own fetch(), and a request made inside a service worker never reaches page.route. Blocking registration in the config puts them back in the page. The mocked dictation test now finishes in 1.6s rather than 7.5s, which is what a real model call costs. Whole suite: 204 passed / 96 failed in 15.8 minutes, now 289 passed / 11 failed in 6.8. The remaining eleven are spread across nine specs with no shared cause and are not touched here. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01Dv6sqaY6Vq3ChZHMem3cnU	2026-09-11 18:54:03 +02:00
Daniel	5577ec301c	refactor: cut the explanatory copy from the sign-in screen and My Resources ... The HIPAA notice is gone from the auth screen, along with the CSS rule that had no other user. The My Resources intro paragraph and the helper paragraphs under each checkbox are gone too; where a paragraph carried something the reader actually needed, it moved onto the control itself — the illustration checkbox now says what it does and where to steer it. Two sentences were kept rather than deleted, both about data leaving the building: that a web search query leaves this network and must not carry anything identifying, and that a PubMed query goes to NCBI. Trimmed to one line each. Say the word and they go too, but a clinical tool should not quietly send a query off-network with nothing on screen about it. The three tests that pinned the removed sentences now assert the behaviour instead: that the header still claims privacy, and that the illustration checkbox is genuinely unchecked in the markup rather than merely described as off by default. Verified as rendered: neither the sign-in nor the sign-up screen mentions HIPAA, the invite-code field is visible, and the generator card is down to its controls plus the NCBI line. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01Dv6sqaY6Vq3ChZHMem3cnU	2026-09-11 17:57:07 +02:00
Daniel	8cca5a4796	feat: instructions can ask for the illustration; library scrolls and searches; Modify ... Three things to My Resources. Illustration was entirely the model's call: someone who wanted a figure of something particular had no way to say so, because the instructions steered the prose and nothing else. The illustration guidance now tells the model to follow the author's instructions when they ask for a figure or name what it should show, and to treat that as the decision already made. Verified live: "include a diagram showing the age distribution and the simple-versus-complex distinction" produced exactly that, both halves in one figure. Exactly one image per generation is a real limit, not a wording choice — the shared imageTool dispatcher rejects more than one tool call per request, and it is used by the assistant and Learning Hub too. So the prompt says to draw the single most useful one if several are asked for, and the screen says the same. The library was an unbounded list that pushed everything below it off the page. It is now a 360px scrolling box with a search over title and topic, filtered locally because the rows are already in hand. "Nothing yet" and "nothing matches" are different messages, because telling someone whose search missed that they have never generated anything is wrong. Measured in a real render: 360px visible of 642px of content, and searching narrows 10 rows to 3. Modify is new UI over the refine endpoint, which existed with no way to reach it. Pick a resource, say what to change, and it is rewritten in place keeping its id, its downloads and its References section. The picker is built from the same library array, so it cannot drift, and a selection survives the refresh that follows a generation. Verified live: "add a Key Takeaways slide before References" inserted exactly that and left the other four slides alone. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01Dv6sqaY6Vq3ChZHMem3cnU	2026-09-11 17:49:27 +02:00
Daniel	79c329ceda	test(e2e): seed an admin account, and fix the sign-in that broke the browser suite ... Adds the admin fixture the Search Sources screen needed, and repairs the reason no browser-driving e2e test could log in at all. The sign-in failure first. The suite drove the app over http on a container hostname, which is not a secure context, so the browser provides no crypto.randomUUID. AccountBoundary calls it to mint a session generation on every sign-in; the call threw, the boot handler's catch swallowed it, and every test landed on the login screen holding a perfectly valid session. Measured: isSecureContext false and randomUUID undefined on http://pediatric-ai-scribe-e2e:3000, both true on http://127.0.0.1:3553, where boundary.enter() returns true and the app enters. Chrome's --unsafely-treat-insecure-origin-as-secure was tried first and does not work: Playwright rejects the --user-data-dir it must be paired with, and the flag alone leaves isSecureContext false. Loopback needs no flags, so the runner now uses the host network and the published port. The seed is new. The e2e user was a registration someone did by hand once that the shared Postgres happened to keep — enough to log in and no more. There was no admin account, so nothing under /api/admin could be tested through a real request, which is how the Search Sources card came to be verified by reading its markup. e2e/seed.js creates both accounts and reconciles an existing one, so a leftover with the wrong role cannot fail the suite for a reason unrelated to the code. It resets passwords and grants admin, so it refuses any address outside @ped-ai.test. The runner seeds before it tests. The new spec covers what markup-reading could not: that an ordinary account is refused the settings and never offered the Admin menu item, that no API key comes back readable, that the Test button reports each source separately, and that every control the save handler reads exists in a real render. Each account gets its own browser context, because AccountBoundary allows one owner per document and freezing the page on a second is the behaviour, not a bug. 10/10 pass on both projects. Two unit tests pin the loopback requirement and the seed's domain guard so neither can be undone quietly. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01Dv6sqaY6Vq3ChZHMem3cnU	2026-09-11 17:24:32 +02:00
Daniel	1270899dcb	feat: PubMed search for My Resources, and an image tool that actually fires ... PubMed joins web search as an optional source for a generated resource: a literature search on the topic, with abstracts, cited by PMID in References. Off by default, admin-enabled, with its own optional API key (NCBI raises the rate limit from 3/sec to 10/sec; it works without one). Neither search is a tool any more, and that is the point. Offering them as function calls meant the model decided whether to search, and with a prompt ending "Output ONLY Pandoc markdown" it decided not to — every time, with and without corpus grounding, no matter how the tool description was worded. Calling callAI with the tool directly produced a correct pubmed_search call, so the plumbing was never the problem. The search only ever needed the topic, and the route knows the topic before it calls the model, so both searches now run up front and their results go into the prompt as findings, exactly the way corpus excerpts do. Ticking the box now means the search happened. Verified live against deepseek-v4-flash: 30 corpus excerpts and 6 PubMed results, and a References slide carrying both the library sources and four real PMIDs (29562151, 38506440, 35721052, 28814254). Three fixes to illustration, which had never once fired: - The dispatch call had been lost in a refactor. The tool was still offered, the model still called it, and the call was dropped, so no job was ever enqueued. - imageContext was passed as a bare topic string where dispatch expects { request, history }, which made the bound request undefined. - The prompt never mentioned the tool existed while explicitly demanding only markdown — the same suppression that killed the searches. It now says an illustration is available and that calling it is not a violation of that rule. my_resources is its own image workflow rather than a reuse of learning_hub, because generated_image_links only accepts learning_hub assets, and that is exactly the barrier that keeps a private illustration out of published content. The illustration renders in the panel, rather than a toast pointing at an image history this feature does not have. Verified end to end: job queued, rendered, and the asset served to its owner as a correctly labelled subglottic-anatomy teaching diagram. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01Dv6sqaY6Vq3ChZHMem3cnU	2026-09-11 16:47:08 +02:00
Daniel	60556aae28	config: the clinical assistant answers from 12 excerpts ... It asked for 8 while the reranker capped everything at 12, so 8 is what it ever actually received — and its answers were judged good on that. Now that the cap defers to the caller, 12 is the number worth asking for: it is what the corpus has been tuned against, and the reranker still decides which 12. Verified live: the assistant returns 12 sources. Also records PubMed as a search source of its own rather than a provider option under web search. It returns structured records — title, journal, year, PMID, abstract — so a reference can be exact instead of reconstructed from a page title, and a model should be able to reach for "the literature" distinctly from "the web". Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01Dv6sqaY6Vq3ChZHMem3cnU	2026-09-11 16:13:55 +02:00
Daniel	571a013d29	feat: optional web search, admin-enabled and off by default ... The one feature here that sends text outside the building, so the defaults are the careful ones: disabled unless an administrator turns it on, opt-in per generation even then, and the option is hidden entirely rather than shown as something a user can tick and be refused. Only the search query leaves. Library excerpts, the generated resource and anything about the user never do. Both screens say so plainly, because a topic typed while drafting clinical material can carry clinical detail and the provider keeps its own logs. Four providers behind one shape, so swapping changes nothing downstream: Tavily, Serper over Google, Brave, and SearXNG — the only one where the query does not reach a commercial third party at all, which is why it is worth supporting even though it needs somewhere to run. The tool description says when NOT to search, because a model handed a search tool will reach for it constantly: not for settled clinical knowledge, which is what the indexed library is for, and one search per resource. That last one is enforced in the route with toolChoice: 'none' on the continuation rather than trusted to the model. A failed search never fails a generation — same contract as corpus retrieval. The resource is written without it and the response says what was searched for and what came back, so a query that left the network is visible rather than silent. The API key is masked on read and preserved when the field is left blank, the handling the OIDC client secret already gets, so changing provider cannot silently wipe a working key. Verified on the running instance: with nothing configured, webSearchAvailable is false, and a request asking for it anyway is ignored rather than honoured. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01Dv6sqaY6Vq3ChZHMem3cnU	2026-09-11 15:47:06 +02:00
Daniel	7eca509b02	fix: slides shrink to fit, and an article is never offered as slides ... A generated deck was losing content. pandoc writes a bare <a:bodyPr/> on every shape, which leaves the body with no autofit even though the slide master has one, so a slide with too much on it is cut off mid-sentence and the rest is not rendered at all. Reproduced and counted: eight bullets went in, three came out, the third ending mid-word. Every generated deck now carries <a:normAutofit/> on its body placeholders. No fontScale, deliberately — the renderer works out the reduction, so a slide that already fits is untouched, where a fixed scale would shrink all of them. The same eight bullets now fit with nothing in the bottom 6% of the slide. This is a floor, not a licence to overcrowd. The prompt still asks for one idea per slide; this stops a long one becoming unreadable. Also: an article is no longer offered as PowerPoint. A deck of paragraphs is not a presentation. Word and PDF suit either kind, and the route refuses the combination rather than relying on the button being absent. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01Dv6sqaY6Vq3ChZHMem3cnU	2026-09-11 15:34:16 +02:00
Daniel	efa84ac0e6	docs: one knob per feature, and record how many excerpts each gets ... Answering a fair question: needing two settings in two repositories to agree before one number changes is not a design anyone chose. It is two codebases each assuming it owned the decision, and the symptom was a caller asking for 30 excerpts and silently receiving 12. rerank_results computes min(reranker_top_k or limit, limit), so RERANKER_TOP_K=0 reads as "however many the caller asked for". The app setting is now the only knob. Verified: Learning asks 30 and receives 30, the assistant asks 8 and receives 8. Zero costs nothing extra — the reranker is billed on documents sent, which is candidate_limit and unchanged; top_n only decides how many come back. A real number there is now what it should always have been: an optional hard ceiling for when someone deliberately wants one, not an invisible default. docs/retrieval-tuning.md covers the per-feature budgets, why the assistant's are so much smaller than Learning's, that My Resources deliberately shares the Learning budget, how to read what actually happened from the MCP logs and the grounding field, and why raising these is not free. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01Dv6sqaY6Vq3ChZHMem3cnU	2026-09-11 15:28:01 +02:00
Daniel	fac8757ce8	feat: My Resources has a menu, a library and three downloads ... The pathway existed but was reachable only by API. It now has a tab of its own next to the Learning Hub — related, not the same thing, and sitting together is how someone discovers the difference — visible to every signed-in user with no role gate in the markup. Generate a deck or an article, see everything you have made, download each as PowerPoint, Word or PDF, delete what you no longer want. The screen says "Private to you" and "Nobody else sees these", because the distinction from published Learning content is the thing a person needs to understand before typing a patient's condition into it. Downloads are fetched rather than linked: an <a href> cannot carry the Authorization header. The blob is saved under the filename the server chose and the object URL is revoked afterwards. Resource titles come from a model, so rows are built as elements and a title is only ever assigned to textContent. The e2e stack now joins danvics_convert too. It could previously reach only Postgres and Redis, so a PDF download failed there in a way production would not — which did at least prove the degradation path works: with Gotenberg unreachable the response is "PDF conversion is unavailable right now. PowerPoint and Word still work", and the other two formats download unaffected. Verified in a browser as an ordinary user: the tab appears and opens, the form swaps slide count for word count when the format changes, the library lists their own work, and pptx, docx and pdf all download with sensible filenames (36360, 13285 and 68310 bytes). Also documents retrieval sizing in docs/retrieval-tuning.md — the per-feature budgets, and RERANKER_TOP_K, which caps all of them and had until now appeared in no configuration file at all. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01Dv6sqaY6Vq3ChZHMem3cnU	2026-09-11 15:00:19 +02:00
Daniel	a0d81789ff	feat: My Resources — anyone can generate teaching material, privately ... Learning is moderator-owned: content published into categories that everyone sees, behind router.use(moderatorMiddleware). That is right for institutional material and wrong as the only way in — an ordinary user could not generate anything at all. So this is a separate pathway rather than a loosening of that one. Learning is untouched; the moderator gate stays exactly where it was. A signed-in user can generate a deck or an article for their own use, keep it, refine it and export it, and nobody else ever sees it. Private by construction. Every statement filters on the owner and there is no route that returns another person's work, which a test asserts statement by statement rather than trusting. The foreign key cascades, so deleting an account takes its drafts with it. There is no category, no publish state and no sharing: adding sharing later should be a deliberate feature, not something that leaks out of a forgotten WHERE clause. Markdown is the artifact. Every format is rendered from it on demand — pptx and docx by pandoc, both carrying the house reference deck, and PDF by Gotenberg, whose LibreOffice preserves a deck's layout in a way rendering from markdown would not. That is what makes "add a slide on when to admit" a text edit rather than a binary patch. Gotenberg was published on the host but on a network of its own, so reaching it from a container went out and back through the host gateway. It now joins danvics_convert, owned by danvics-net like the others. PDF is the one export allowed to fail: if that service is down, the deck and the document still download and the error says which. Verified end to end as a plain user: the moderator route still refuses with 403, generation returned a deck grounded on 12 corpus excerpts, the library lists only their own, pptx/docx/pdf all downloaded valid, "add a slide on when to admit" put the slide in the right place and left References last, and an unauthenticated request gets 401 while someone else's id gets 404. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01Dv6sqaY6Vq3ChZHMem3cnU	2026-09-11 14:50:54 +02:00
Daniel	99a72734fb	feat: the Learning screen can ask for grounding, and says what it got ... The route accepted useCorpus and reported grounding, but nothing in the admin screen sent the flag or showed the result — so the feature existed and was unreachable. Opt-out in the UI rather than opt-in. For clinical teaching the library is nearly always the right source, so someone who never notices the checkbox should get the grounded version. The help text explains when to turn it OFF, which is the non-obvious case: a topic the library does not cover is better written without grounding than padded with the nearest unrelated excerpts. Afterwards it says what happened — "Written from 12 library excerpts", or "Not grounded — nothing indexed matched. Written from the model alone." Ungrounded material presented as grounded is the failure worth preventing here, so the wording never implies the library was used when it was not. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01Dv6sqaY6Vq3ChZHMem3cnU	2026-09-11 14:39:49 +02:00