Split the relevant courses into two tiers: ~25 courses described in detail
(paths, tags, about), and ~6 of those get their full video titles when the
question points at them specifically. This lets the advisor answer what a
course covers without sending the entire library's 3,000 video titles every
time. Video titles now become searchable terms weighted alongside tags and
learning paths, so courses are found by their actual content.
The advisor was a card inside the home page, so reaching it meant navigating
back to #/ and it was rebuilt on every render of that view. It is now a bubble
in the bottom-right corner of every screen bar the player and a quiz, opening
a chat panel that stays put as you move around — ask what to take next, follow
a recommendation, keep asking. It is built once into #overlay-root, outside
the view that gets replaced, and syncAdvisor() called from renderChrome() is
the single path that decides whether it is on screen; that covers every route
change and every loadMeta(), and makes the Settings toggle work with no change
to its handler.
Being permanent removes the reason for the routeSeq guards in ask(). An answer
that arrived after the user navigated was written to sessionStorage and then
dropped unrendered; now it lands, and the bubble carries an unread dot if the
panel is closed. showAuth() wipes the transcript and hides the dock, both of
which are new requirements: the dock is a sibling of #app-shell, so hiding the
shell no longer hides the advisor, and the widget now outlives a sign-out.
Courses named in the answer itself are links. advisorResolveCourses() is
lifted out of advisorValidateRecommendations() so an id written inline goes
through the same existence check and the same redirect to whichever duplicate
copy the learner has progress on — a sentence and the card under it can never
point at different copies. The map is keyed by the id as the model wrote it,
since that is what the browser has to find in the text. The client splits the
raw line before escaping rather than linkifying escaped output: esc() writes
an apostrophe as ', and the course-id pattern matches the #39 inside it.
Under 900px the panel is a full-height sheet, and the mobile toast strip lifts
clear of the bubble only while the bubble is actually on screen. The z-index
sits on the dock rather than #overlay-root, which would otherwise become a
stacking context and trap the confirm dialog behind the panel.
README's advisor section is brought back in line, including the parts left
stale by 05f0abc about sections and the whole library going out per question.
Claude-Session: https://claude.ai/code/session_01LPZRrwVj6L9ckoRtK1yHAN
Every question used to carry the full catalogue (~39k tokens on this
library, 41% of it per-section sizes). Now the model gets one line per
course as a stable, cacheable index, and paths/tags/blurb only for the
~25 courses a local term match against the conversation picks out (plus
anything in progress or already cited). Typical prompts land at 7-10k
tokens and the model reads the relevant courses in depth.
Duplicate copies of a course (same folder key) are collapsed before the
prompt is built; progress on a dropped copy counts for the kept one and a
recommendation is steered back to the copy the learner is part-way
through. Finished courses become a single id list. Reasons are capped at
15 words / 120 chars, which also shrinks every replayed follow-up turn.
Thinking mode is off for advisor calls (ADVISOR_REASONING = 'off').
Claude-Session: https://claude.ai/code/session_011gkaBpxtBdCBwdLL3R6Pfo
Course metadata: a scan now gives every course a course_meta row from the
downloader's sidecar JSON when the folder has one, else from catalog_courses
matched by the folder name the downloader would create (folderKeyFor, a port
of ine.py's sanitize_name). tools/build-catalog-seed.php turns the INE export
into seeds/ine_catalog.sql, an idempotent patch that creates its own table and
upserts by id, so it can be applied to the live server before or after the
code and re-run freely. Course pages show the blurb, difficulty, vendor and
tags; search matches tags and blurbs; Settings reports how each course got its
details. Against the live library 163 of 179 courses match.
Course advisor, off by default: api/advisor.php answers "which course should I
take to learn X / what comes before Y" with courses from the library, in
order, plus vendor courses not yet downloaded. includes/ai_client.php is a
small curl wrapper for any OpenAI-format chat endpoint (DeepSeek V4 Flash by
default, no composer); the prompt is a byte-stable catalogue of the library so
the provider's prefix cache covers it. Every course id the model returns is
validated server-side, a bad JSON reply gets one retry then a plain-text
fallback, and a per-user daily cap with token counts lives in
advisor_requests. With AI_ENABLED false the app makes no outbound requests
and the box does not appear; users can also hide it per browser.
Claude-Session: https://claude.ai/code/session_012enquL2RZdjy3D5AcKP5yx
Harwinder Sekhon