The hand-packed icns was pixel-clean (verified by unpacking and scanning
every size), but macOS 26 composites legacy .icns icons into the Liquid
Glass system at render time; stacked on Icon Composer's baked PNG-export
lighting, that clips the rim highlights into the near-corner specks and
abrupt highlight ends (owner report 2026-10-13). The fix is the real
layered icon: mac.icon now points at the Icon Composer document
(build/icon.icon, copied from ~/Documents/Affinity/lh-logo) and
electron-builder compiles it via actool into Assets.car — macOS 26+
renders the true Liquid Glass design (dark/tinted variants included).
actool's backwards-compat icns caps at 256px, so scripts/after-pack.cjs
swaps in the existing full-res build/icon.icns before signing: pre-Tahoe
macOS keeps today's quality, Tahoe gets the layered icon.
Windows/Linux get the owner's flat design (rounded squircle with baked
transparency): scripts/make-icons.mjs renders build/icon-flat.svg via
resvg into build/icon.ico (16→256 PNG entries) + build/icon.png (512),
and electron-builder.yml pins win.icon/linux.icon.
desktop:dist injects DEVELOPER_DIR (actool needs Xcode; the machine
selects CommandLineTools by default, CI's setup-xcode takes precedence
via the :- default). The release workflow's mac job selects Xcode 26+
explicitly. Verified locally: mac --dir build (Assets.car + checksummed
full-res fallback + CFBundleIconName/IconFile) and the win-unpacked exe
carries our ico bytes; NSIS/AppImage assembly is CI's (cross-build
tooling can't run on arm64 macOS). Runbook updated. 1610/1610.
Titles become text-sm font-semibold in destructive red — JetBrains Mono
via the UI default, matching the abort-turn/dismiss-question dialogs the
owner named as the ideal look. Node and session names below the title
drop the sized-up text-lg for normal text-sm, still white mono, no
code-block chrome. Settings and Setup keep their serif titles by owner
ruling; abort-turn already renders this way. 1610/1610 both configs;
verified live on the running dev server.
The mount-time jump never fired on real graph-link arrivals: the
transcript scroller's ref is attached by a later sibling, and the queued
rAF was cancelled outright whenever use-stick-to-bottom re-created its
context while the initial scroll settled — every link landed at the
bottom (owner report 2026-10-13).
Rewrite the mount path as a synchronous 64ms retry watcher that keeps
running past its first success (hydration/adoption can replace the
jumped-to node), with a ~20s watchdog for cold-compile SSR streams
(observed 6.8s). jump() now reads the stick-to-bottom context through a
ref so effect identities stay stable, and the unmount cleanup no longer
strips a live flash. Verified live on real data: full load, SPA link
navigation, and in-session hashchange all land the anchor 16px from the
scroller top with the 2s highlight.
Also gives the node card's session/file rows matching horizontal and
vertical padding (px-2 py-1.5).
micromark-extension-math's inline closer is code-span logic — any
size-matched $ run closes, backslash-escaped dollars included — so
"$x = \$5$" closed at the escape, errored in KaTeX, and could cascade
into later text. KaTeX itself renders the escaped dollar fine, and its
own auto-render scanner skips escaped dollars, so the author's intent
is unambiguous.
Fix at the render seam (transformMathDelimiters): inside a balanced
inline span, an escaped dollar becomes the dollar-free
\text{\textdollar} before tokenization, with an escape-aware closer
(odd-backslash dollars never close; even-backslash row-break closers
keep closing). Display spans are untouched (a lone $ can't close $$ —
the escape already survives there), prose escaped dollars, code, and
unbalanced spans pass through, and spans without escaped dollars emit
byte-identical text — corpus-verified over 489 dollar-bearing fields
in 40 real sessions (35 changed, all escaped-dollar spans; the
economics session's dollar-amount math now renders).
The latex-testing session's Test-6 question renders 14 KaTeX spans
with zero errors (was 12 + 2 errors). 1578/1578 tests both configs;
typecheck, lint, build, and Electron-runtime smoke green.
- switch the math authoring/rendering contract from \(...\)/\[...\] to the
standard $...$ inline / $$...$$ display: prompts (20-question-tools,
40-formatting), singleDollarTextMath on the renderer, legacy reader kept,
and escape-debris recovery extended to dollar spans (mirroring micromark
delimiter-run rules)
- exact answer membership now precedes debris cleanup in
coerceCorrectAnswer (calculus incident: an exact sentence-final-period
label was stripped as model debris, then rejected); option keys are
passed at the boundary/grading seams, the dock no longer pre-coerces
- a complete, phase-correct JSON-escaping example rides the probe/quiz
tool descriptions (serialized from one canonical body; quiz leads with
type+nodeId), plus the one-sentence JSON-escaping rule in 40-formatting
- ask option descriptions render markdown/LaTeX in the pending dock;
settled cards keep the details line and option descriptions
retroactively, read from the persisted call arguments (chat passes
part.input to AnsweredCard)
1566/1566 tests on both navigator configs; typecheck, lint, production
build, standalone packaging, and Electron-runtime smoke all green.
Demos execute JS and need a browser; the app's own binary now provides
one: main.cjs boots a hidden window + CDP port under --browser-service
(before the single-instance lock - a helper is deliberately a second
instance of this binary), lifetime tied to the server's stdin (EOF =
quit, no orphans). headless.ts connects over CDP as PRIMARY (owner
ruling 2026-10-10), system Chrome stays the fallback, graceful refusal
when neither. Electron's CDP has no Target.createTarget, so the helper's
pre-created window is THE page, reused and reset per job behind a
serialized queue. PDFs and SVGs never reach a browser anymore (native
since stages 1-2); stale Chrome wording in model-facing tool copy made
engine-neutral. Live-proofed end-to-end from the dev tree (helper seen
by process watch, js-ran 42 probe, no orphan). Gates 1523/1523.
Visual-maker SVGs never execute script, so a browser added nothing but a
process. resvg rasterizes in-process now (white background and fitTo
width mirror the old Chrome-page semantics; svgViewport unchanged,
quirks included for parity). The old Chrome path stays verbatim as the
fallback for exotic-SVG parse failures - it was the only path before,
so nothing regresses.
The resvg binary (4MB per platform, one per build) rides the standalone
via the same tracing pattern. Gates 1519/1519; native path pinned
against real SVGs (intrinsic size, viewBox-only 900 default, opts
override, system-font text).
The browser was never the PDF engine - pdfjs always painted the pages;
Chrome only hosted the canvas tab. Now pdfjs-dist v6's own Node
integration (@napi-rs/canvas, its optionalDependency) hosts in-process:
same paint code, Skia canvas, zero browsers, zero temp files, no base64
hop. The whole pdf-reader runs browser-free - the Chrome gate on
pdf_render is gone.
The Skia binary (27MB per platform, exactly one per build - npm installs
only the building machine's) rides the standalone via the same
outputFileTracingIncludes pattern as pdfjs-dist. Verified: builds green
(1515), native render pinned against the real fixture (canonical
single-IDAT PNGs, dpi/72 arithmetic, page-range skipping), and the
canvas loads + paints from inside .next/standalone.
Stages 2-3 (resvg for SVG, the --browser-service demo helper) follow.
Exposed by the update drill: a packaged launch whose stdout consumer
exits (piped run, launcher) died on its next log line with an uncaught
EPIPE dialog. The stdout write is now guarded; the durable channel is
userData/logs/main.log. Stray main-process exceptions log and continue -
the window + child server are the product.
electron-updater in the shell (packaged builds only; dev never updates).
Checks at startup + every 24h, downloads silently, and the renderer shows
a Restart banner once the update is staged (it installs on quit either
way). Manual "Check for Updates..." under the app menu answers via dialog.
Every updater event routes through the shell log and is caught - the
updater can never take the app down (verified live: a feed failure logs
three lines and the app runs on).
LH_UPDATE_FEED overrides the baked GitHub feed with a generic local
server - the pre-release drill: serve latest-mac.yml + zip + blockmap
from localhost and a real installed copy updates itself, no release, no
public repo.
The asar allow-list re-includes electron-updater + its 16-package runtime
tree explicitly (gitignore last-match-wins; asar verified at 2.1 MB).
The banner rides the preload bridge (onUpdateStatus/installUpdate) and
renders nothing in the browser. window.lh typing centralized in
sidebar.tsx's declaration.
Chat records gain version (CURRENT_CHAT_VERSION), stamped by a single
writeChat choke point in the store; absence reads as v1 — existing files
migrate on their next write, no mass rewrite. The graph family (four JSONL
ledgers, sources.json, render-settings.json) shares one lazy sidecar,
graph/format.json; positions stays unstamped (rebuildable) and jobs stay
unstamped (transient).
Downgrade scenario ruled honest-refusal: records stamped newer than the
running build read as absent (chats, logged) or throw (graph family) — a
read-modify-write store never rewrites what it doesn't understand.
Golden fixtures stay the unstamped legacy shape on purpose (lazy v1 is
current-valid by design); the stamped write path is pinned by
tests/schema-stamps.test.ts. Reads never write — the golden byte-identical
pin holds.
Owner-manufactured teach master (chat_mv27rrra: compaction, subagent-abort,
every teach tool) + practice lifecycle (chat_mv293wv5: the full marker cycle
including the first-ever problem-dismissed), the real graph ledgers, five
jobs across the kind x status matrix, and settings. tests/golden-fixtures
reads all of it through the stores via the LH_DATA_DIR seam and pins the
fixture law: real data parses, flattens to a legal LLM wire, derives the
graph, covers every custom-marker kind, and reads never write.
Fixture law: an intentional format change updates these files in the same
commit; an unintentional one fails the suite.
THE START BUG (owner report, reproduced live both kinds): new sessions
(?prompt flow — practice and teach alike) sat dead until a second
message, which then persisted WITHOUT the first. Root cause, proven
both directions: the submit POST rode the view's consumeAbort
controller; React StrictMode's simulated remount fires Diet A's
unmount cleanup (model.stop()) ~8ms in and killed the POST mid-flight;
the page's sentInitial guard had already flipped, so it never retried.
Teach was equally broken — only unnoticed, because every teach chat in
use predates the diet and practice starts are always fresh. Fix: the
submit law — POSTs are not abortable by view teardown (they are ~100ms
acks since Diet B and cannot starve the pool); after the ack, a
consumeAbort identity check guards the stream attach. Pinned with the
stop-mid-submit regression test. Live: StrictMode on, practice draft
POST 200/67ms, stream attached, message persisted, problem posed.
PRACTICE PROMPT (00-base): the opening rule — material arrives two
ways, both first-class (picked propositions primary; free-form —
topics, course material, files/PDFs — fully supported, never refused);
as soon as material exists the reply poses, at most one short line
before the problem call, no warm-up interview. Plus the hard law: no
problem is ever conversation — every problem through the tool (no
thinking field/no reveal/no verdict attaches to prose).
GRAPH-CONTEXT PREAMBLE: custom markers flatten to user-role on the
wire, so the old "Contextualization — the learner picked…" read like a
system broadcast in the learner's voice channel. Now the bracket
convention: "[Session context] — picked from the learner's knowledge
map…".
SIDEBAR: busy rows unify to "Working…" for both kinds (the spinner's
default label included); "Practicing…"/"Teacher is working…" retired.
Gates: 1490/1490, typecheck, lint. doctrine's stale doc pointer fixed.
The README becomes the storefront for the public repo — what it is, why
it exists, BYOK/local/inspectable values, download, quick start —
written from the canonical articulation (owner doc, 2026-10-05): the
dependency structure of understanding, measure-first teaching, one
reasoning step at a time, the accumulating world model. All developer
content (commands, env, architecture, mechanisms, testing, lineage)
moves verbatim to docs/development.md (stale suite count and a private
machine path dropped on the way).
Owner will edit before the release; this is the base coat.
- LICENSE: AGPL-3.0-only (holder Amos Blomqvist), package.json gains "license".
- Product/package name unified: package.json "name" is now "learnharness"
(matches the GitHub repo); electron-builder productName/appId already
were LearnHarness/com.learnharness.LearnHarness.
- Deterministic userData: app.setName("LearnHarness") early in main.cjs
— Application Support dir is exactly "LearnHarness" in dev and
packaged runs (packaged Electron would otherwise derive it from the
package name). Dev shell becomes "LearnHarness-dev". Pre-release
rename: zero external users, no migration code; the local test dir
was moved by hand.
- docs/ diet: nine vision-*.md splinters, the completed pi-ai migration
plan, and research/ notes leave the tree (retained privately in the
git-ignored memory; full history stays in git). Public docs remain:
architecture, vision, plan, probe-teach-split, pi-engine-swap,
keyboard, teaching-methodology (+ release-0.3 until the ship).
Gates: 1489/1489, typecheck, lint.
Plan canvas (owner report: arrows clipped at the viewBox edge in dist —
measured 14px/20px cuts in the owner's real plans): blockLayout now
unions node boxes with every edge path's control-point hull (a bezier
never leaves it; the layout calls edgePathD with the renderer's exact
inputs). Dagre's node extent stays the floor, so ordinary plans render
byte-identical; the viewBox origin shifts only when ink bulges past
zero. Verified on both real chats; pins use the owner's real
linear-algebra tree (golden-real-data).
Keyboard-first (owner's law: Enter always submits, arrows always cycle;
text fields the only exception):
- Enter guard split (enter-submits): a focused BUTTON keeps native
Enter only when keyboard-walked (:focus-visible) or inside an open
overlay (menus/dialogs); pointer-clicked transcript buttons submit,
preventDefault suppressing the re-activation underneath.
- Focus hygiene (focus-hygiene): pointer clicks on transient controls
(agent/user thinking toggles, copy, mermaid expand, plan chip) blur
to body — the state clicking empty space produces.
- Session search owns the walk: Up/Down wrap-cycle the filtered rows,
Enter opens the walked row or the first match (quick-open), then
blurs and keeps the query. The Alt-arrow session cycle is RETIRED.
- Registry: model-menu walk/pick entries removed (obvious in-field
navigation, not settings material); question-dock keys stay
displayOnly per owner ruling.
Gates: 1489/1489, typecheck, lint. Verified live end-to-end.
Root cause (postmortem, verified live): Chromium caps 6 concurrent
HTTP/1.1 connections per origin. Every running turn pinned slots for
its whole life — the POST /api/chat response stream plus a reattach
GET stream per visit that never closed on navigate-away — so two
concurrent thinking sessions starved navigation RSC fetches in the
browser queue (147s measured) until the turns settled. Server was
exonerated (3ms curls during the freeze); main thread idle throughout.
- COMPACTION LAW + WIRE DIET (ui-stream): store the states, not the
chunks — replace-in-place snapshot retention keyed by updateSlots,
~100ms coalesced latest-wins fan-out, slim wire drops pi's duplicate
assistantMessageEvent payload.
- DIET B: POST /api/chat acks and closes ({ok:true}) — the turn wire
is GET /stream for everyone, submitter included; send() reattaches
via refetch-first resumeStream(true).
- DIET A: unmount cleanup calls model.stop() — navigating away aborts
the view connection (turns are server-side background objects).
- Answer-flicker fix: resumeStream's eager tail clear moved off the
refetch path (the refetch swaps base+tail atomically; clean-slate
guard restored on refetch failure) — the optimistic answer card
survives refetch latency (1.7s stalls measured on dev).
- Tests converted to the ack contract (drainTurn/drainRunning helpers,
dietFetch mock) + a pin that the POST body reads while the turn runs.
Gates: 1489/1489, typecheck, lint. Live: dual streaming sessions,
navigations 335-648ms, in-page settings 19ms (was 147s).
The 2026-10-08 flip, verified end-to-end (notarized build accepted,
stapled, spctl clean): Developer ID auto-detect, LH_NOTARY keychain
profile, zip + blockmap beside the DMG for Squirrel.Mac, publish block
pointing at GitHub Releases (repo goes public before 0.3.0; builds never
auto-upload — publishing is a release-time act).
The tools_fails_2 diagnosis (glm-5.3 dropping trailing required fields):
- probe-phase quiz/explain renamed to probe/recall — a shared name carried
the probe call shape into teaching (the first teach quiz dropped nodeId);
display seams accept the old names forever
- teach quiz+explain schemas lead with the required pair (type, nodeId);
explain gains the predict/check type and joins the reveal law (checks
collapse, predictions arrive open; no-type records render check-style)
- plan_init/probe_done results go context-slim (owner, context diet):
content = validated nodeIds + binding-id reminder / fixed ack; details
keeps the tree for the popup and {ok:true} — serializer-verified that
details never reaches the model; failure echoes dropped
- the cards carry the tool's own name (Probe/Recall) and the quiz type
reads on the label (Prediction/Check) in composer and transcript
The whitehead incident (2026-10-09): glm-5.3 emitted quiz type with
quotes baked in ('"predict"') — rejected ten times, byte-identical
retries, the lesson never started — and correctAnswer with the JSON
key-value colon glued on ([':parmenides']). Third GLM wobble incident,
and each prior heal was wired only where its incident landed; this one
landed precisely where the practice heal never reached.
Layer 1 — healEncodedArguments moves to core/argument-heal (the display
seam needs it too — the tool-parts law) with withArgumentHeal +
healToolRecord (agent/tools/heal) wired at EVERY registry: prober,
teacher, practice, bareQuestionTools (it had no prepareArguments at
all), demo-builder, and submit-agent (composing after
healStringifiedArrays; the auditor's extraTools wrapped).
Layer 2 — stripModelValueDebris (core/value-debris): leading/trailing
colons, quotes, backticks, whitespace strip at the MEMBERSHIP seams —
coerceCorrectAnswer (boundary check + display + grading in one),
normalizeQuizType (persisted wobbled records keep their role),
resolveTargetId (copy-syntax on targets).
Layer 3 — schema description hygiene: the quiz type and practice
mode/outcome/rung/missClass values now read as bare words — the quoted
JSON-key style invited quote-copying.
Display mirror — parseQuizInput heals the raw record on read, so a
boundary-healed quiz whose record keeps raw args renders instead of
becoming an invisible pending quiz (learner lockout).
Pins: tests/tool-heal.test.ts (the incident shapes end-to-end through
pi-ai validation, the wiring, the strips, the mirror). The stringy-
options pins flip to boundary-parity per tool-parts' own law; the
chat-flow garbage pin keeps its honest-reject intent with an unhealable
shape. 1471 tests, typecheck, lint green.
- The setup pickers opened invisible: React attaches a host element's ref
AFTER descendants' layout effects, so MenuShell's mount-time place()
measured a null anchor and gave up — the menu stayed hidden until a
window resize re-ran place(). The stage-2 slot and thinking menus now
anchor by anchorTestid (querySelector at place time, the settings-menu
pattern) instead of the conditionally-attached triggerRef.
- Saving models NEVER auto-sends the queued message (owner ruling,
2026-10-09): the draft waits in the composer and the learner sends it.
The replay/draft-send calls removed from onDone in new-session, home,
and chat; the dead force machinery removed from send/start/begin.
- The setup CTA pills get the app-wide hover language (hover:bg-accent +
cursor-pointer) — hover:bg-muted was invisible because --muted and
--card are the same #1c1c1c.
Empirically diagnosed via live browser measurement: Source Sans 3's
font box is 22px at 16px, but lh-input's 1.25 leading gives a 20px line
box - the font's content area sticks 1px below the line, Chromium
counts it in scrollHeight, and the zero-padding field-sizing textarea
turned it into a permanent 1px scroll: the bar showed EMPTY and at
every expansion size below max-h (owner report). The composer and dock
fields never showed it because their py-2 absorbs the surplus
implicitly. Fix: py-0.5 on the reading field - the same absorption,
line-height untouched (the caret law, typography round 2), and the
max-h-40 cap still scrolls honestly past 160px of text.
Lucide's Contrast leaves its half-moon unfilled (fill=none default) -
the owner's catch. Hand-rolled InvertDiscIcon: same geometry, the half
carries fill=currentColor, so the filled side is filled. The demo cards
get the identical disc + 'Invert colors' tooltip on their invert
toggles, matching the figures.
Owner round: the figure card loses its container entirely (no border,
no padding box) - the figure is document material. Image first, then
the CAPTION ROW below it: 'Figure' label aligned to the image's left
edge, the controls (invert toggle, download, fullscreen) on the same
row aligned to its right edge. The wrapper carries the natural-sizing
law, so for a declared-width svg the row hugs the centered figure's
own edges, not the column's.
The invert affordance: the house Toggle gains an optional icon disc -
the figure's toggle renders lucide's Contrast (the half-filled circle)
AS the sliding disc, plus a native 'Invert colors' tooltip; every
plain toggle elsewhere is unchanged. Interactive demo cards keep their
boxed stage (owner ruling: an interactive surface reads as its own
stage).
Owner verdict on the accent step: way too light - in-card thinking
drops the band entirely (rail only, identical heights). The rail
narrows again: 14px gap inside the 3ch inset (rail calc(3ch-14px),
about 8px) - the wide rail was distracting. Total inset still 3ch,
chevron law intact.
Owner round four on the user rows: (1) the rail thins - the gap inside
the 3ch inset grows 5px -> 8px, so the rail is calc(3ch-8px); total
inset still 3ch, chevron law intact. (2) The band padding drops again,
5px -> 4px (py-1). (3) The in-card thinking band steps to bg-accent -
the same ONE-STEP lift the chat band gets (background->card on the page,
card->accent on the card); card-color on card-color was invisible, the
accent gives the card rows a visible band with identical heights. (4)
pr-4 on the band rows - the expanded text wraps EARLY instead of hugging
the band's clear right edge.
Owner round three on the user rows: (1) the band padding halves,
py-2.5 -> py-[5px] — subtle, not tall; same for the messages. (2) The
in-card thinking joins the band: identical styling and heights for
collapsed thinking in the chat and inside question cards (card-on-card
is quiet by design; revert if it reads too light). (3) The left gap
MATCHES the new 5px padding, so the rail fills the rest of the 3ch
inset - border-l-[calc(3ch-5px)] pl-[5px]: a thick muted rail, total
inset still 3ch (the chevron-alignment law survives). The banded prop
dies - one geometry, cards included.
Owner round two on user-vs-agent clarity (the chat-comeback was tried
and set aside): we go all in on documents. No chat sides. User messages
and every transcript user-thinking row are full-width CARD-COLOR BANDS
- the band is the authorship mark against the agent's transparent
prose. No rounded corners, no borders besides the rail.
The rail keeps its muted color but is now 6px wide (was 2px): the
artificially wide 3ch inset (the chevron-alignment law with the agent's
flush rows) is re-split - rail eats the air, the rail-to-text gap drops
from 3ch-2px to a natural 3ch-6px, and the total inset stays 3ch so the
chevrons still align. The 'You' label stays on messages - one register
with the thinking labels.
In-card thinking (inside answered cards) takes the rail but NO band:
card-color on card-color would be invisible; geometry alone carries it.
Transcript rows arrive expanded (clearly sent) - the reading lift now
opens on arrival too; the in-card row stays collapsed. Cmd+T still
rules all rows.
Also this round: the answered-card last-child rhythm law
([&>*:last-child]:mb-0) - on ask cards the thinking row's my-3 stacked
with the card's p-4 into a 28px void below (12px above); the card's own
padding is the bottom rhythm when a row ends the card.
- desktop:dist cleans dist/ + .next/standalone before next build: the
trace sweep was shipping the PREVIOUS build's DMG inside every new one
(standalone/dist = 8.0GB inside the app; measured 7.4GB DMG → 144MB).
- the same sweep shipped the repo-root auth.json (dev BYOK keys,
byte-identical) in every DMG ever built — excluded in the trace and
REFUSED by a new build-server.mjs guard (.env* belt-and-braces).
- WARNING (next.config.ts): dir-style trace excludes (dist/** etc.,
"./"-anchored or not) strip server-external packages to bare
package.jsons — pi-ai/pi-agent-core lost their dist/, the packaged
server 500'd on ERR_MODULE_NOT_FOUND. Only the three verified-safe
entries stay.
- new scripts/smoke-server.mjs gates desktop:dist between payload
curation and packaging: boots the standalone exactly as the Electron
shell does (temp LH_DATA_DIR, empty auth) and requires 200s on / +
graph/models/settings. Caught both defects above on its first day.
- entitlements: allow-unsigned-executable-memory dropped (2019-era
relic per current electron-builder docs; weakens hardened runtime).
- pi-ai + pi-agent-core pinned exact 1.0.2 — persisted chat records
carry their wire shapes; post-release a bump is a migration event.
Gates: typecheck, lint, 1459/1459 tests, full desktop:dist + smoke PASS.
Blockquotes: the callout override had displaced Streamdown's own styled
blockquote component, so plain quotes rendered as bare unstyled text
since that commit. The pass-through branch now restores Streamdown's
verbatim classes (border-l-4, muted italic).
Explanations tucked for ALL questions (owner): the prediction's Why /
Hide-why toggle is universal — every quiz answered card sits its
explanation behind the same collapsed button. Same pattern, one law.
The thinking row's rhythm: my-1 -> my-3. The 4px margin hugged the
options on non-graded answered cards (owner report); 12px matches the
card's mt-2/mt-3 spacing grammar everywhere the row renders.
Side-based authorship (owner round, 2026-10-07): the document shape
blurred user into agent — three left-rail idioms (user text, user
thinking, quoted prose) collided once quotes rendered. User rows go
RIGHT: 'You' messages and practice thinking cards hug the right edge at
85% width with the rail mirrored (border-r), text still reading
left-to-right. In-card thinking (inside the answered card, the agent's
context) stays left. Transcript thinking rows arrive EXPANDED — the
thinking was clearly sent. Cmd+T (toggle-user-thinking, chat scope,
cheat-sheet row) sets every user-thinking row to one state via a module
store: explicit sets stomp individual toggles, late mounts honor the
last set (the state initializer reads the store, so static render
reflects it too).
Limits out (owner rulings): the 'Typically 5-15 nodes' cap and the
'keep the map small' pressure are dead — a plan is as elaborate as the
subject needs; what keeps it readable is correct structure (unconditional
truths as the hanging points), not compression. Labels stay short; size
is free. The method section now carries the goal reiteration: thorough,
deep, real understanding — sessions run as long as the understanding
needs, plans as elaborate as the subject requires; no schedule, nothing
to rush. (Code was already limit-free: no maxTurns, no parser caps.)
The adaptation + detour law, merged into one section ('The map moves
with him'): when the ground shifts mid-lesson — a missing foundation, a
deeper confusion — never rush the repair; no wall of text that 'covers'
the ground to get back on plan. Update the map if the structure changed,
then build the missing ground the only way anything is built here: one
reasoning step at a time, verified, at the session's one and only pace.
Detours are the teaching, not interruptions to get through.
The probe enters at the top (owner-observed failure: eigenvectors, then
high-school chain rule, then partials — strands opened at textbook
floors). New entry-point rules: open near the hardest thing the goal
needs and let a miss walk you down; strands inform each other (a strand
crushed at height says neighbors start high); the verdict + confidence +
thinking field TOGETHER set the next question's difficulty — confident
correct jumps sharply, hesitant correct steps gently. Efficiency is
resolution: never descend into already-bracketed territory; the
questions you save buy the specific, nuanced ones that locate the edge.
The one emphasis device for load-bearing ideas: GitHub-style alert
callouts. The prompt teaches only > [!IMPORTANT] (one device, one
meaning — an unconditional truth laid down, a payoff landed, a hinge
definition) with the restraint law: everything else stays in prose, if
everything is boxed nothing is. The renderer accepts all five GitHub
markers and boxes them identically — a stray [!NOTE] still meant 'this
deserves emphasis', and a marker line must never leak as text.
Rendering (owner rulings): SHARP corners, white stroke, no background,
no label, no icon — just the outline — and the box hugs its text
(width: fit-content capped at the column) instead of stretching
full-width into empty space. Same grammar as the plan blocks: this is
structure, not chrome.
Seam: the Streamdown blockquote override in md-blocks. parseAlertMarker
is pure and exported — Streamdown wraps blocks in its own styled
components, so detection reads the hast node's tagName, and a standalone
marker drops its <br> hard-break artifact along with the empty
paragraph. Plain blockquotes pass through untouched.
Rider (owner, same round): tables hug their contents — width:
max-content on the data-streamdown table (width:auto would still
stretch, the classic table trap); wide ones keep the md-blocks
overflow-x scroller. Graceful in exports: a callout is still a
blockquote there.
6 render pins: the box mounts with white stroke + fit-content + no
rounding, all five markers box, standalone-marker drops the empty
paragraph, multi-paragraph structure kept, plain quotes untouched.
Accuracy (shared, one law across every agent): the target shifts from
surface facts to understanding itself — 'the deeper risk isn't a wrong
name or date, it's a wrong understanding: an interpretation of an idea
that seems solid and isn't.' The bar is the doubt itself: checks are
cheap, and weighing whether a doubt is worth checking is how errors slip
through. Thinking stays the agent's own.
Research (teacher 40-jobs): floor framing — verify your understanding
before it becomes his; dispatch before the fact crosses your lips, teach
around the gap until the brief lands. Planning (10-planning): planning is
the natural moment for verification; the map is a hypothesis and the
probe only its first sketch — rewriting is a normal, common event (the
session adapts to what teaching reveals); and plan_init IS the
presentation — never draw the plan again in prose (the double-plan bug).
New shared segment 16-mental-model.md: the thinking fields are the
session's primary data — read them in full, measure what they reveal,
adapt everything on the model they build. Composes into teacher, prober,
and practice (shared-fragments list updated).
Practice: the rarely-needed framing is dead — fact checks get their own
law (a wrong model solution grades the learner against a lie; pose once
the brief lands), files and map readers become plain side tools.
LaTeX: the question-tools craft list gains the point-of-use bullet
('LaTeX everywhere — plain-text math in a question is a craft failure'),
the formatting law names EVERY field, and the practice problem statement
line carries it too.
Prompt pins updated: felt-doubt bar + mental-model doctrine in both
teacher and practice compositions, no-double-plan rule, rewrite-is-
normal law, rarely-needed framing asserted gone.
web_fetch_pdf downloads (25MB cap, probe-validated) into a per-job temp
dir and returns the triage summary — page count, size class (the pdf
doctrine's <=10/<=40 thresholds), flagged math/image-heavy pages. The
researcher's toolset then gains pdf_info / pdf_extract / pdf_search over
that dir, scoped by a resolve-and-startswith wall: the researcher's
filesystem ends at the temp dir. Render is deliberately absent — the
researcher slot has no vision; visual reading stays the pdf job's job.
web_fetch: a PDF URL now returns a POINTER to web_fetch_pdf instead of
dead-ending, and every output carries 'Source: <final url after
redirects> (HTTP <status>)' — the 'ill-formed request' lesson: a
200-served error body must be distinguishable from a real page.
The runner owns the temp lifecycle (mkdtemp before, recursive cleanup in
the job's finally — never enters the record) and web_fetch_pdf urls ride
the sources ledger. Doctrine step 5 added: fetch, triage, extract only
the pages that matter, cite as (p. N).
Tests build a REAL minimal PDF programmatically (correct xref offsets —
pdfjs parses it like any download): zero network, real parse.
URL.protocol carries a trailing colon ("https:", WHATWG); the colon-less
comparison (a direct port from the py harness, where urlparse().scheme
has none) rejected EVERY valid URL ever fetched — the tool never fetched
a page in its life. Owner-reported via OpenRouter logs: 8 valid URLs in,
8 'invalid URL' results out.
The new pin stubs globalThis.fetch so the happy path runs hermetically —
a valid https URL must REACH the fetch stage (and ftp:// still rejected).
dist rebuilt (desktop:dist) — bundle verified: "http:"!==t.protocol
- bridge measures the BODY (scrollHeight + rect + margins) instead of the
viewport-clamped documentElement — documentElement.scrollHeight latches
at the parent's own height, so demos could grow but never shrink on
window resize (owner-reported); smoke: 1036 → 2386 → 1036 px
- invert now PERSISTS as details.ui.inverted on the steer record, written
by the silent artifact-ui route (dismissal pattern: no wake, no SSE) and
seeded through the projection (SubagentArtifactData.ui); survives
navigation + reload, never enters teacher context or exports
- invert restyled as the house Toggle switch (settings controls) on both
cards; demo header is the caption grammar: 'INTERACTIVE DEMO — title'
left (title normal-case, same size/tracking), switch + download far
right; figure header: label left, switch/download/open far right
1439 tests green (route + pure-seam pins added); live smoke passed
amosblomqvist