Three unit-1.1 lessons with gpt-oss-20b through LM Studio and the server —
the first real model on the reworked turn. The letter-level check went out
right, and the +25 clamp held: a reported 80 on the first answer was stored
as 25. What failed was how the model wrote its blocks, a different way each
session. All three transcripts are in test/fixtures/, verbatim, and each
failure below is a test against them.
Marks lost. The prompt asks for `여덟 | wrong | 여덜`. The first session wrote
`we | wrong | 우라 → 우리`, English prompt first; the second wrote no ::result
at all and marked only in prose, `✗ 나 | I (humble) → 저`. evidence.ts keys on
the first field of a ::result row, so nothing was ever recorded — no
evidence, no schedule, no confusions, and a 다지기 review that could never
close. The artifact would have lost them the same way. domain/marking.ts
attaches each mark to its word only where that is unambiguous: one Korean
word first, or through a prompt of the exercise he answered, read via that
exercise's ::words as the letter check reads it. With no ::result block the
✓/✗ lines are read on the same terms, so a mark can never name a word the
exercise did not ask for; a mark on a whole sentence is still dropped. What
he mistook a word for is taken from what he actually wrote whenever the mark
itself gives no other word — the third session put the right answer there.
Its third session, marked through all of this: 20 evidence rows, 20 cards.
Progress on requests. The prompt allows marks, ::confirmed and ::progress
only in reply to an answer. The model wrote ::progress on every message, and
three requests for a new exercise took the unit from 50% to 80% with nothing
answered. A reply to anything but an answer now changes none of them.
Feedback swallowed. The model closed no blocks, so lib read what followed
each one as rows: "your score is about 5%" became a result row the student
never saw, and a "---" became a recall item he was asked to write in 한글.
Another session fenced every block in ```. gloss.ts now decides every
block's extent from the raw text — at "::", the next block, a rule or fence
line, a blank line with no row after it, or for the piped blocks the first
line without a "|" — and hands lib the blocks properly closed. The gate
audit, now run through the parser the lesson uses, still flags 7 and 2.
Answers given away. Translate rows came with their meanings ("나 | I") and
recall hints were the answers ("two | 이"). A translate row keeps only its
Korean line, and a recall hint that is the expected word, or any word the
message declares, is dropped.
Also: the spelling a recall prompt expects now keeps its qualifiers. With 나
"I, me (casual)" and 저 "I, me (humble)" in one list, "I (humble)" matched
나: no letter check was sent, and the mark for 저 was filed under 나.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
README: the bundle's checks and what they measure, stable lemma ids and
curriculum words as cards, the one resolver and the audit it matches, the
turn's rules (earned progress, recall evidence, the 다지기 checklist, the
letter-level check, the prompt's cached prefix), the five-destination shell
with history, answer mode and word lookup, and sync's three rules. The known
limitation is the one lib/blocks.js still has.
server/README: the endpoints as they are — systemTail, the paged pull, the
compare-and-swap push, the protocol header and its 426 — and how protocol 2
works: hydrate first, a counter not a clock, the copy that holds more wins,
shrinking only ever declared. The Caddy setup is unchanged.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
읽기 연습. Each mode opens with what the round is and a Start button; the
round is timed, and a finished one reports accuracy and time, keeps the
best per mode and says when it was beaten. In sound mode the written
spelling is among the options — reading it as written is the habit the
drill exists to break, so it is the distractor that matters. Answers stay
marked 450ms when right, 900ms when a rule is shown, 1400ms when wrong.
The best rounds sync (trainer.drill), merged mode by mode: the higher
accuracy, then the faster time.
문법. An All chip, a count on every category, 반말 selected first — it is
what manhwa speech is made of — and each row names its category, so the
list still reads under All.
활용 연습. Words come in random order, not the pool's; the running score
comes back after a reload; a wrong answer shows what was typed beside what
it should have been, then the rule. The pool adds the roadmap's own verbs
and adjectives to the deck's. A form lib/conjugation.js cannot build no
longer leaves the Check button stuck.
문장. The artifact's lesson on the ending word and the 서술어 legend, a count
on each level, and every sentence's place in review — as far along as its
least-known chunk, since the chunks are the cards.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
단어. The search and the filters stay put above the list; the filters are
chips with counts — 내 진도 My units (the default), All, Due, New, Learning,
In review, Secure, then every topic. Below 840px the words are a list, 40
rows at a time as it scrolls, and a row opens its actions — Know it or
Reset, Delete for his own words — instead of three buttons in 360px. From
840px it is a table. The artifact chose between the two once, at first
render, so turning a tablet left the wrong one; this follows the width.
The add form sits behind ＋, and searching still reaches past the deck into
the dictionary — whose words can now be added from there.
오늘. One number and one action: what a review started now would hold (the
same number as 복습's badge), how long it should take, and Start. Below it a
line — streak, answered today, words secure — and the daily goal. The road,
the study log and the deck in numbers fold away underneath. The heatmap
shows as many weeks as fit, with weekdays, months and a legend.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
The reworked artifact made every Korean word on the lesson screen a way in
to the dictionary, in three tiers. This ports them.
1. The underline. Every Korean run in a message, a gloss or an exercise is
a button whose underline says what the word is to him: amber new, blue
learning, jade in review, faint when secure or merely explainable,
dotted when nothing knows it. A form carries its dictionary word's state
— 갔어 is 가다's — through the one resolver the gate uses
(domain/words.ts). The artifact gave "in review" no colour at all.
2. The popover, beside the word: the meaning, "+ Add to deck" (the
dictionary entry, so 먹었어 adds 먹다), "Ask 선생님" for a word he was
shown without being taught, a search, the word list. English on an
exercise's English side opens its Korean, from an index of the course's
own material — never the frequency bands, whose thousands of glosses
would bury the answer. An answer chip's tap belongs to the exercise, so
a 420ms hold glosses it, or a tap with 힌트 on; the hold's click is
swallowed before the exercise sees it. Pressing a word never takes focus
from the answer being typed, and the popover no longer closes when the
window resizes — on a phone that was the keyboard moving.
3. The word list gains + / ✓ on every row, search results included.
Every lookup, whichever tier, goes into the answer's "I had to look up"
and the peek tally — counted under the Korean looked at; the artifact
counted an English lookup under the English word.
A dictionary word added this way is studied: a card on an entry outside
the curated sources now counts as a deck word, in review and in "my
units". Before, it got a card that no review would ever show.
The word list glosses through the same resolver as the popover and the
gate, rather than the surface table alone.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
With a phone keyboard up there are about 350px left. Below 840px, focusing
an exercise's answer now strips the lesson back to what is being answered,
as the reworked artifact does: the nav, the header, the roadmap, the quick
replies and the earlier messages go, and an answer bar takes the composer's
place — ✕ · 2 / 4 · 가 · 한 · ↑ ↓ · 제출. Leaving the fields leaves the mode.
The 한글 keyboard follows the field touched last, message box or answer,
each with its own composer, so a half-built syllable stays in the field it
was typed in. The exercise picks the keyboard — up for recall, down for
translate, so one is not left up from the last — until he switches it
himself; that choice holds for the rest of the exercise. With it up, fields
ask for inputmode="none", and a focused field is refocused so the change
takes effect. 가 parks the field and opens the word list over the answer;
closing the list, or Back, returns to the field.
iOS ignores interactive-widget=resizes-content and lets the keyboard cover
the page; there the shell takes the visual viewport's height instead.
Also, from the plan's list:
· a send that gets nothing back takes the message out of the transcript
and puts a typed one back in the box; an exercise keeps its answers.
Stopped with nothing received, the message is withdrawn too. The answer
counts toward the unit once the tutor has it — counted before, a failed
and resent answer counted twice.
· Skip is local: the exercise steps aside and nothing is sent.
· a choice, tapped again, stays chosen.
· Enter moves to the next answer and submits from the last.
· an unreachable server says so, not "Failed to fetch".
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
The reworked artifact went mobile-first; this ports its shell, minus the
quirks it shipped with.
Routes. #lesson #today #settings #words #learn #sent #grammar #cj #hangul
#drill, each a history entry, so the phone's back gesture works. The word
sheet, a popover and the review screen are layers: each owns an entry while
open, and Back closes the topmost before it leaves a route — the artifact
read window.__onBack but never set it. shell/history.ts holds the rules,
free of React and tested against a history whose traversals land late, as a
browser's do. Unlike the artifact's router:
· re-tapping the current destination pushes nothing;
· leaving a route closes what is open on it, rewinding its entries first;
· a sub-page's ← goes back only when history leads to its parent, and
otherwise becomes the parent — history.length > 1 let ← leave the app.
Shell. The page never scrolls; each route owns one scroller, in svh. A
bottom bar under 600px, a 76px rail to 840px, a labelled rail beyond, with
the artifact's icons and a due badge on 복습. Routes mount on first visit and
stay mounted, so a draft or a drill survives a trip elsewhere. Settings,
the conjugation trainer and the reading drill are pages of their own; 학습
is a hub. The design tokens gain the layout set and a second register:
reference panels stay square, what a thumb works is rounded.
복습 is its own screen: tap anywhere to reveal, grades in the thumb zone,
and an empty queue says so instead of doing nothing. Its pool is "my
units" once there are twenty such words, as in the artifact — plus his own
words, which the artifact dropped from review at that point.
The lesson is a column that fits the screen: roadmap strip, conversation,
composer with the quick-reply chips, and a ··· menu for focus and clearing.
The word list is docked from 840px; below that it is a sheet with peek,
half and full detents and a drag handle, whose height comes off the shell
so the exercise above shrinks rather than being covered. Focusing an answer
drops it to peek; a new exercise takes it down from full.
Fixed on the way: the log now follows a reply by what the learner did, not
by distance — its own scroll event arrived after the stream had added more
than the threshold, which read as scrolling away and stopped the follow
mid-reply. Sending re-sticks. The keyboard's focus guard moves to
pointerdown; React 19 attaches touchstart passively, so preventDefault()
there was ignored.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
"The client enforces; the prompt only explains." Every rule the artifact's
tutor was merely asked to follow, it broke: it certified words on one
correct answer, scored a unit before anything was answered, used a word
from three phases ahead, answered in Korean, and invented spelling
diagnoses. The reworked app fixed each by making the client refuse. This
ports those refusals; domain/turn.ts holds the turn, testable without React.
The gate. A reply is scanned before he sees it — the side of the exercise
he must decode, through the one resolver, and its prose for Korean. A
refused draft is never stored, shown or applied: the tutor is asked again
and told exactly why. After two retries the reply is shown with its words
flagged, and the next turn names them. (The artifact's follow-up told the
tutor it could declare such a word in ::words; that contradicts the gate
and is left out.)
Marking. ::result feeds recall evidence per word. lib/srs.js is looser
than PORT.md, so the call site tightens it: one outcome per word per round,
and "learned" also needs five rounds between the first and last CORRECT
answer — lib alone counted a wrong answer as the start of the span. A
lookup is never recall. What he mistook a word for is kept. The schedule
takes at most one good grade a day from marking; in the artifact five good
rounds in one afternoon made a word "secure" by interval alone.
Phase reviews. The client holds the 다지기 checklist — each unit's rule and
every word the phase introduced, 132 items for Phase 1 — worked in batches
of ten. ::confirmed ticks a rule on the tutor's word but a word only on
evidence; "-item" puts one back; anything off the list is ignored.
Progress is earned: ignored until the unit has an answer, +25 at most per
message, a fall honoured in full, and the next unit only at 85% with three
answers — plus, in a review, nothing open. advanceUnit() enforces it too,
not only the banner.
The prompt gains a per-round tail after the shipped prompt — the practice
set (scored on the evidence, round-robin by word class, each word with the
words one letter away), the checklist, retry notes — sent as a second,
uncached system block so the stable prefix still caches.
Also: recall answers carry the letter-level jamo comparison (kept out of
his own bubble, since it is written to the model); match chips are keyed by
pair index, the bug PORT.md names; and the stand-in tutor exercises every
path offline — recall, ::result, ::confirmed, progress only after answers.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
The bundle's audit-gate.mjs measures the word gate against 54 real tutor
messages. Scored the same way, the port's gate was not the gate that was
measured:
port's allowed set 13 · 2 over-fires on words he had met
port's dictionary resolver 4 · 2 too weak: 마셔 → 마시다 again
lib/lexicon.js, DB behind it 7 · 2 the shipped gate
domain/resolver.ts makes lib/lexicon.js the one resolver for word taps and
the gate, built from the shipped data in the audit's order — roadmap words
first, so a scheduled form cannot inherit its stem's permission — with the
dictionary consulted only where lexicon.js has no route. That fallback is
what the port adds over the artifact: a word the curated data does not know
is still recognised as a real word.
The allowed set gains every word he has met (any card out of "new"), as
the shipped gate has it; the frequency band stays on top from Phase 2, as
PORT.md specifies. The words the tutor is TOLD it may use and the words it
is CHECKED against are one set, so it is never refused for a word it was
offered.
test/domain/gate-audit.test.ts reproduces the audit through the app's own
code — database, loader, resolver — and gets the shipped 149-word allowed
set and the same 7 and 2 messages, word for word.
Also: {{VARIETY}} and {{FOCUS}} take the artifact's wording and know the
fifth exercise type, recall; and the prompt's maintainer header, which
explains the placeholders, is no longer sent to the model.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
The reworked bundle's PORT.md makes row-level last-write-wins conditional
on three gates, each learned by losing real data. The port's sync broke
all three, and had four more ways to lose or stall work. Both ends change,
so this is one protocol version, refused by the other side if mismatched.
Gate 1, hydration. A device now pulls every page the server holds before
it may push anything; it used to push first. The 14 Sep laptop — a
week-old copy re-stamped at boot and pushed over a week of phone work —
is now a test, and the phone's week survives it. Boot writes nothing
syncable either: the lesson opens with the app's own words (the artifact's
seeded turn) and waits for Start, instead of stamping a reply and a
progress edit before a server can even be configured.
Gate 2, a counter. Every row remembers the change_seq it last agreed with
(base_seq). The server applies a write only if that still matches —
compare-and-swap under an advisory lock — and otherwise returns its copy
as a conflict. No clock is compared anywhere: a device an hour fast used
to win every conflict for an hour, and a slow one's newer edit was
silently dropped with HTTP 200. dirty and rev replace the timestamp
watermark, which lost edits whenever a clock moved backwards.
Gate 3, no silent shrinking. A conflict is settled by what each copy
holds (sync/resolve.ts): more reviews, more evidence, a finished unit, the
further roadmap position, the union of learned grammar. A deliberate
shrink is explicit: a reset or a cleared lesson raises a marker every
device obeys, including its own unsynced edits, so a reset is not undone
by a device that had not heard of it. Trimming the transcript is local
and tombstones nothing — it used to delete the other device's turns.
Also fixed on the way:
· keys travel as JSON arrays — a space in 몇 명 used to stop every
device's pull at that row, permanently;
· pulls take a shared lock against pushes, so a change_seq committed
out of order can no longer be skipped;
· study_log and peek are per device and summed, so two devices' reviews
of one day both count;
· each user's data has an epoch; a server that lost it is detected,
and the device re-hydrates and offers its data back;
· a protocol-1 client is refused with 426 rather than half-understood.
Migration 9 adds the columns, re-keys the counters and tombstones; the
server drops protocol-1 rows once (none were deployed). Verified: 14
two-device scenarios against a real Postgres, and two browser profiles
syncing a lesson through the UI.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Three storage changes the reworked app needs, as migration 8.
Progress. "Which unit is current" was a 'now' state on each unit's row. Two
devices that advanced could leave two of them, and leaving a finished unit
through the roadmap panel wrote it back to 'todo' — goToUnit un-finished
work. Where he is now lives in one place, meta road.unit; a unit's row
records only what is true of that unit: done, confidence, and room for the
answer count and the tutor's note that earned progress needs. The migration
carries the most recently written 'now' row across with its own stamp.
Chat. Turn ids were INTEGER PRIMARY KEY — max+1 on whichever device wrote
them, restarting at 1 after a clear — so two devices continuing a lesson
both wrote turn 201 and sync treated two different turns as one row. Ids
are UUIDv7 now, and the transcript is ordered by (created_at, id). Existing
turns become legacy:<device>:<n>, zero-padded so turns sharing a timestamp
keep the order they were written in; their tombstones are renamed with them.
The learner model gets its tables: evidence (lib/srs.js's record, plus the
rounds of the first and last CORRECT answer, which PORT.md measures and lib
does not), confusion, and phase_ledger for the 다지기 checklist, with a
confirmed flag so putting an item back is an edit rather than a delete.
A roadmap reset now clears the ledger and road.*; a full wipe also clears
evidence, confusions and learner.* — the artifact's wipe left its round
counter and confusion list behind. What a learner knows about words
survives a roadmap-only reset, as it should.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
The reworked app's learner model hangs off cards: recall evidence, the
phase-review checklist, the practice set, and the gate's "words he has
met". So every word a unit introduces has to be studiable — and 16 of the
371 were not. Eleven existed only as sentence chunks or dictionary rows
outside the review deck, and five (봐 읽어 갔어 봤어 먹었어) nowhere at all.
The build now marks exactly one reviewable lemma per roadmap word with the
unit that introduces it. Where the deck has the word, its row is chosen
deterministically (deck order, then source, then part of speech) — the
artifact tagged whichever card came last, which put the evidence for 이, 눈
and 저 on the wrong meaning. The sixteen get a `curriculum` lemma of their
own, glossed from the curated verb they conjugate (자 is "sleep", the 반말
of 자다 — not the dictionary's "ruler"), else from the sentence that uses
them, else the dictionary.
Lemmas also carry their topic, which the vocabulary filters need.
dict:assert gains the guarantee: 371/371 roadmap words as one card each.
Migration 7 adds the two columns; the rows arrive with the dictionary
reload a changed build now triggers on its own.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Cards point at lemmas by id, and an id was the entry's position in the
sorted build. One word added near the top of the dictionary would have
moved every card below it onto a different word — silently, because loaded
bands were recorded by number and a rebuilt dictionary never reached an
existing install anyway. A custom word took max(id)+1 on whichever device
added it, so the same id meant different words on a phone and a laptop.
An id is now lemmaId(headword, pos), a 53-bit hash defined once in
shared/ and used by the build, the loader, custom words and the migration.
The build asserts all 30,520 entries are collision-free, and a test pins the
function itself, since changing it re-keys every card.
Band files are format 2: they carry no ids at all. The loader derives each
id from the word, and a surface names its lemma by row index in the same
file. Writing hashed ids out cost 0.5 MB of incompressible digits; leaving
them out makes the files smaller than before (1.1 MB -> 1.0 MB).
The loaded dictionary is now versioned by its band hashes, so a rebuild
reloads on the next boot — safe only now that a reload cannot move a card.
Migration 6 re-keys an existing install without stamping anything: cards
and their tombstones move through the lemma rows still loaded, custom words
become custom_word rows (the learner's data, which can travel) carrying the
time their card was made, and the dictionary is dropped to reload.
Sync is paused until the protocol that replaces it lands: the server still
holds rows under the old ids, and exchanging them would plant cards that
name no word.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
HOUSE_STYLE was appended after tutor-system.md because the old prompt
never stated which language to teach in, and gpt-oss-20b delivered a
unit-1.1 lesson in Korean. The 16 Sep prompt states the rule outright,
twice — "everything you SAY is in English" — and the client now has
proseIsKorean() to send a Korean explanation back. The appendix said
itself that if the shipped prompt ever covered this, it should go.
{{VARIETY}} also learns the fifth exercise type, recall.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
The 16 Sep lib fixed both defects gloss.ts used to work around: a gloss
block closes at each "=" line, and every block is collected. So the
sentence splitting goes.
Three defects remain in how a block ENDS, measured on the new lib:
· deleting one block still deletes the "::" of the next, so a whole
::words block reaches the prose as "words" plus its rows;
· an unclosed gloss still swallows the following paragraph as parts;
· two gloss blocks written back to back share one terminator, and lib
never sees the second.
The old rebuild also only re-parsed the FIRST gloss block, which against
a lib that accumulates would have dropped every block after it.
One pass over the raw text now yields both the prose and each gloss
block's content, a gloss block ending where its rows end; lib parses each
block on its own, so the row format is still entirely lib's.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
The artifact was reworked after real incidents: a week of lost data, a
student taught out of order, and spelling diagnoses the model invented.
This takes the new export in verbatim; the port catches up in the
commits that follow.
Copied byte-identical from the bundle:
lib/ lexicon.js and sync.js are new; gate.js gains enforcement,
hangul.js letter-level marking, srs.js recall evidence,
conjugation.js deconjugate(); blocks.js now takes the last
block, closes gloss at "=", and parses recall, ::result and
::confirmed
data/ curriculum.json v5 — six 다지기 phase reviews; the 371
roadmap words are unchanged and no band moves
prompt/ English-only rule, recall, LETTER-LEVEL CHECK, marking
audit-gate.mjs, run-checks.sh, fixtures/ — the word gate measured
against 54 real tutor messages
CI runs run-checks.sh in place of validate.mjs alone, and `npm run check`
gains the audit. Baselines: validate PASS 0/0; audit 7 of 41 and 2 of 13.
types/lib/ declares the new API, and test/lib/ pins it: letterCheck on
the prompt's own 짧다/빫다 case, deconjugation, the roadmap-first order
that keeps 마셔 out of Phase 1, sync's three gates, and recall evidence —
including the two ways lib's evidence is looser than PORT.md, pinned as
they are so the call site that tightens them is visibly needed.
TaskHost gains a plain recall renderer so the tree typechecks against the
wider Task union.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Reported again: "::task translate" and its four lines showing as text above
the exercise they had been rendered into. The previous fix dropped
everything from the first line beginning "::", and could not see this one,
because by the time it ran the colons were gone:
**Example sentence**
*나 바다*
task translate <- the "::" removed, the items left as prose
나 바다
lib/blocks.js removes each block by string surgery, and a block's
terminator `(?:\n::|$)` is inside its own match, so removing one block
takes the two colons belonging to the NEXT one with it. Last time the
casualty was ::task before ::words; this reply put ::gloss first and the
casualty was ::task. Repairing that body cannot be made to work in general.
So the body is no longer repaired, it is derived: proseOf() reads the RAW
text, where the markers are always intact, and keeps every line outside a
block. The grammar is small -- a line of exactly "::" closes, any other
"::" line opens, a block runs until closed, until the next opens, or to the
end -- and nothing compares exact strings, because this model ends every
line with markdown's two trailing spaces.
An unclosed ::gloss is handled too, and every local model tested forgets
that closer. lib then reads the following paragraph as gloss parts, so two
sentences of English rendered as Korean example text inside the card. A
gloss row is `한글 | English | note` and a translation starts with "=";
anything else ends the block. The stranded prose goes back into the
message, and the gloss is now rebuilt unconditionally rather than only for
multi-sentence blocks, since the single-sentence case is exactly where this
bites.
Separately, MessageBody degrades gracefully on markdown instead of showing
it raw. The prompt forbids all of it and says so outright, but the backend
is pluggable now and a local model ignores the rule: a bulleted list of the
ten consonants arrived as lines starting with a hyphen, "*What we learn:*"
kept its asterisks, "---" showed as three dashes. Bullets, headings, rules
and *italic* now render quietly.
Verified by serving the exact reported reply through the real app: 10
bullets rendered as rows, no literal markup, no directive text, the gloss
card holding only 나 | 바다 = "I sea", and the exercise as real UI.
Also worth recording: three browser checks in this session were reading a
stub-generated turn synced down from Postgres, not the model. The app boots
on 수업, so the local stand-in writes the opening turn before a server can
be configured, and a fresh client then pulls the old transcript. Sending a
message after connecting is what actually exercises the remote path.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
On a reasoning model the reasoning is spent from the same completion budget
and it goes first, so a small ceiling truncates the actual reply away. It
fails silently and looks exactly like a model that cannot follow the system
prompt, which is the expensive way to debug it.
Measured on gpt-oss-20b against the real ~3.5k-token system prompt, same
prompt and same model, only the ceiling changing:
1400 an empty string, 0 bytes, after 27s
2048 a truncated half-Korean fragment, none of the required blocks
8000 99% English prose, no romanization, 0 batchim violations in the
task lines, all three blocks, 11s
The old comment justified 2048 by worrying about 4k-context local models.
That reasoning was wrong: the system prompt alone is ~3.5k tokens, so such
a model cannot run this app at all and there was nothing to protect.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
assemblePrompt used template.replace("{{GATE}}", ...), and String.replace
with a string argument substitutes only the FIRST occurrence. All three
placeholders appear twice in tutor-system.md, because the document names
them in its own header paragraph before using them:
Assembled per turn. `{{GATE}}` is `renderGate()` from `lib/gate.js`;
`{{VARIETY}}` and `{{FOCUS}}` are one-liners built from recent state.
So the rendered gate replaced the backticked mention mid-sentence, and the
real slot further down was sent to the model as the literal text "{{GATE}}".
The mechanism that decides what the tutor is allowed to teach was delivered
in the wrong place, with a template token standing where it belonged, and
the same for VARIETY and FOCUS. Confirmed by capturing what the app
actually put on the wire: three unfilled placeholders at lines 60, 86 and
141.
Substitution is now anchored to a whole line, which is what distinguishes a
slot from a mention -- the header's are inline and backticked. The
replacement is a function because renderGate() output contains "$"
sequences that String.replace would otherwise interpret.
The existing tests could not have caught this. They ran against a synthetic
template naming each placeholder exactly once, which is precisely the
property the shipped file lacks. The new ones run against
prompt/tutor-system.md itself: no slot may survive unfilled, the header
must come through intact, and the gate must land between the profile and
the pre-flight check. Both fail against the old code.
Also appends a HOUSE STYLE section after the shipped prompt -- an addition
by the app, not an edit to the file, which still ships byte-identical.
It covers two things the prompt leaves to inference. The language of
explanation is never actually stated: "한글 and English only" is a rule
about retiring romanization, and everything else only implies English. A
strong model infers it; gpt-oss-20b delivered a full grammar lesson in
Korean to a student on unit 1.1 who cannot yet read it. And no-markdown is
stated outright at line 96 and was ignored anyway, so it is restated where
the consequence is visible: the app renders **bold** and nothing else, so a
table arrives as rows of literal pipes.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Reported from the app: "::task translate" and its five sentences appeared
as text in the message, directly above the exercise those same lines had
been rendered into.
It is lib/blocks.js, not the model. parse() removes ::words by truncating
the body at its index, then removes ::task by substring:
if (w) body = body.slice(0, body.indexOf("::words"));
if (t) body = body.replace(t[0], "");
RE.task's terminator (?:\n::|$) is part of the match, so t[0] ends with the
"\n::" belonging to the ::words that follows -- the two colons the line
above just truncated away. The substring no longer occurs, replace() is a
no-op, and the whole task block stays in the body.
Order is the whole trigger. The stub tutor emits ::words before ::task and
is therefore fine; the local model emitted ::task first. Nothing in the
prompt requires either order, so this was always reachable -- Claude would
hit it too. It survived every test until a real model chose the other way.
lib/ ships unchanged, so the fix is at the call site, next to the ::gloss
workaround that is there for the same reason: parseMessage() drops the body
from the first surviving directive line on. Safe precisely because parse()
has already removed the blocks it handled correctly, so a "::" still in the
body is by definition one that leaked. That also subsumes the streaming
filter added earlier, which is now one rule instead of two.
A turn can also be nothing but blocks -- this model writes no prose around
an exercise at all -- which left an empty bubble above it. The bubble is
skipped when there is nothing to put in it, and the typing dots stay up
while the reply so far is only markup, since there is genuinely nothing to
read yet.
test/domain/block-leak.test.ts pins the lib behaviour as it is, and the
workaround against a verbatim capture of the model output that produced the
report.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
HANKAN_TUTOR_BACKEND=openai talks to anything serving
/chat/completions -- LM Studio, Ollama, llama.cpp, vLLM, LiteLLM,
OpenRouter, OpenAI. The TutorBackend seam already existed for this, so the
model becomes a config line rather than a code change.
Written against fetch rather than the openai package. The Anthropic SDK
alone is 14MB in the image, this backend uses one endpoint with no tools
and no retries, and local servers are the ones most likely to deviate from
an SDK's expectations. The real risk in hand-rolling it is SSE reassembly,
so that is where the tests are: a JSON payload split across two TCP reads,
an event whose blank-line terminator lands in the next read, heartbeat
comments, CRLF framing, and a stream that ends without [DONE]. The two
split cases both fail against a naive per-read parser, which is what makes
them worth having.
<think> blocks are stripped from the stream, tags split across chunks
included. Reasoning models served locally often emit chain-of-thought
inline in `content` rather than in a separate field, and left in it lands
in the lesson transcript where the block parser reads it as prose.
WHAT THIS COSTS: prompt caching. The Anthropic backend marks the ~12k
character gate as a cached prefix, so every turn after the first reads it
at a fraction of the input price. There is no portable equivalent, so
against a paid hosted endpoint the system prompt is re-billed every turn --
the biggest cost lever in the design, gone.
Against a local model it costs nothing, and the shape still pays: llama.cpp
and LM Studio reuse their KV cache for an unchanged prefix. Measured on a
6,948-character prompt against gpt-oss-20b, first token 1,563ms cold and
324ms warm, so the system prompt goes first and stays put here too.
Verified against LM Studio running openai/gpt-oss-20b, not only a fake: a
turn streams from the browser through this server to the model and back,
rendered in the chat, no page errors.
Also makes test/server/http.test.ts backend-agnostic. It asserted the echo
backend's wording and so failed the moment the server was pointed at a real
model -- precisely the case a transport test should survive.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
The Android build always talks to the Pi cross-origin -- a Capacitor
webview serves the app from http://localhost, not from your domain -- so
the browser sends a preflight OPTIONS first. It is not permitted to attach
an Authorization header to that. Auth ran before anything else, so the
preflight came back 401 and the real request was never attempted.
There were no Access-Control-Allow-* headers either, so even a successful
preflight would not have helped. Verified from an actual page before the
fix: GET /api/sync and POST /api/tutor both "Failed to fetch" -- an opaque
network error that points at the network rather than at middleware order.
CORS now runs first and answers OPTIONS itself. Any origin is allowed by
default, which is not a hole: the gate is a bearer token rather than a
cookie, so a hostile page gains nothing from being allowed to send a
request it cannot authenticate. HANKAN_ALLOWED_ORIGINS narrows it.
backends/echo.ts is a keyless backend that reflects the request back in
chunks. Deploying involves a container, a reverse proxy, a token, CORS and
an SSE stream that has to survive compression -- five things that break
independently, none of which involve Anthropic. HANKAN_TUTOR_BACKEND=echo
proves all five from the phone before a key exists and before anything is
billed. CI now runs the server that way, so the tutor endpoint is
exercised over real HTTP rather than only against an injected mock.
test/server/http.test.ts covers the preflight, the allow-origin header on
real responses, Vary: Origin, that a bad token is still refused, and that
the SSE stream parses and terminates with a done event.
Verified end to end in a browser: the Pi configured through the settings
panel, sync pushing 2 rows and a second sync moving 0 (the pushedAt
watermark holding), the header switching from "local stand-in" to
"connected", and a turn streaming back over SSE with the 8,859-character
system prompt intact.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
The previous pass measured with a 50ms timer and reported zero. A
MutationObserver plus the Layout Shift API disagreed: three log-height
shrinks per turn and two places where the reply's text went backwards.
Timer sampling had simply missed the frames.
Raw markup was flashing. A directive line arrives before the block it
opens is complete, and until then parse() has no reason to treat it as
anything but prose -- so "::task match" and "::words" rendered as visible
text for one frame and then vanished when the next line closed the block.
The streaming preview now drops :: lines outright, since blocks are not
rendered until the turn commits. It also trims the blank line that
preceded the directive, which MessageBody renders as a 9px gap: that gap
opened and closed on every block boundary.
The answered exercise no longer collapses. It was replaced by a one-line
"exercise answered" the instant Send was pressed -- around 170px out of
the log in a single frame, the largest jump in the view, and it threw away
what the learner had typed. It now stays rendered read-only, dimmed, with
the answers still legible.
Autoscroll never worked on a phone. Below 900px tutor.css sets the log's
max-height to none, so it has no internal scroll and setting scrollTop did
nothing -- a new reply just landed below the fold. It now scrolls whichever
element is actually the scroller. In the page-flow case the target is the
composer, not the bottom of the document: the word rail renders below the
chat, so following the document would scroll past the whole rail on every
turn. A reader who has scrolled up is still left alone.
Measured per turn, desktop: text regressions 2 -> 0, log-height shrinks
3 -> 0, DOM mutations 9 -> 5, CLS 0.0026 (only the word rail's own footer,
which is real content changing). Mobile: the composer stays in view and
the newest message is on screen; scrolled-up reader unmoved at y=200.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
TaskHost.tsx used a literal NUL as the delimiter in its drag-and-drop
payload, and tools/dict/build.mjs used one to join headword and part of
speech. Both work at runtime. Both also make the file *binary* to every
text tool: git shows "Bin 12259 bytes" instead of a diff, and grep prints
nothing at all for a match.
That is not hypothetical. Searching TaskHost.tsx for "<input" came back
empty three times while reviewing it, which is how its four exercise
inputs came to be reported as absent -- and why the accessibility defect
in them went unseen. Written as the escape \u0000 the value is identical
and the file stays text.
test/source-hygiene.test.ts fails on any control byte in a source file, so
this cannot come back quietly.
With the files readable again, the sweep the NUL had been hiding: eleven
form controls had no accessible name. The exercise blanks announced only
an ellipsis, and the part-of-speech select announced nothing. A
placeholder is not a label -- it disappears the moment you type. All
eleven now carry one, named after the thing they answer.
`npm run lint` gains --max-warnings 0. exhaustive-deps is configured as a
warning, so a hooks-dependency bug would have passed CI silently.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Three separate causes, all measurable with a MutationObserver over a turn.
The reply vanished and came back. On completion the code cleared
`streaming` and only then wrote the turn, trimmed the transcript and read
it back -- three awaits during which the message was unmounted, the typing
dots took its place, and the log jumped by the message's height. The write
now happens first and the placeholder is swapped for the committed turn in
one render.
The dots blinked again underneath the finished reply, because `busy`
stayed set until the finally block: one render with the turn committed and
the pending slot still mounted with `streaming` back to null. All three
state changes are now in the same batch.
The log shook while streaming. scrollIntoView() scrolls every scrollable
ancestor, so each token nudged the page as well as the log, and it fired
unconditionally, so scrolling up to re-read something was undone by the
next token. It now writes scrollTop on the log alone, once per animation
frame, and only when the reader is already within 80px of the bottom.
Also: the pending bubble is one keyed node rather than two sibling
conditionals, so waiting-to-streaming no longer remounts; and the
transcript is parsed once per change instead of on every render -- every
keystroke in the composer was re-parsing up to 26 messages of blocks.
Measured before: 1 bubble-count drop, 2 log-height shrinks, page scrolled
per token. After: 0, 0, and the page never moves.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
The text-entry rule was an allowlist of type= values, which fails twice.
An <input> with no type attribute is a text field but matches no
[type=...] selector at all, and any type nobody listed falls through
silently. Both had happened: the add-a-word fields carry no type, and the
server URL and token are url and password. All four rendered as white UA
boxes on a dark page.
A denylist of the controls that must keep native rendering — checkbox,
radio, range, colour, file, the button-like ones — cannot rot the same way.
color-scheme now sits on :root per theme rather than nowhere. It is what
tells the browser to paint the parts CSS cannot reach: the select dropdown
list, spinner arrows, scrollbars, the autofill wash.
Separately, .tiles draws its hairlines with its own background, so as a
stretched grid item it filled the rest of the hero's height with a solid
block of line colour. align-self: start.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
The seed pre-marked the artifact's 30 headwords secure on first run. Two
reasons it is gone rather than merely disabled:
It was never 30 cards. The match was on headword, and homographs each
carry their own lemma, so 그 as pronoun and as determiner both matched —
47 rows for a 30-word list.
Worse, "Reset everything" deleted the cards and then deleted the
'seed.known' guard along with the other meta keys, so the next boot
re-seeded and the deck looked untouched. The one thing a wipe exists for,
undone by the wipe itself.
Migration 5 clears the seed from installs that already have it. seedCard()
is the only writer that leaves updated_at = 0 on a card and the seed was
its only caller, so `DELETE FROM card WHERE updated_at = 0` removes exactly
the seeded rows and nothing the learner graded — the timestamp rule paying
for itself a second time.
test/domain/reset.test.ts pins both scopes: what a full wipe must leave
empty, what a roadmap reset must keep, that the deletions are tombstoned
so a sync cannot restore them, and that migration 5 really runs against a
schema-4 database rather than the test performing the delete itself.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
CI starts postgres:16-alpine and the sync server before npm test, so the
round-trip runs for real instead of skipping. A fake would not exercise
the change_seq trigger, the last-write-wins upsert or cursor paging, which
is exactly where sync goes wrong.
Also corrects the stale advisory count in the validate.mjs comment: the
updated export bundle carries data/gloss-extra.json, and the baseline has
been 0 blocking, 0 advisory since it landed.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
compose.yaml joins the Pi's existing network as external and reaches the
Postgres and Caddy already there by name; nothing is published to the
host. The schema applies itself on boot, so there is no migration step to
run by hand.
The Caddy snippet in the README is the part worth reading. Excluding
/api/tutor from `encode` matters more than flush_interval: compression
delays the header flush until body bytes arrive and holds already-flushed
events inside an unfinished frame, which presents as a stream that hangs
rather than as an error.
Untested against the actual Pi — written from PORT.md and verified only
as far as building the image and running it against a local Postgres.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
POST /api/tutor takes the assembled system prompt, the transcript the
client owns, and the new message, and streams tokens back. It holds
nothing between requests, so a dropped connection costs one turn rather
than the conversation.
The seam moved first: Sample took the assembled prompt as messages[0] with
role user. It now has an explicit system field, which is what lets the
backend put it in the API's system parameter as a cached block. The gate
is ~12k characters and is byte-identical for as long as the learner stays
in one unit, so every turn after the first reads the prefix at a fraction
of the input price. That is the single biggest cost lever in the design,
and it was unreachable through the old shape.
prompt/tutor-system.md still ships unchanged; only where the string is
placed changed.
SSE has three rules that are silent when broken, and all three are
handled: every event ends with a blank line, payloads are JSON-encoded
because a raw newline in Korean text would break the framing, and a `:`
heartbeat every 15s keeps intermediaries from timing the stream out.
Cache-Control is set on the returned Response rather than inside
streamSSE, which writes its own and would overwrite it; `no-transform` is
there because compression, not buffering, is what usually makes SSE look
like it hangs behind a proxy.
The client uses fetch + getReader, not EventSource — EventSource cannot
POST, and the body is {system, history, message}. Aborting closes the
connection, the server aborts upstream, and a cancelled turn stops
billing. With no server configured the app falls back to the stub, so the
offline build is untouched.
backends/anthropic.ts is the default. backends/agent-sdk.ts is deliberately
unimplemented and documents why the plain API was chosen over PORT.md's
Agent SDK — chiefly that its prompt accepts only user-role messages, so
the transcript would have to be flattened into a single turn.
The endpoint's own tests use a mock backend and need no API key.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Hono and pg, run under --experimental-strip-types, so the deployed thing
is the source. GET /api/sync?cursor=N pages rows above the cursor;
POST /api/sync upserts last-write-wins. Bearer token on everything under
/api; /health is open, for the container healthcheck.
Rows are stored generically — primary key as text, body as JSONB —
because the server never reads inside a row. It stores and orders them and
the client interprets them, which keeps the two schemas from having to
move in lockstep.
change_seq is bumped by a BEFORE UPDATE trigger rather than by the write
path. A row edited after a client last pulled would otherwise keep its old
sequence, sit below that client's cursor, and never be delivered; putting
it in the database means no future write path can forget.
The last-write-wins comparison is in the ON CONFLICT clause itself, so a
losing row is not written at all and does not bump change_seq — a
conflict does not become traffic for every other device.
test/sync/roundtrip.test.ts runs two clients against a real Postgres and
asserts what actually goes wrong in sync: that a fresh client's seeded rows
cannot overwrite the server's history (the artifact's bug, as an executable
test), that a delete propagates, and that dict.loadedBands never crosses
the wire. It skips without HANKAN_TEST_SERVER, so npm test still runs
anywhere.
POST /api/test/reset exists only when HANKAN_TEST_MODE=1, so it cannot be
reached on the Pi even if the token leaks.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Row-level last-write-wins on updated_at, cursor-based on a server-assigned
change_seq. The schema was built for this in step 1, so the work here is
the three things it did not yet have.
Tombstones (migration 4). Row-level sync cannot express a delete: with the
row gone there is nothing to compare timestamps against, so the other
device pushes its still-live copy back and the row silently returns. Every
delete path now writes a tombstone inside the same transaction.
The wire format lives in shared/sync-protocol.mjs and is imported by both
sides, so there is one definition rather than two that drift. It carries
the syncable-meta allowlist, which is the load-bearing part: meta mixes the
learner's preferences with bookkeeping that describes one install, and
replicating dict.loadedBands would tell a phone that had loaded bands 0-2
it holds every row the desktop has — the word rail would then fail to find
words it believes are present.
The sync loop pushes first, then pages the pull. Two details it would be
easy to get wrong, both commented at their site:
- The pull cursor advances only as rows are applied, never from the push
response. The server's newest change_seq includes rows this device has
not seen; adopting it skips them permanently, and nothing ever asks for
that range again.
- Pulled rows advance sync.pushedAt too, bounded by the instant the sync
started. Otherwise they look like local edits and get pushed straight
back, and an edit made during the sync is not swept up with them.
Seeded rows carry updated_at = 0, so a fresh device is never dirty and can
never win a conflict — the artifact's clobbering bug stays unrepresentable.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Six gaps the last review named, closed.
FOCUS SELECTOR. FOCUS_MODES and focusLine() already existed and already fed
{{FOCUS}}; nothing in the UI ever set prefs.focus, so it was permanently
"auto". Now a seven-mode picker in the chat header, with a compile-time
check that every listed mode exists in FOCUS_MODES — a typo would otherwise
render an empty {{FOCUS}} silently.
ADD YOUR OWN WORD. Custom words live in `lemma` beside the dictionary, with
ids from a reserved range starting at 10,000,000. The build assigns ids
sequentially from 1, so a custom word placed in that range would be
overwritten the next time the band files reloaded.
`lemma` is UNIQUE on (headword, pos) and the shipped dictionary is large, so
"add a word" collides with an existing entry regularly — 각성 already being
there is the normal case, not the exceptional one. Adding an existing word
now gives it a card and says so, rather than throwing an unhandled UNIQUE
violation into the console, which is what the first cut did. Its curated
gloss is kept; overwriting one from a text field would be a poor trade.
Only custom rows can be deleted outright.
GRAMMAR NOTES. Per-point textarea, saved on blur. Shares one JSON-in-meta
helper with the learned flags and the trainer score.
SEEDED KNOWN WORDS. The artifact's 30-word SEED_KNOWN list, applied once
after the bands load — they have to exist as lemmas to be matched. Applied
through seedCard(), so updated_at stays 0: it matches the artifact's own
stampInit() behaviour, and it keeps the seed invisible to sync when that
lands. 47 cards, because several headwords appear as both a curated word
and a sentence chunk, and he knows both.
FULL RESET. Two scopes, each spelled out before the second press. Neither
touches the dictionary — it is reference data, rebuildable from the assets,
and wiping it would leave the app unable to gloss anything.
ABOUT PANEL. lexicon.stats() was written and unused. It now reports what is
loaded here against what shipped, the storage driver, and the attribution —
which is a licence obligation, not decoration.
Verified in a browser: focus persists across reload, notes persist, a
colliding word is adopted, a new word round-trips through add and delete,
and 47 cards seed secure on first run.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
The PWA manifest declared icon-192.png, icon-512.png and favicon.svg. None
of the three existed, so install-to-home-screen and the browser tab both
resolved to nothing. That was a defect in what the last pass claimed as
done.
Drawn procedurally with Pillow rather than authored as SVG: there is no
dependable SVG rasteriser here to convert from, and ImageMagick falls back
to its MSVG renderer, which cannot be trusted with text. Pillow's
FreeType + RAQM stack shapes Hangul correctly, which is the only hard
requirement. The mark uses the app's own register — the serif face already
used for 한칸 on the boot screen, in --jade on --bg.
Two glyphs read well at 180px and up; at favicon sizes 한칸 turns to mush,
so the small sizes carry 한 alone.
icon-192 · icon-512 · apple-touch-icon (180) 한칸
icon-512-maskable 한칸 at 56%, inside
Android's circular crop
favicon.ico (16/32/48) · favicon-32 한
Sizing is a binary search on the drawn ink extent, and centring uses the ink
bounds rather than the font's line box — CJK metrics leave asymmetric space
above and below, which would otherwise sit the mark visibly high.
Re-running the generator writes byte-identical files.
Android launcher icons are untouched; Android work is out of scope.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
CI runs `npm ci`, which fails outright without package-lock.json — the
workflow could never have passed as committed. *.tsbuildinfo is incremental
typecheck state and does not belong in the tree.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
README covers how the pieces fit: the storage interface, the inverted
frequency join, the gate's three refinements, and the checks that guard
each one.
server/ holds a README only. Steps 1-3 give a fully working offline app with
no server at all, and that is the version that gets used first.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
npx cap add android, plus the SQLite and Keyboard plugins and
android:windowSoftInputMode="adjustResize" so the on-screen 한글 keyboard is
not covered by the system one.
npx cap sync copies dist/ — including the dictionary band files — into the
APK's assets, so the phone build is as offline as the web one.
Build output is gitignored; the project itself is committed. No APK has been
built: this machine has no Android SDK, so sqlite.native.ts has never
executed. test/db/conformance.ts is exported precisely so it can be pointed
at openNativeDb() on a device.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
React + Vite + TypeScript, PWA, offline-first. Six tabs: 수업 오늘 단어 문장
문법 한글, plus the full-screen SRS review overlay, the reading drill, the
conjugation trainer and the 두벌식 keyboard.
The visual language is carried over deliberately: two hand-tuned palettes,
three type stacks, about a dozen component classes, zero border-radius and
no icons anywhere — Korean glyphs do the work icons would.
THE GATE is the reason this app exists. buildGate() already took a
vocabQuery hook; filling it with a band query is what turns 371 hand-typed
words into something that scales. Three refinements sit inside that hook,
all of them narrowing:
1. words a not-yet-finished unit is the first to introduce are excluded,
so a frequency ceiling cannot smuggle 3.4's material into 2.1;
2. Phase 1 is filtered by the phonological ladder;
3. the list is capped at 800 by frequency, because renderGate() inlines
it into the prompt — strictly more restrictive than the band, so it
cannot leak.
prompt/tutor-system.md ships unchanged with {{GATE}} filled by renderGate().
Confidence is clamped per turn. The artifact wrote the model's ::progress
number straight into the sole gate on advancement, so one hallucinated 95
skipped a unit.
stub-tutor.ts stands in for the model on the artifact's exact contract —
onText receives cumulative text, an aborted turn keeps what it streamed —
so the real endpoint drops in without touching the UI. It rotates all four
task types and climbs progress gradually, which makes every render path
reachable with no server.
Two artifact bugs are not ported: task state lived in the full-page
re-render, so anything arriving mid-answer wiped typed text and placed
chips; and the day number was computed once at module load, so a session
left open overnight scheduled against yesterday.
Verified in a browser: all six tabs work, and after a hard reload with the
network cut every tab still works — including dictionary search out of OPFS.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Closes REVIEW.md §2. 167 of 371 roadmap words had no lexicon entry, so the
word rail silently showed nothing. Now:
Roadmap words: 371/371 resolve
Spiral targets: 35/35 resolve
Deck words: 386/386 resolve
and npm run dict:assert makes it a blocking build failure, not a silent
empty rail.
No runtime morphological analyser ships. lib/conjugation.js surfaceForms()
runs at BUILD time over every verb and adjective, so looking up a conjugated
form is an index hit on the surface table.
The frequency join had to be inverted. A subtitle frequency list holds
surface forms; a dictionary holds lemmas whose -다 citation form barely
occurs in running text, so joining on headword gives verbs a frequency of
roughly zero. Expanding each lemma into the forms it plausibly takes and
summing recovers 하다 from 118 to 89,041. Forms claimed by more than one
lemma are dropped rather than split, so homographs don't inherit each
other's mass. Those expansions score frequency only — the surface table
itself stays strictly surfaceForms() output plus the headword.
Bands are one per curriculum phase. Phase 1 admits no frequency band at
all: during the writing-system phase every word must be phonologically
legal for the unit reached, and a rank ceiling would hand the learner a
겹받침 during unit 1.4. shared/phonology.mjs lifts validate.mjs's own
feature ladder to enforce that; it agrees with the validator on all 371
words.
Sources are chosen automatically — KRDICT when vendored, otherwise the
kaikki.org extract. KRDICT's download is a JS form behind anti-bot
protection, so it cannot be fetched by CI; the derived band files are
committed instead, which the app needs offline regardless. Attribution and
the share-alike terms are in NOTICE.md.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Nothing above db/ knows which driver it got. sqlite.web.ts runs
@sqlite.org/sqlite-wasm in a dedicated Worker; sqlite.native.ts uses the
Capacitor plugin. Both apply the same migration array.
The web driver uses the OPFS SAHPool VFS rather than the plain opfs VFS:
the latter needs COOP/COEP cross-origin isolation headers, which neither a
static host nor the Capacitor webview reliably provides. Same storage,
fewer deployment constraints.
THE TIMESTAMP RULE. The artifact had a sync bug where a fresh device
stamped its own empty defaults as newer than the server's real history and
clobbered it. Three layers make that unrepresentable rather than merely
avoided:
1. updated_at INTEGER NOT NULL DEFAULT 0 on every syncable table, so
forgetting the column is the SAFE failure — a row that loses every
last-write-wins comparison, not one that wins them all.
2. writes.ts splits every mutation into seedX() (never stamps) and
editX() (always stamps), and is the only file allowed to read the clock.
3. An ESLint rule enforces that, and the conformance suite asserts a
freshly seeded database has no non-zero updated_at anywhere.
Two batching limits are sized for the platform we cannot test here: Android
links the system SQLite, historically capped at 999 bound parameters, while
sqlite-wasm allows 32766. A multi-row insert sized for the browser would
fail only on the phone, at first launch, loading the dictionary. Both the
band insert and the word-rail lookup batch under the smaller ceiling, and
test/db/limits.test.ts fails at 3600 if that regresses.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
lib/ ships unchanged, so its types live in types/lib/ and are wired up by a
tsconfig path mapping. Adding types costs nothing; reimplementing the logic
would cost the two things that make this port possible.
The tests exist so a later refactor cannot silently drift them:
hangul the nine Composer cases named in the export README —
먹어 · 왔어 · 읽어 · 괜찮아 · 값 · 의사 · 뭐야 and backspace
conjugation all seven irregular classes, and every form in
IRREGULAR_FORMS reachable through haeche()
srs the SM-2 transitions, ease and interval clamps
blocks parse → answerText round-trip for all four task types
101 tests.
One is a pinned defect rather than a guarantee. blocks.parse() never closes
a ::gloss block on its "=" line, so a multi-sentence gloss — which the tutor
prompt explicitly invites — collapses into one run-on line keeping only the
last translation. lib/ ships unchanged, so the test records the real
behaviour and the app works around it at the call site.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
The export bundle is the input to this port, not a sketch: the curriculum,
the tutor prompt and the five logic modules are finished and tested. They
land here byte-identical and stay that way.
diff -r export/data data && diff -r export/lib lib
diff -r export/prompt prompt && diff export/validate.mjs validate.mjs
data/, lib/, prompt/ and validate.mjs sit at the repo root so validate.mjs
runs verbatim with no path edits. All four are excluded from lint and
formatting — they are not ours to restyle. Types for lib/ live alongside in
types/ rather than as sibling .d.ts files, so the verbatim check stays a
plain directory diff.
CI runs the curriculum gate first, before anything else can pass:
node validate.mjs PASS — 0 blocking, 0 advisory
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
