Eighteen tajweed rules, two instruments for one displacement, and a register that knows the difference - #87
Merged
Merged
Conversation
…x belongs
The tajweed shards carried seven families because that is what the first
reader of the source collapsed them to. The source names eighteen rules, and
a reader who wants to colour madd differently from madd cannot say so through
a vocabulary that has already thrown the distinction away. So the shards are
rebuilt at rule granularity, core reads the wider vocabulary, and gate:tajweed
holds the two in step.
Beside that, the marks. probe-mark-ink measures whether the rectangle we draw
for a diacritic sits on the ink the print drew for it, and finds a per-page
displacement worth correcting. That correction is a number nobody has looked
at, and this commit adds the two instruments that let somebody look:
- a forced choice. Two rectangles, one panel each, nothing on screen saying
which is ours, and no key written to disk until the scorer rebuilds it
from the seed. It carries its own controls: obviously wrong rectangles to
catch a session worked while distracted, identical pairs to catch a person
who always picks something, and same-size displacements in other
directions to measure whether an eye can resolve a shift this small at
all. What it returns is a preference.
- a placing session. One rectangle, started a fixed distance out in a
direction that changes every trial, and the reader puts it where it
belongs. Subtracting what we proposed from where it landed leaves a
magnitude, reported against the reader's own precision, measured in the
same sitting from the marks shown twice. A residual smaller than the hand
that produced it is not a finding, and the scorer says so.
They are two sessions and not one page with a drag handle, for a reason worth
writing down: adjusting a rectangle inside a two-panel trial and then scoring
the landing against the two starting positions IS the answer key, and
adjusting after answering teaches the direction of the correction, which
points the same way on nearly every page.
Also here: make session, which opens a manual check as a sitting and writes a
transcript while the walkthrough happens rather than after it, and the
decisions register with its gate.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01EuhvbUKjGesE3uMhjCzBGt
Two rows where there was one. placement-correction-by-eye was carrying both halves of a question it can only answer half of: a forced choice returns one bit per trial, so it can confirm that our correction points the right way and is structurally incapable of confirming that it goes far enough. Every threshold in mark-registration.md's section 7 is a claim about distance. So placement-residual-by-hand joins it in the ledger, section 10 item 7 states the gap in the design doc, and the issue that was open on the first question is scoped rather than rewritten -- it now says what it can never deliver, so that nobody six months from now reads a preference as a distance. The two verdicts are scored separately and never summed, and the sessions should not be worked back to back by the same person. The other registers follow: four pointers in the code map, two rules in the extend skill about instruments and about measuring the hand in the same sitting as the thing the hand measured, and a transcripts README saying why a sitting is written down as it happens. Also here, from the same stretch of work: the decisions register and its rendered face, the tajweed-colours record and its options page, the mark granularity options page, mark-labels.md, and CLAUDE.md -- which says the thing all of this is in service of, that a decision is explained in plain language or it is not explained. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01EuhvbUKjGesE3uMhjCzBGt
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
What is here
Two commits, code then docs, over three threads that share a subject: what a mark is, and where it sits.
The tajweed shards widen from seven families to eighteen rules
The source names eighteen rules; the first pass through it collapsed them to seven families, and a reader who wants madd_2 coloured differently from madd_246 cannot say so through a vocabulary that already threw the distinction away. The shards are rebuilt at rule granularity,
@hifth/corereads the wider vocabulary, andgate:tajweedholds the two in step.The colour decision itself is open —
docs/decisions/tajweed-colours.md, with its options page — and nothing reader-facing ships until it is answered.Two instruments for one displacement
probe-mark-inkfinds a per-page displacement between the rectangle we draw for a diacritic and the ink the print drew for it. The correction is a number nobody had looked at. Two ways to look, and they are deliberately two:pnpm adjudicate:markspnpm nudge:marksNeither page knows the answer while it is being worked: the plan is a pure function of the seed, so no key exists on disk until the scorer rebuilds it afterwards. The forced choice carries its own controls (catch trials, identical pairs, same-magnitude decoys in other directions). The placing session measures the reader's own precision from marks shown twice, in the same sitting, and refuses to call a residual smaller than that hand a finding.
Why not one page with a drag handle. Adjusting a rectangle inside a two-panel trial and then scoring the landing against the two starting positions is the answer key; adjusting after answering teaches the direction of the correction, which points the same way on nearly every page. The two sessions are scored separately, never summed, and should not be worked back to back by the same person.
The registers learn that a preference is not a distance
A forced choice yields one bit per trial, so it can confirm direction and is structurally incapable of confirming distance — and every threshold in
mark-registration.md§⑦ is a claim about distance. So there are now two ledger checks, two issue rows, and a §⑩ ⑦ that states the gap. The older issue row is scoped rather than rewritten, so nobody later reads a preference as a distance.Also along for the ride, from the same stretch:
make session(a manual check opened as a sitting, with a transcript written while the walkthrough happens rather than after it), the decisions register and its gate, andCLAUDE.md.Verification
make cigreen (118.3 KB gz / 150 KB budget).pnpm --filter @hifth/etl exec vitest run scripts/lib/adjudication.test.mjs— 16/16, including that no trial can be answered by leaving the rectangle where it appeared, that the starting directions cancel, and that a repeated mark's two showings are far enough apart to be a fresh judgement.grepover the built page confirms the proposed correction reaches no markup, no data attribute and no comment.gate:validation,gate:issues,gate:map,gate:decisions,gate:tajweed,lint— all green.What this does not do
It does not bank a verdict. Both sessions are
owner: userand both blockmark-C; the hour is a person's and cannot be spent by an agent.🤖 Generated with Claude Code
https://claude.ai/code/session_01EuhvbUKjGesE3uMhjCzBGt