Repository path: workshop/experiments/E-20260810x-legend-again/README.md · rendered 2026-09-09

Page metadata (front matter)

type	`note`
id	`E-20260810x-legend-again`
status	`active`
created	`2026-08-10`
updated	`2026-08-10`
senses	`style-correspondence`, `voice`, `accuracy`
internal-judgment-only	`true`
provisional	`true`
links	`wiki/findings/results/RS-20260810x-legend-again.md`, `workshop/translations/szent-peter-esernyoje/R05-v1/translation.md`, `wiki/arms/ARM-legend.md`

E-20260810x-legend-again — there is no frozen design page here, and that is the deviation

continue-prompt.md §6 requires frozen written design → independent pre-run critic pass → run → post-run verification. This unit has no design page and had no pre-run critic. What it has instead:

The predictions were frozen inside the translation. D77 of workshop/translations/szent-peter-esernyoje/R05-v1/translation.md, committed at 7eeeb4d, states P-A, P-B and P-C and the direction of the self-similarity bias, before R04-v1 or R04-span2-v1 was opened. The decision frame decision_frame.py scores is the set of renderings named in D63–D76 of the same commit.
An independent post-hoc critic read the finished result page instead: critic.py, runs/critic-out.md — NEEDS-REDESIGN, 14 findings, 6 BLOCKING, all 14 accepted, none overruled. It is not a substitute and the result page does not report it as one (RS-20260810x-legend-again §12).
What the missing step cost is visible: P-C's registered band was built by applying the precedent's per-English-word site rate to a Hungarian word count, and failed for that reason alone. Note (blw).

Scripts

file	what it does
`diff_sites.py`	the divergence count under `E-20260806d-first-span-again` §4.3's site rule, verbatim; asserts the paragraph inventory before diffing; prints the agreement statistic note (bkc) requires
`sensitivity.py`	the site count at merge bridges 0–4, and with the one-to-two paragraph pair dropped
`decision_frame.py`	the 22 logged decisions of `D63`–`D76`, each with its test on the inherited rendering
`karoli_phrases.py`	the Károli frequency check at the five marked narration sites, with exact denominators
`critic.py`	the post-hoc critic call

karoli_phrases.py reads karoli-full.txt from the session scratchpad. The corpus is not stored in the repository (copyright hygiene aside, it is 4.9 MB of a public-domain text that is one fetch away); it is MEK-00161 fetched chapter by chapter from https://mek.oszk.hu/00100/00161/html/, and its defects — per-chapter navigation boilerplate, and an edition that is almost certainly the 1908 revision — are stated in RS-20260810x §11 and bound every figure in §8.