# Outside read (GPT-6 Astra via Codex, xhigh): the v2.4 clarification branch and Code It Yourself, 5 October 2026

Prompt: `READ_PROMPT.md`. Verbatim final message below; local file links shortened to repository paths.

Release: **ADOPT AMENDED**. Page: **HOLD**.

1. **High — the page can mislabel results.** At [line 123](site/lab/code-it-yourself.html:123), “the pre-registered label for it” promises too much. Starting from **Run 2’s coders**, change **Like touching and being touched → content**: the page gives **5.030, Supported**; the full analysis gives **4.964, Weakly supported**. Changing **Emotions → content** produces **Weakly supported** where the full analysis is **Inconclusive**. Replace with “an approximate application of the numerical decision rule”. Label computed badges **“Approximate rule output: [label] — not an H1 verdict”**; reserve unqualified recorded labels for exact recorded results.

2. **Medium — incomplete release disclosure.** [state.json:67](site/state.json:67) follows C4’s wording, but C4 itself omits part of the spec’s overarching requirement. Replace:
   > “The author fixed the pairing after the exploratory H1 analysis had been seen; the results stay exploratory and move no claim.”

   With:
   > “The author fixed the pairing after the exploratory H1 analysis had been seen, and under this pairing that analysis leans slightly toward the pre-registration’s prediction. The stated grounds are the Positions page’s readings and the traditions’ usage; the results remain exploratory and move no claim.”

   The changelog, research page and direction note disclose both timing and favourable direction plainly.

3. **Medium — relative difference presented as absolute orientation.** On the [page](site/lab/code-it-yourself.html:101), replace its opening question with:
   > “Do the sampled Zen and TM groups differ in their content-minus-awareness scores?”

   Replace “Positive means Zen leans toward content.” with:
   > “Positive means Zen’s mean content-minus-awareness score exceeds TM’s; both groups may still score higher on awareness.”

   Replace “points: Zen minus TM, on a scale from −100 to +100” with:
   > “points: Zen minus TM; each group’s orientation score ranges from −100 to +100.”

   The difference itself can range from −200 to +200.

4. **Medium — sampled codings are not validated interpretations.** At [line 303](site/lab/code-it-yourself.html:303), replace “the distribution is narrower than the space of defensible codings” with:
   > “The sampled universe retains known problematic codes and excludes alternatives nobody proposed. Its shares depend on treating candidate combinations uniformly; your selections can also fall outside that universe.”

   This preserves the all-codings record’s substantive caveat.

5. **Medium — restore the recoding qualification.** At [line 286](site/lab/code-it-yourself.html:286), replace “Two blind readers working from the paper’s own definitions landed at about 3.5 points, which the pre-registered rule calls Inconclusive.” with:
   > “Two blind readers used an unfiled instruction to prioritise the paper where it conflicted with the registered definitions; 63 items remained inherited. Both exploratory recodings gave about 3.5 points, labelled Inconclusive.”

6. **Low — provenance wording.** At [line 283](site/lab/code-it-yourself.html:283), replace “the preset buttons load recorded codings” with “Four presets reproduce recorded codings; ‘Lean awareness’ and ‘Lean neither’ are constructed examples.” Replace “none of the questionnaire’s wording” with “no full questionnaire items; abbreviated item labels are shown.” The payload contains aggregate statistics and coding metadata, **no individual responses**.

7. **Scope and consistency pass.** Apart from the separately proposed page, the diff stays within the specified clarification and version changes. No claim node, tier, confidence or falsifier moves. Part IV fits Part III. I found no surviving current assertion of the old pairing outside preserved historical records. The quoted outside-review verdicts are accurate; the page explicitly discloses the late pairing and that H1 remains untested.



## What would count against this

- One reader, one run.