ProtocolSealed record
The precommitted discussion — written before the data, corrected after
Four full discussion branches sealed pre-dispatch; verdicts selected Branch 2; review then invalidated its headline.
Discussion branches (sealed)Branch selection recordScope-seal status
Before Phase 5 dispatched, four complete discussion texts were written and sealed — one for each combination of the registered verdict axes (corner mixture / interior persona / temperature grading), with both P5-2 variant paragraphs embedded. “The author does not get a vote after data exists.” The registered selection rule fired on axis B (at-least-one interior persona) and spliced Branch 2 verbatim.
External review then showed Branch 2’s headline — an unconfounded incentive-response existence result — was not prospectively established (no family control; the exact post-adjudication family is underpowered). The sealed text remains byte-identical in the record, quoted in §5.2 beside a four-row correction table, because its evidentiary value lies partly in making interpretive error visible.
The companion scope seal froze the experimental boundary: no new arms, no Phase 6 before publication, exit by registered adjudication + replay + anchored release. Its header still reads “PROPOSED — UNSEALED” because the file sealed byte-exact before dispatch — editing it now would destroy the chronology it proves; a living addendum explains the apparent mismatch instead.