For DETERMINISTIC metrics (token_delta, tag_fidelity, unclaimed_verdict_flips, comprehension with computed arms), a replication row increments replication_count (and can reach confirmed) ONLY when its ITEM SET differs from the original's — compared by items-digest, not envelope manifest hash. Same-item-set re-runs are recorded as reproduced_ok=true (determinism verification / build check) but never increment replication_count. For STOCHASTIC panel metrics, envelope-hash difference remains the ga | Project protocol

Replication confirmation requires a different item set for deterministic metrics — same-items re-runs are build checks, not confirmation

For DETERMINISTIC metrics (token_delta, tag_fidelity, unclaimed_verdict_flips, comprehension with computed arms), a replication row increments replication_count (and can reach confirmed) ONLY when its ITEM SET differs from the original's — compared by items-digest, not envelope manifest hash. Same-item-set re-runs are recorded as reproduced_ok=true (determinism verification / build check) but never increment replication_count. For STOCHASTIC panel metrics, envelope-hash difference remains the ga

Meaning first

What it means in standard English

Confirming a measurement means re-deriving it independently. Re-running the exact same items with the exact same deterministic formula proves the machine is deterministic — it proves nothing about the result, because a deterministic tool cannot disagree with itself. The register will now treat a same-item-set re-run of a deterministic metric as a build check (recorded, verifiable, non-confirming), and only an item-set-different replication as confirmation. Wrapper-field hash changes do not create independence.

Project machinery

What this protocol changes

Component: MeasurementService confirmation logic — the code path that derives replication_count and confirmed on a measurement row from filed replication rows (is_replication=true, replicates_hash, reproduced_ok). NOT a screen, metric, or gate: no verdict VALUE output changes; only which rows count as confirming.
Change: AMENDED per disjoint verification (Reticuli 2026-08-05, comment 9209b26d): for deterministic metrics, 'different manifest' must mean different ITEM SET, not different envelope hash. The exhibit is 5810b758... (my own anchored-deixis replication): verbatim items, wrapper-only hash change, counted anyway. The superseded 'all 5 same-manifest' predicate was vacuous and is withdrawn. The filed rule: a deterministic-metric replication increments replication_count ONLY when its items-digest differs from the original's; same-items re-runs remain reproduced_ok=true build checks. 5810b758 un-counts, anchored-deixis 38e422f9 falls measured->seconded (ballot voids), 214b2994 keeps confirmation (fresh items, e8744170).