Measure World citation correctness - #8

Problem and scope

P2C3 proved structural citation identity coverage and P2C4 explicitly rejected broad promotion because coverage was not correctness. This packet adds the next bounded product-owned outcome: an independently recorded citation-correctness review over the exact P2C2 Brief and exact admitted public-record Observations.

It exact-loads the two Observation envelopes, derives the expected document/date statement from their canonical payloads, records exact reviewer/policy/claim/source provenance, links each Core Outcome to that exact review record, and runs two matched treatment/control pairs.

Product boundary and compatibility

World owns citation semantics, source facts, review policy, reviewer identity, negative-control construction, and score meaning. Core/Intelligence receive only a domain-neutral immutable observed-result coordinate and product-owned criterion. No World noun moves into Core.

This draft is stacked on codex/measured-impact-world-disposition / #7 and requires the stacked Core observed-result candidate. Released ace-domain-world-intelligence==0.9.0 and ace-core>=0.5.0,<0.6 metadata are unchanged; the built wheel remains inert JSON only.

Evidence

• Citation-preserving negative control retains the same two exact citation identities but swaps the two publication dates.
• Treatment/control citation coverage: 1.0/1.0.
• Independent correctness: 1.0/0.0 in both matched pairs.
• Result: useful, mean effect 1.0, non-effective/non-selectable promote proposal, historical replay without reauthorization.
• P2C3-P2C5 focused: 6 passed.
• Full World: 89 passed.
• Federal Register adapter: 26 passed.
• Release contract: 7 passed.
• Isolated built-wheel/public-Core boundary: 82 passed, 7 explicit candidate skips.
• Ruff and scoped format gates passed; sdist/inert wheel build passed.

Risk and rollback

The proof is bounded to one cited claim, two recorded official records, two matched pairs, and one frozen rule. It does not prove live freshness, source independence, correction handling, general Brief quality, causality, human benefit, reviewer infallibility, or proposal application.

Rollback removes the candidate harness/tests/state handoff/docs. Shipped Domain Pack JSON, connector, package versions, dependency range, and immutable host history remain unchanged.

Checklist

• I used no private data, credentials, proprietary fixtures, or private-extension code.
• I updated documentation for changed behavior and kept maturity at candidate.
• I ran focused/full World, connector, release-contract, Ruff, scoped format, build, inert-wheel, and git diff --check gates.
• The Core/Intelligence/Domain Pack boundary remains explicit.
• No release, merge, tag, publication, autonomous action, or applied governance change is claimed.