EM-0027 execution plan

Source content

EM-0027 execution plan

Status: complete. Three independent source-readiness packets were exact-head reviewed and admitted

through protected PR #50 at merge commit cbb029bd336404e5bc654e939308247317eaa153.

Objective

Verify whether the GPT-4 bar-exam percentile, Mehrabian 7–38–55, and fact-check-effectiveness

candidates have bounded, credential-free, license-aware primary-source paths suitable for later

case research.

Authority and accepted base

• immutable task: tasks/contracts/EM-0027.json;
• accepted base: e9ad62b18f21594258643694c55709f78b4f9a50;
• owner and editorial direction: docs/editorial/how-we-know-case-roadmap.md;
• allowed changes: this plan, the research summary, three separate preliminary packets, and

append-only run receipts.

Method

1. Assign one candidate per context-isolated source scout.

2. Prefer primary research, official records, author or institutional manuscripts, registered

data/code, and authoritative examination statistics.

3. Record exact work and edition identity, URLs, identifiers, retrieval date, media type, access,

license or quotation treatment, and unresolved artifacts.

4. Keep circulated propositions separate from tested or reported propositions.

5. Map data, author, method, material, citation, comparison-class, and derivation dependence

without converting expected topology into an evidentiary count.

6. Search for corrections, replies, reanalyses, supersession, and material counterevidence.

7. End each packet with a provisional readiness disposition and bounded cost estimate.

8. Obtain fresh-clone independent review before protected merge.

Boundaries

• no complete literature review, pooled effect estimate, or scientific verdict;
• no dossier, feature manifest, catalog admission, lens change, or public claim;
• no credentialed source, account, secret, billable provider, or redistributed restricted text;
• no cross-candidate evidence count;
• unknown and inaccessible artifacts remain visible.

Validation

• compare every packet with EM-0027 acceptance and the retained editorial retrieval questions;
• verify identifiers, edition labels, and access/license language;
• verify all numerical statements have an authoritative locator or are clearly planning estimates;
• compare exact changed paths with task authority;
• run git diff --check and full make check;
• require independent source-identity and recommendation review on the exact head.

Draft-stage result

• GPT-4 bar-exam percentile: HOLD pending at-launch comparison-source and derivation closure.
• Mehrabian 7–38–55: HOLD pending primary-access, 7% derivation, and edition closure.
• Fact-check effectiveness: HOLD pending corrected-analysis artifacts, outcome boundaries, and

report-level overlap mapping.

The packets are preliminary retrieval maps only. They do not authorize a Case 002 choice,

evidence admission, a scientific verdict, catalog generation, or deployment. Before any protected

merge, an independent reviewer must verify every source identity, access/license statement,

lineage boundary, unresolved gap, and disposition on the exact candidate head.

Validation record

• Relative repository links resolve to the three separate packets and their index.
• The exact changed-path set is confined to the task's research/, docs/, and append-only

runs/** authority.

• git diff --cached --check passed.
• make check passed 80 tests, accepted-input validation, disclosure audit with no findings,

deterministic independent rebuild, and unchanged accepted source state.

• Run receipt: runs/proposals/20260823T160854Z-28cbe2de225f.json.
• Independent-review corrections preserved the Martínez 45th/48th discrepancy, added the exact

*Silent Messages* scan URL and unstable-carrier limitation, and pinned the 2025

Chan–Albarracín reply identity and dependence. Corrected-head validation receipt:

runs/proposals/20260823T164540Z-c67b9ff664a4.json.

• Second-round review corrections bound the current NBU scan bytes/digest, described the P2

anonymous session-dependent access flow, corrected Nathan Walter's name, recorded known

Ecker/Swire-Thompson author-program overlap with Case 001 while leaving item overlap unknown,

and pinned the Porter–Wood boundary review. Validation receipt:

runs/proposals/20260823T170317Z-e8eef36305f6.json.

• Third-round review correction records the confirmed Swire-Thompson et al. (2022) report-level

overlap between the Chan–Albarracín correction and Case 001 while leaving effect-size, item,

and participant-data overlap unresolved. Validation receipt:

runs/proposals/20260823T174049Z-f624b35ab550.json.

The validation environment reused an already installed project virtual environment while forcing

this checkout's src directory first on PYTHONPATH; it performed no dependency installation or

new provider retrieval.

Independent review and protected admission

• accepted base: 50af5bb295634bc17ec6ad0697e1806e6cee6ba9;
• independently reviewed head: ae5b2025a8400a1e12bd17ad6b70143b56d57d39;
• required Validate run/job: 32655948195 / 97234603952, conclusion success;
• protected squash merge: PR #50 at cbb029bd336404e5bc654e939308247317eaa153;
• merge tree: 5ef201c025b4a9f49e43251af569c0a951a18662;
• merge parent: exact accepted base 50af5bb295634bc17ec6ad0697e1806e6cee6ba9;
• source branch deleted after merge; no ruleset bypass was used.

Independent review retained all three dispositions as HOLD. Admission of this preliminary

mapping did not admit a dossier, establish a scientific verdict, or publish any of the three

candidates as a How We Know case.