Em 0027

Source content

{

"$schema": "https://epistemedia.com/schemas/task-contract-v1.json",

"id": "EM-0027",

"title": "Map primary-source readiness for the next How We Know library",

"status": "ready",

"change_class": "research",

"objective": "In parallel with Case 002 setup, independently verify the primary-source identity, availability, licensing, claim boundary, likely dependence structure, counterevidence seeds, and bounded corpus feasibility for the GPT-4 bar-exam percentile, Mehrabian 7-38-55, and fact-check-effectiveness candidate cases without producing or admitting a dossier.",

"depends_on": [

"EM-0025"

],

"authority": {

"allowed_paths": [

"research/how-we-know/library-preliminary/**",

"docs/execution-plans/EM-0027.md",

"docs/research/how-we-know-library-preliminary.md",

"runs/**"

],

"forbidden_paths": [

"constitution/**",

"policies/**",

"schemas/**",

"governance/events/**",

".github/**",

"tasks/contracts/**",

"catalog/**",

"src/**",

"tests/**",

"README.md",

"docs/product-direction.md",

"docs/editorial/**"

]

},

"required_evaluation": [

"primary or authoritative work, DOI, edition, and publication-identity verification for each candidate",

"credential-free access, open-access or quotation boundary, license, snapshot, and stable-retrieval assessment",

"circulated proposition versus tested or reported proposition comparison",

"known, shared, and unresolved data, method, author, material, citation, comparison-class, and derivation-lineage mapping",

"active counterevidence, qualification, correction, reply, and supersession seed search",

"bounded corpus-size and review-cost estimate with explicit uncertainty",

"fresh-clone independent review of source identities and every go, hold, or fail recommendation",

"full deterministic make check, disclosure audit, exact path-scope comparison, and clean tracked state"

],

"acceptance": [

"three separate preliminary packets cover GPT-4 bar-exam percentile, Mehrabian 7-38-55, and fact-check effectiveness without combining them into one dossier or one evidentiary count",

"each packet distinguishes verified source identity and access facts from unverified retrieval leads, interpretation, expected topology, and editorial recommendation",

"the GPT-4 packet checks the original model report, underlying score study, authoritative comparison-population data, the Martinez re-analysis, and the provenance of the top-10-percent derivation",

"the Mehrabian packet checks both 1967 primary studies, the cross-study combination step, the author's later public qualification, and a bounded sample of current recirculation while keeping the circulated and tested propositions separate",

"the fact-check packet checks the four-country experiment, broad evidence review, 2023 science-misinformation meta-analysis, 2025 reply, durability and behavior boundaries, and source-program overlap with Case 001",

"each packet records exact URLs, identifiers, retrieval dates, media types, access status, license or quotation treatment, and unresolved artifacts without inventing missing values",

"each packet ends with a provisional go, hold, or fail recommendation tied to source closure, corpus size, risk, and machinery value rather than perceived importance",

"reported numbers from prior model-generated notes receive no evidentiary credit until independently read from an authoritative source",

"the work retains negative searches, inaccessible sources, conflicting identities, and unresolved derivations",

"no candidate is described as researched, accepted, reviewed, settled, published, or ready to feature merely because preliminary mapping succeeded",

"the result remains preliminary research and does not alter the public catalog, Case 001, application code, policies, workflows, or generated outputs",

"make check passes without changing accepted source state"

],

"limitations": [

"Only credential-free public reads are authorized; do not create accounts, accept provider terms, use secrets, purchase access, or initiate billable research services.",

"This task records source readiness and candidate topology, not complete literature reviews, effect-size estimates, verdicts, or admission receipts.",

"Copyrighted works may be represented by metadata, access receipts, digests where available, and quote-minimal fair-use spans; do not commit unauthorized full text.",

"The three candidates may be researched in parallel, but their source artifacts, conclusions, lineages, and recommendations must remain independently reviewable.",

"No agent trace pilot, dossier construction, feature selection, policy change, compiler change, workflow mutation, deployment, publication, DNS action, package release, or spend is authorized."

]

}