Context
Spun out of Stackbilt-dev/stackbilt-web#177 (route tarotscript's Old Farmer's Almanac codex through the Evidence Engine for attestation). That issue's "Path A" (curated public-domain library seed + existing pipeline) needs nothing here. This issue is Path B — the real authority/sourcing-discipline work, which lives in evidence-core because policy and requirements are owned here, not in stackbilt-web.
Problem
The Evidence Engine currently cannot enforce three things the almanac use case (and any tradition/folklore/historical-sourcing content) needs. All three are absent because:
- a domain is just a routing/label tag in stackbilt-web — it carries no policy;
- the policy lever is
EvidencePolicyVersion → EvidenceRequirements, a global set of 13 requirement types, none of which express sourcing provenance or claim modality;
- the EEAT score is a completeness metric — it counts whether required pillars are present, so a fabricated citation scores identically to a real one (citations are counted structurally, never verified).
Net effect: a "blessed when EEAT ≥ threshold AND receipt non-tampered" gate would launder fabricated citations rather than catch them.
Asks
- New requirement type — claim modality. Distinguish "the tradition holds X" (a sourceable historical claim) from "X is physically true" (a scientific claim the tradition does not license). Needs a new
EvidenceRequirement value + validator logic, not just a label.
- New policy preset for sourcing discipline. A preset (e.g.
historical_public_domain) that requires public-domain / primary-source provenance and rejects modern copyrighted source text as a primary citation.
- Citation-resolution tripwire. A validation pass that fails any in-content citation that does not resolve to a real evidence-library / research-cache asset. This is the actual anti-fabrication gate; today nothing resolves a citation string back to a stored asset.
Non-goals / boundaries
- No stackbilt-web changes required for Path A; once (1)–(3) land here, stackbilt-web wires the new preset/requirement through
/attest and surfaces the tripwire outcome.
- Threshold tuning + the
approved → blessed promotion gate stay tarotscript-side (tarotscript#167).
References
Stackbilt-dev/stackbilt-web#177 triage comment (full architectural analysis)
- AEGIS wiki:
almanac-codex-evidence-engine-attestation, evidence-engine-compose-and-trust-surface
- Honesty framing: receipts prove non-mutation, not truth — enforce authenticity upstream (here), not at the trust page.
Context
Spun out of
Stackbilt-dev/stackbilt-web#177(route tarotscript's Old Farmer's Almanac codex through the Evidence Engine for attestation). That issue's "Path A" (curated public-domain library seed + existing pipeline) needs nothing here. This issue is Path B — the real authority/sourcing-discipline work, which lives inevidence-corebecause policy and requirements are owned here, not in stackbilt-web.Problem
The Evidence Engine currently cannot enforce three things the almanac use case (and any tradition/folklore/historical-sourcing content) needs. All three are absent because:
EvidencePolicyVersion→EvidenceRequirements, a global set of 13 requirement types, none of which express sourcing provenance or claim modality;Net effect: a "blessed when EEAT ≥ threshold AND receipt non-tampered" gate would launder fabricated citations rather than catch them.
Asks
EvidenceRequirementvalue + validator logic, not just a label.historical_public_domain) that requires public-domain / primary-source provenance and rejects modern copyrighted source text as a primary citation.Non-goals / boundaries
/attestand surfaces the tripwire outcome.approved → blessedpromotion gate stay tarotscript-side (tarotscript#167).References
Stackbilt-dev/stackbilt-web#177triage comment (full architectural analysis)almanac-codex-evidence-engine-attestation,evidence-engine-compose-and-trust-surface