README.md
KESPA-NOTE-012 — Trusted Research Release Promotion Pipeline
Date: 2026-09-14 Status: Verified — end-to-end release proof completed Project: KESPA AI / NexLabs Studios
Purpose
Before the later natural knowledge-gap automation was built, KESPA had already proven a separate research-release path for moving evidence-backed research into trusted knowledge.
The path separated research construction from production promotion:
evidence -> audit -> claims -> support verification -> release policy -> frozen release -> candidate -> trusted knowledge -> index -> NexLedger
Research-release stages
The historical pipeline contained distinct stages for:
1. safe evidence acquisition;
2. evidence-quality audit;
3. evidence-bound claim construction;
4. claim-support verification;
5. research-release policy;
6. trusted release packaging.
The resulting release could then cross into the application lifecycle as a research_release candidate.
That separation matters because research output was not allowed to become trusted merely because a model produced it.
End-to-end systemd proof
The verified proof used:
What is systemd?
Recorded release:
forge-release-b668d809f8b517bb46cdb92b
The release contained:
- 6 verified claims
- 1 evidence record
Application-side result:
- candidate: #80
- candidate source type:
research_release - trusted knowledge: #72
- status: active / public / indexed
The live Brain subsequently retrieved the new trusted card.
Importantly, the recorded live response did not rely exclusively on that one card; normal response composition still operated.
NexLedger proof
The promoted trusted record also completed NexLedger attestation:
- attestation: #73
- token: TK121
- asset: FORGEAI#TK121V1
- ledger height: 91
The existing generic NexLedger worker remained untouched.
Why this mattered
This proved that KESPA could maintain a clean boundary between:
research artifact
and:
production trusted knowledge
A research result had to survive explicit release gating before entering the authoritative trusted-knowledge store and local semantic index.
The later autonomous natural-gap lifecycle reused the same general research principle, but was intentionally implemented as an isolated path rather than by modifying these older shared scripts.
Evidence boundary
Search snippets were not accepted as evidence.
The evidence path depended on fetched/audited source material and explicit claim-support verification.
Limitation
This is one end-to-end promotion proof, not a population-level accuracy study.
The historical systemd release used one evidence record for six verified claims. Later natural-research work adopted stricter independent-source requirements; this note preserves the earlier pipeline as it actually operated rather than applying later policy retroactively.