Evidence Layer

AIEP’s evidence layer is the set of protocol specifications that govern how evidence is ingested, normalised, combined, checked for gaps, and recalled. Every claim admitted to an AIEP substrate has been through this pipeline.

The pipeline at a glance

Raw evidence
     │
     ▼
[Normalisation] P10, P17
Canonical form: UTF-8 + sorted keys + no whitespace → SHA-256
     │
     ▼
[Stitching] P25
Cross-source evidence assembly — links artefacts to claims
     │
     ▼
[Evidence Weight] P16, P37, P44
Temporal gap detection + divergence scoring
     │
     ▼
[Admissibility Gate] P02, P03, P04 → see /admissibility
     │
     ▼
[Recall] P22, P41, P42, P83
Archived branches surfaced when new evidence arrives

Evidence Normalisation — P10, P17

Evidence Normalisation →

Deterministic canonical form for any evidence artefact. The normalisation algorithm: UTF-8 encoding + lexicographically sorted keys + no whitespace + minimal number representation → SHA-256 hash. Two nodes that normalise an identical artefact produce an identical hash — regardless of machine, jurisdiction, or time.

P17 extends normalisation across jurisdictions: multi-jurisdiction normalisation profiles ensure that legal and regulatory artefacts from different jurisdictions resolve to a canonical form that any node can verify.

The aiep-normaliser repository is published under Apache 2.0 and available immediately.

Evidence Stitching — P25

Evidence Stitching →

Cross-source evidence assembly. Stitching links artefacts from multiple sources into a coherent evidence chain for a single claim. Each stitch is recorded as a machine-verifiable relationship — not an editorial judgment. The stitched chain can be replayed by any node from the committed artefacts.

Evidence Weight — P37

Evidence Weight →

Formal weight assigned to evidence in the reasoning graph. Evidence weight governs how the system arbitrates competing claims: a claim backed by a higher-weight evidence chain overrides a claim backed by a lower-weight chain, deterministically, with the weighting computation recorded in the ledger.

Divergence Detection and Temporal Gap — P16, P37, P44

Divergence →

Temporal gap detection (P16): The substrate records not only what evidence it has, but cryptographic proof of what it lacks. A time period with no evidence gets a committed NegativeProofRecord. Absence is proven, not merely asserted.

Claims divergence (P37): The DivergenceProof records where two evidence chains conflict. Conflict is not silently resolved — it is committed, signed, and surfaced.

Drift detection (P44): Long-running reasoning substrates can drift from initial evidence state. P44 detects accumulated drift and triggers re-arbitration when drift exceeds threshold.

The aiep-divergence-detector and aiep-divergence-proof repositories are published under Apache 2.0.

Recall — P22, P41, P42, P83

Recall →

The most important property of the AIEP evidence layer is not verification. It is recall.

Every knowledge system in history has discarded what did not fit the current consensus. AIEP preserves everything — not because everything is true, but because we do not yet know what will be true.

P22 governs deterministic multi-cycle recall state: a substrate can reconstruct any past evidence state from the ledger, artefact-by-artefact, and verify it produces the same hash as the committed record.

P83 governs recall divergence control: when a previously-archived branch becomes active again, the divergence between the original archival context and the current substrate state is computed, committed, and surfaced.

The aiep-recall-substrate repository is published under Apache 2.0.

Open-source repositories

Related

• Architecture — where the evidence layer fits in the seven-layer stack
• Admissibility & QA — the gate between evidence and execution
• GENOME & Swarm — the frozen kernel that enforces evidence binding
• Downloads — all specifications and repositories