Iteron v0.0.22 evidence-bound claim sheet

This page is the public one-page claim boundary. A number is included only when the repository contains the cited machine-readable or source evidence. Paths are relative to the repository root.

Current claim	Repository evidence	Scope
The workspace version is 0.0.22.	Cargo.toml workspace package version; matching Iteron package entries in Cargo.lock.	Source version, not proof that every release target is published.
The optimization census schema is version 4.	governance/optimization-census.json field schema_version.	Census format only.
The census contains 2,971 independent candidate rows: 2,084 runtime-settable, runtime-applied, and externally addressed; 0 unaddressed; 0 binding-required; 887 invariant/read-only.	governance/optimization-census.json summary fields.	Addressability census, not performance or quality.
The audited harness surface describes 28 modules and 66 services.	docs/development/deepseek-harness-gap-audit.md.	Audit inventory under that document's definitions; not universal trainer completion.
The research bridge protocol is iteron-research/1.	docs/reference/research-harness-protocol.md.	Protocol identifier, not evidence of a completed campaign.
Model adapters and model weights are structurally refused by manifest validation.	crates/evolve/src/lib.rs, PolicyManifest::validate, plus its reserved-artifact tests and xtask/src/conformance.rs.	Iteron admits harness artifacts only.

Explicit non-claims

Iteron does not claim production readiness, confidentiality isolation, complete microkernel conformance, autonomous live promotion, model training, universal harness optimization, parity with another coding agent, or a completed real research campaign. It makes no performance, latency, cost, quality, or benchmark-superiority claim. Counts above describe repository structure and contract coverage only; they are not empirical outcome measures.

Any future empirical claim must identify an immutable campaign manifest, dataset/task definition, baseline, model identity, evaluator, raw retained results, analysis code, uncertainty, and failure cases. Until those artifacts exist and are reviewable, result cells remain PENDING.