Evidence

Evidence register

Current proof items only — each with scope, source, what it proves, what it does not prove, and artifacts.

Open the Evidence Lab → Recorded demonstrations, the receipt wall, measured cost, and roadmap status in one place.

Agentic coding reliability — Case 001

EVIDENCE SNAPSHOT · 2026-08

Scope: One multi-day incident in Company New: internal green signals, customer-visible delivery failures, and the later repair record.
Source: Public app postmortem, dated receipts, immutable repository records, and an explicit owner-observation boundary.
Proves: A customer effect can fail while component tests pass; acceptance must include the signed-in request through an unauthenticated public artifact.
Does not prove: A universal ranking of Cursor, Claude Code, Codex, or any model; current end-to-end closure remains unproven.
Artifacts: Evidence lab · Claim-level JSON
Supersedes: The unpublished Case 001 prototype and its over-broad current-state claims.
Superseded by: —

FIRST-PARTY · 2026-07

Scope: One real governed run end to end: authorized goal, exact candidate digests, failed verification, bounded repair, independent pass, human promotion, receipt.
Source: Case Study #2 evidence bundle (run CS2-GISF-2026-07-14T005554Z)
Proves: The governed path operates on real work for its first operator, with the failure and repair preserved.
Does not prove: External adoption, revenue, universal correctness, or certification.
Artifacts: Walkthrough · JSON bundle
Supersedes: —
Superseded by: —

DEMONSTRATED · 2026-03

Scope: Internal first-party workflow — retain/wrap/replace with verification and bounded repair.
Source: First-party commissioning bundle
Proves: Failure capture, bounded repair, independent verification within infrastructure scope, receipt promotion.
Does not prove: External customer deployment, revenue, universal correctness, or certification.
Artifacts: Case study · JSON
Supersedes: —
Superseded by: —

FIRST-PARTY · 2026-07

Scope: Noetfield operating its own delivery, evidence, and deployment workflows on the Noetfield runtime.
Source: First-party operations under founder-operated alpha status.
Proves: The runtime performs real work under its stated controls for its first operator.
Does not prove: External adoption, customers, revenue, or certification.
Artifacts: Company-stage page
Supersedes: —
Superseded by: —

LIVE · 2026-07

Scope: Corporate www and public narrative under Noetfield Systems Inc. control.
Source: Domain canonicalization release NOETFIELD_DOMAIN_CANONICAL_V2
Proves: Current claim boundaries, retirement of stale enterprise and pricing narratives, and honest status labeling.
Does not prove: External adoption, ARR, SOC 2, or production deployments at third parties.
Artifacts: Case study · JSON
Supersedes: Parent-company self-audit (superseded)
Superseded by: —

LIVE · 2026-08

Scope: The release check that screens retired language on this site, across two consecutive releases.
Source: Release records for the control-testing page and the routes-and-language sweep, August 2026.
Proves: A check reported clean twice while the condition it screened for was present; the record shows both causes.
Does not prove: That current checks are complete, or that this failure mode is absent elsewhere.
Artifacts: Case study · JSON
Supersedes: —
Superseded by: —

LIVE · 2026-08

Scope: A research agent that screens a crypto watchlist once each morning, has a model draft a trade plan for anything that survives, and then checks that plan with arithmetic that does not consult the model.
Source: The agent's own cycle receipts, published in full including the runs where nothing qualified.
Proves: That a deterministic check runs after the model and rejects its output, on a schedule, in public, with the specific reason attached to each refusal.
Does not prove: That the gate is well calibrated. It has passed nothing so far, which is equally consistent with a correctly cautious system and one too strict to ever act. No performance is claimed and no position has ever been taken.
Artifacts: Research Trader
Supersedes: —
Superseded by: —