Fairway assurance package pilot — 2026-07-12

Decision

• Fairway Sovereign Deployment Ready starter: repeat-with-fixes after the product controls, disconnected rehearsal, customer evidence, and independent assessment exist.
• AI Cloud NIST SSDF starter: promote for bounded internal assessment preparation and evidence-gap discovery, not for an SSDF, legal, procurement, certification, compliance, or authorization claim.

Both signed packages were internally consistent and verified against a pinned Ed25519 public key. Both correctly returned control_sufficiency=insufficient, external_certification=not_evaluated, and ok=false.

Scope

The pilot used the released profile contract and the current local Fairway binary built from source after FW-360/FW-341. It did not import consumer source files or artifact bodies.

Package	Profile	Project and task-set scope
Fairway sovereign	fairway-sovereign-deployment-ready-v1-starter@v1	fairway-assurance-core-20260712: FW-341 and FW-355 through FW-360
AI Cloud	nist-ssdf-1.1-starter@v1	aicloud-node-security-20260712: healthy node-identity renewal priority, strict node-agent task signing, and Schemathesis promotion evidence tasks

The consumer config was read only. Package output was written under .fairway/artifacts/assurance-pilot-20260712/ in the Fairway repository, which is an ignored local evidence root.

Reproducibility identity

Package	Profile SHA-256	Manifest SHA-256	Files	Signature
Fairway sovereign	85ae2eebdf483d4b875fd71211245d517cfc94f72ab78d4e638551d281ebb95c	928497a6beb9ecdac8fe447e7ad2e3d0b0fea219b986565472825621dd250825	17	verified_pinned
AI Cloud	542af5f16789138e91199fc12a44b9e996555d11e9cb8592a86fed9070342212	e87f7b999a20efac8cc7bfc0a6bd45b03a0cc57c66fe3df8bb71383b643c6c80	17	verified_pinned

Both manifests use evaluation time 2026-07-12T21:52:00Z. The signing seed was generated for this disposable pilot, passed through an environment variable, and not written to argv, the packages, the assessment, or Fairway records.

Timing

Machine timings were captured with /usr/bin/time -p:

Operation	Fairway sovereign	AI Cloud
Export	0.30 s	0.02 s
Offline verify	0.01 s	0.01 s

Selecting the two bounded task sets, running the four commands, and producing the first machine-readable gap summary took about four operator minutes. That does not include framework applicability, legal analysis, sampling, testing, customer evidence collection, risk acceptance, or assessor judgment.

Results

Fairway sovereign starter

Status	Controls
satisfied_by_recorded_evidence	0
partial	1
missing	3
customer_responsibility	1
external_assessment_required	1

The package contains 84 metadata-only facts across task, review, ci, and evidence classes. It reports ten gap rows. The selected assurance-core tasks provide partial evidence for the offline-rehearsal objective but do not provide the required deployment configuration and decision, release provenance, offline rehearsal, customer identity/audit, backup/restore, or independent assessment proof. This is expected because FW-342 through FW-353 have not yet implemented or assessed those controls.

AI Cloud NIST SSDF starter

Status	Controls
satisfied_by_recorded_evidence	1 (PO.2)
partial	2 (PO.1, RV.1)
missing	1 (PS.3)

The package contains 13 metadata-only facts across task, review, ci, and evidence classes and reports six gap rows. PO.2 is supported only for the starter's narrow objective: the selected completed tasks have accountable task state and positive independent review facts. This does not assert that the consumer implements the complete SSDF practice or full framework.

The remaining gaps are an accepted decision record for PO.1, explicit release/provenance facts for PS.3, and vulnerability-class proof for RV.1.

Stale, conflict, and false-satisfaction review

• Neither readiness report contains a stale or conflicting control status.
• The evidence indexes preserve 56 Fairway and two AI Cloud facts as conflicting, primarily because historical changes and later approve review rows share one evidence class. That is conservative, but it obscures resolved latest-domain review state. FW-362 tracks deterministic latest review resolution while retaining prior rows as superseded history.
• No control was found satisfied by changes, partial, blocked, or fail. FW-360 regressions prove those non-positive results cannot satisfy shipped starter requirements.
• The sampled PO.2 support consists of done task facts and approve review facts. No failing or private artifact content was used.

FW-362 replayed the same Fairway task set after the pilot. The normalized index contained 33 current approval facts and 28 superseded review facts, with no review-history conflicts. A fresh AI Cloud replay after later independent changes verdicts correctly marked the review requirements conflicting and did not preserve the earlier PO.2 satisfaction. Its two blocked/pass evidence facts remained conflicting because they are genuine evidence history, not review supersession. The original signed pilot bundles remain immutable point-in-time evidence at 2026-07-12T21:52:00Z; this replay is regression evidence for the corrected current-state projection, not a rewrite of them.

Privacy and trust checks

• Evidence indexes contain no /Users/ paths, command text, kind-recovery artifact path, bearer value, password marker, prompt, transcript, raw tool body, source body, or artifact body.
• Package references are stable Fairway task/evidence/review identifiers.
• Integrity, control sufficiency, signature trust, and external certification remain separate verifier fields.
• The packages require a separately pinned public key before signature trust is verified_pinned.
• Verification is read only and wrote no finding back to either Fairway DB.

Manual assessor work remaining

Fairway materially shortened deterministic collection, indexing, digesting, signing, gap generation, and offline verification. It did not remove the need to:

• decide legal/framework applicability and the exact system/product boundary;
• review whether each starter mapping is suitable and complete for that scope;
• inspect authoritative source artifacts referenced by metadata;
• collect customer identity, network, key, backup, retention, and operating evidence;
• sample and test control operation;
• assess vulnerability and release provenance quality;
• resolve exceptions and residual risks through authorized owners;
• perform independent or accredited assessment and issue any external outcome.

Recommendation

Promote the assurance accelerator for internal readiness preparation because it turns selected Fairway facts into a signed, offline-verifiable, assessor-shaped packet in under one second of machine time and exposes missing proof without a false positive. Repeat the sovereign reference package only after the remaining controls are implemented and independently exercised. Complete FW-362 before using review-history conflict counts as a quality metric.

This pilot is evidence about package utility. It is not evidence that Fairway or AI Cloud is certified, compliant, authorized, FIPS validated, CUI authorized, EU CRA conformant, Common Criteria evaluated, or approved for any jurisdiction.