railiance-clock/specs/acceptance-v0.1.md
tegwick c29bf39b5d Verify reproducible release artifacts and map acceptance evidence
Assistant: codex
Assistant-Model: gpt-6-astra
Assistant-Session: 01a0e6f0-a6f6-7b93-9124-55ef584930e2
2026-09-28 10:22:52 +02:00

3.3 KiB

Acceptance and threat model — draft v0.1

IDs below are behavioral acceptance obligations, not a count of implementation coverage. The current coverage map identifies local test evidence, gaps and the existing owner records for native acceptance.

ID Scenario Required evidence
C01 Corporate Windows/WSL wall clock ±5 minutes, wrong zone and wall-clock step Process-local interval remains correct under supported bootstrap; OS settings unchanged
C02 Asymmetric delay and deliberately delayed authentic response True test time stays within computed interval or sample is refused; no symmetry assumption
C03 Wrong key, nonce, authority, version, signature or modified uncertainty Refusal before any consumer validity check
C04 Replay, duplicate nonce, response after timeout/restart Old response cannot establish a fresh anchor
C05 Authority upstream outage, large dispersion, restart or UTC rollback Explicit unusable/holdover policy; no unearned healthy state or rollback acceptance
C06 Client reboot, suspend/resume, monotonic discontinuity and virtual clock drift Old mapping invalidated; no frozen-clock lifetime extension
C07 Expiry/not-before boundary overlaps uncertainty Refresh/wait/refuse; entirely valid interval accepted; signed fields unchanged
C08 Offline drift/sample-age budget exhausted Protected use refuses; no wall-clock fallback
C09 TLS bootstrap fails due to OS time; unknown/revoked authority key No insecure transport or circular trust acceptance
C10 Concurrent approval consume and delayed side effect Existing CAS and action/lifetime binding still enforced by consumer
C11 Local event before calibration or across clock step Original retained, translation unknown unless supported by evidence
C12 Two event intervals overlap / arrival order differs No fabricated total occurrence ordering
C13 Bad numeric units, nonfinite/overflow, malformed/oversized requests Bounded parsing and stable refusal without sensitive output
C14 Key rotation/revocation, stale trust cache and authority compromise drill Adopted trust recovery evidenced; old samples cannot resurrect trust
C15 Authority inaccessible/slow and sustained request load Bounded resources and honest unavailable state; liveness distinct from usability
C16 Consumer migration Native positive plus wrong caller/owner/tenant/signature/expiry negatives; no wider access grant

Threats include compromised authority/upstream, network delay/replay, malicious producer timestamps, privileged local clock manipulation and configuration rollback. A fully compromised client process can falsify its own evidence; server-side owners must not treat client-supplied time as authority. One-host deployment does not provide Byzantine consensus. Host administrator trust and upstream error assumptions must be stated with results. No microsecond-accuracy or universal offline guarantee.

RCLK-WP-0002 reviews the matrix. RCLK-WP-0003 supplies fake-clock/property tests and an isolated authority/client exercise. RCLK-WP-0004 supplies measured native evidence, consumer acceptance and failure/recovery drills. Performance thresholds become versioned, measured policies before production admission.