Decisions

2026-11-01 S187 combines #540 and #541 because longer observation must remain aligned with managed-launch authority and release evidence. The 30-second launch and 60-second calibration defaults remain, while explicit launch and observation ceilings are five and ten minutes. Capture gets a separate five-second return allowance in both modes. Steam publisher roles are declared before effects; undeclared descendants remain observed exit obligations without process-control rights. Inception-time socket table evidence is bound to one process generation and exact launch ancestry after process collection, independently of packet-flow correlation. A bounded independent collector handles the socket-table read so an accept cannot stall the single proxy runtime worker; full, unavailable, or unfinished collection stays explicit. Missing or ambiguous data cannot become a final-client claim. The isolated installed-binary timing fixture uses the certification account's actual default paths and retains no public private-path data. Real-title compatibility and first-second protocol failures remain unverified under #542.

2026-11-01: Advance current v0.10.11 publication markers, documentation coverage, the frozen review candidate and six exact asset authorities only after merged source, annotated tag, all ten post-merge main workflows, all four tagged jobs, normal protected registry approval, six independently downloaded files and ten non-yanked registry archives reconcile. Retarget the published replay label to these verified bytes. Preserve the v0.8.0 predecessor, unsigned policy, historical evidence and unperformed independent whole-product review status. Hosted clean-install certification does not claim a local installed-product run or real-title Deep Capture success.

2026-11-02: The v0.10.12 release bump changes only eleven first-party package versions across the three governed dependency graphs. Rebind their measured digests and unsafe-review inventory without changing third-party packages, feature edges, package and edge counts, review expiry, or exceptions. Synchronize the nine first-party identities in each isolated fuzz and native-performance lockfile; their third-party resolutions remain unchanged. The independent-review record remains an empty not-started template for the new product version.