Krometrail

Appearance

CDP transport qualification evidence (historical) ​

This directory contains retained version-1 machine-readable evidence. It is obsolete and is not accepted by the strict version-2 contract under ../v2/; reports and the decision remain byte-for-byte historical records. Raw runner logs and browser profiles remain outside Git under target/.

Selected transport ​

The historical Rust decision selected exact published cdpkit 0.4.0 (Cargo.lock checksum c3fdb566d913b31e0014391a94c0db4ed871dbb76577dd1b2f2c5f6df158bfaa) because both reports passed the then-current 13 gates. That is not a current qualification claim. Phase 2 wire-authenticity remediation changed the qualification harness after these reports were captured; the reports and decision.json are retained unchanged and must not be treated as refreshed decisive evidence until both platforms are requalified.

  • Linux report: cdpkit-linux.json
    • SHA-256: 081259729e2495e999745bcd7caa509ec7effc844f50b2a4d786d6cc744c7feb
    • Platform: Linux x86_64; Chrome 149.0.7827.155
  • macOS report: cdpkit-macos.json
    • SHA-256: 3ffe94f405038fd8d9efd9fa7f8acbf15e8cb02c1f9e19bf24397f180981d401
    • Platform: macOS aarch64; Chrome 149.0.7827.201

Both reports use schema version 1, the same exact candidate, fixture digest, and unchanged configuration: at least 60 seconds, 1,000 frames, 10 seconds of capacity-1 saturation, and 100 saturation attempts. Every gate is present and pass; no platform exception or threshold waiver exists. They are now explicitly historical/obsolete inputs: the strict contract additionally requires a positive hard_stop_seconds and observed disconnect/rebuild elapsed fields, which these unchanged reports do not contain. They must not be edited or used for a new decision; both platforms require fresh qualification.

Validation and decision reproduction ​

From the repository root:

bash
cargo run -p krometrail-cdp --features cdp-spike --bin cdp-transport-gate -- \
  schema --output /tmp/cdp-transport-schema.json
cargo run -p krometrail-cdp --features cdp-spike --bin cdp-transport-gate -- \
  validate-and-normalize --input docs/evidence/cdp-transport/v1/cdpkit-linux.json \
  --output /tmp/cdpkit-linux.normalized.json
cargo run -p krometrail-cdp --features cdp-spike --bin cdp-transport-gate -- \
  validate-and-normalize --input docs/evidence/cdp-transport/v1/cdpkit-macos.json \
  --output /tmp/cdpkit-macos.normalized.json
cargo run -p krometrail-cdp --features cdp-spike --bin cdp-transport-gate -- \
  decide --linux-report docs/evidence/cdp-transport/v1/cdpkit-linux.json \
  --macos-report docs/evidence/cdp-transport/v1/cdpkit-macos.json \
  --output /tmp/decision.json
sha256sum docs/evidence/cdp-transport/v1/cdpkit-linux.json \
  docs/evidence/cdp-transport/v1/cdpkit-macos.json
cmp /tmp/decision.json docs/evidence/cdp-transport/v1/decision.json

The decision command verifies schema decoding, strict fields, complete gate registry, observed deadline measurements, measured thresholds, candidate/version/checksum consistency, Linux/macOS platform identity, fixture/configuration consistency, and redaction before hashing the exact report bytes. Against the retained historical reports it now fails during strict decoding/validation by design; re-run it only after fresh Linux and macOS reports are captured. The schema is generated from the Rust evidence types; schema.json is not a separate source of truth.

Gate interpretation and limitations ​

The 13 gates cover deterministic routing, typed domains, flat-session isolation, browser/session raw commands, named raw event parameters, protocol-drift survival, sustained screencast, prompt acknowledgement, bounded handoff saturation, RSS trend, disconnect cleanup, and explicit reconnect/session rebuild.

The decision preserves these limitations exactly:

  • cdpkit exposes named event parameters, not wildcard or full-envelope receive. The remediated runner binds unknown-event/additive-field/unknown-enum fixtures to a candidate-only observed-wire trace; it does not call those fixtures real-Chrome measurements.
  • cdpkit's event subscriber is unbounded and its queue depth is not inspectable; the continuously drained RSS and handoff counters are process-level proxies only and do not prove the hidden queue is bounded.
  • acknowledgement latency is a receive-to-ack-completion proxy, not a wire-enqueue timestamp.
  • Krometrail owns reconnect/session restoration, bounded handoff, backpressure, capture gaps, cancellation, and flush behavior.

chromey was not tested because no cdpkit lifecycle, ordering, or sustained-capture failure was demonstrated. An owned Tokio/tokio-tungstenite transport remains the fallback if a later cdpkit integration needs a fork, loses raw protocol evolution before a preserved boundary, misroutes sessions, or obscures acknowledgement/backpressure. Neither fallback is wired or selected by this evidence.

The spike remains non-default and is not root-wired. This rollup selects the mechanism for the later production adapter; it does not implement production lifecycle, capture, reconnect, or core-port changes.

Last updated: