Skip to content

P6_A1_DECISION.md

Latest commit

 

History

History
90 lines (67 loc) · 4 KB

P6_A1_DECISION.md

File metadata and controls

90 lines (67 loc) · 4 KB
Raw

ExecSurface — P6-A1 Same-Workload Observation Decision

Date: 2026-09-30 Parent program: #100 P6 issue: #111 Protocol: docs/development/P6_A1_SAME_WORKLOAD_PROTOCOL.md Extraction: docs/development/P6_A1_EVIDENCE_EXTRACTION_001.md Machine matrix: experiments/p6-competitive-falsification/p6-a1-observation-matrix.json

Decision

P6_A1_OBSERVATION_CORPUS_COMPLETE_BOUNDED

This is a bounded factual-observation decision for the A1 lanes authorized by the preregistered protocol. It is not a product ranking, semantic-equivalence claim, performance claim, or release decision.

Accepted executable evidence

Corrected same-workload run:

  • source SHA: 110bf2ce400d50ac2cc2ae950de2de4e3a9cba12;
  • run: 36770813526;
  • conclusion: success;
  • frozen workload blob: c5986f1acd8bc79a3945e613acc128d5c8a9d168.

All authorized lanes completed S0–S3:

  • ExecSurface alpha.4;
  • Harden-Runner;
  • cicd-sensor.

Tetragon was not an authorized live lane because the A0 exact-execution pin condition was not yet satisfied. It therefore remains NOT_TESTABLE_YET and is not counted as a negative comparative result.

Retained historical negative evidence

The first A1 run remains immutable evidence:

  • source: bcf0da8e42648541b67a383db3c3c76ebc5b5790;
  • run: 36770551463;
  • ExecSurface artifact: 11123635852;
  • product evidence: S2 returned native ERROR because raw observation was incomplete;
  • harness defect: the workflow incorrectly stopped on stable exit 2 before S3.

The correction did not change workload, comparator pins, propositions, thresholds, or acceptance vocabulary. It only retained exit 2 as native evidence and continued the corpus.

Bounded findings

ExecSurface public alpha.4

  • S0: native pass;
  • S1: native review, added process-exec drift;
  • S2: native ERROR/INCOMPLETE, fail-closed before trusted canonical surface;
  • S3: native review, file open/write drift observed.

Harden-Runner

  • exact example.com:443 endpoint was correlated to curl in retained job log;
  • process monitor and project file monitor initialization were evidenced;
  • exact S1 child lineage was not emitted in the retained local job output;
  • exact S3 /tmp/execsurface-p6/s3.txt write was not emitted; the retained log stated the project file monitor scope as the repository workspace;
  • job/security-insights correlation was exposed, but no attestation-equivalent machine artifact was demonstrated by A1.

cicd-sensor

  • process execution + ancestry was retained for S1;
  • process-bound domain/network evidence was retained for S2, including explicit TCP destination evidence for curl;
  • exact S3 file write/read events were retained with process ancestry;
  • a machine-consumable Runtime Trace predicate was produced with GitHub run/job/workflow identity.

Native cicd-sensor predicate passed is not mapped to ExecSurface PASS.

Scientific boundary

A1 establishes only that the same frozen workload produced a reproducible factual corpus with meaningful differences in exposed evidence and semantics.

It does NOT establish:

  • an overall stronger/weaker product;
  • event-count superiority;
  • cross-backend authority equivalence;
  • that Harden-Runner cannot observe processes/files outside the retained local evidence;
  • that cicd-sensor network_connect equals P4 success-authority semantics;
  • that ExecSurface's S2 incompleteness is a competitor advantage/disadvantage;
  • any performance result.

Next gate

P6-A2 — adversarial false-PASS / identity-binding corpus.

A2 must target ExecSurface itself first. Required attack classes include:

  • baseline/evidence replay under a different expected run/source identity;
  • current-surface or evidence substitution;
  • workflow/source identity substitution;
  • observer incompleteness laundering into PASS;
  • process/file/network drift that might be incorrectly accepted;
  • CP4 predicate/attestation replay or unsigned-artifact substitution where applicable, reported only against each system's demonstrated verification contract.

No A1 result may be used to weaken A2 acceptance criteria.