Repository navigation
Post-M9 — eBPF completeness closure & same-workload value requalification gate #85
Description
Activity
E1 lifecycle attribution — CLOSED
Formal result:
docs/milestones/POST_M9_EBPF_E1_LIFECYCLE_DIAGNOSTIC_RESULT.md
Commit:80722931e29d375f489307fcfcc3e11ec2e678b5
Run: https://github.com/AETHERXGLOBAL/execsurface/actions/runs/36300797733Frozen diagnostic result:
casey/just: 3/3 lifecycle-drain complete, dropped events 0, exit 0; artifact10925149811, digestsha256:d3cb40bed68ba57f129b1ead448bd7e4b18275e6bdfda6cb566988ac117d152f.junegunn/fzf: 3/3 lifecycle-drain complete, dropped events 0, exit 0; artifact10925267846, digestsha256:47073c39c2caab249e9e00fa76ea331a785c0a86a9c83393717e583b4207a31e.- all six runs reported only
incomplete_capability, which is not a hard M9.1 lifecycle failure.
Classification: historical intermittent lifecycle failure not reproduced in the preregistered six-run diagnostic. Previous M9.1
incomplete_lifecyclefailures remain authoritative negative evidence and are not relabeled.No timeout increase or tracker heuristic is authorized. The old M8.4d per-invocation collector remains unsuitable for renewed value qualification because its syscall-exit membership propagation boundary had already been killed by M8.7.
E2 direction: build a research-only value-requalification collector carrying forward the M8.7 session-epoch/root-barrier/
sched_process_forklifecycle discipline. First gate is lifecycle/health only on the same pinned workloads. No performance timing or authority change is authorized yet.E2 persistent real-workload health — CLOSED / PASS
Formal result:
docs/milestones/POST_M9_EBPF_E2_PERSISTENT_REAL_WORKLOAD_HEALTH_RESULT.md
Commit:d71284de06ac7f50ffcc163ab2c09a01cc408f71
Run: https://github.com/AETHERXGLOBAL/execsurface/actions/runs/36301162578Results:
casey/just: 3/3 persistent invocations PASS, 6/6 sessions PASS; artifact10926175328, digestsha256:9de1bd5dd4b005b2eec79144277df1616eb29ec3576fd7bad66f3ce74fd64941.junegunn/fzf: 3/3 persistent invocations PASS, 6/6 sessions PASS; artifact10925861295, digestsha256:3afee45753c3d823a26b3e640fae7b66a720e8fc10b9b96c887b0d55a2c56edd.
No accepted session had lifecycle leak, stale epoch, routing error, producer drop, decode error, integrity error, event-limit hit, residual membership, or authority-boundary violation.
The committed M8.7 persistent collector source was unchanged; only a runner-local barrier adapter was used to hand off the pinned real commands after root epoch registration.
E2 proves lifecycle/health only. Full ptrace semantic parity and performance value remain unestablished. E3 is now authorized to test only shared lifecycle propositions before any performance timing.
E3 shared lifecycle proposition parity — CLOSED / PASS
Formal result:
docs/milestones/POST_M9_EBPF_E3_SHARED_LIFECYCLE_PARITY_RESULT.md
Commit:812ddfaca52bc9a471fc0c8b8239be2bf7394183
Run: https://github.com/AETHERXGLOBAL/execsurface/actions/runs/36301688725
Artifact:10925328847
Digest:sha256:6709544a4e3023d661c52d0cc81c8e40e66a4d861cf2db627e5115cf7dda9cc2Result:
- ptrace reference/reference proposition determinism PASS;
- ptrace projection: 2 root-child spawns, success child exactly 2 successful exec occurrences, failed-exec child 0 successful exec occurrences, exit 0;
- persistent eBPF epoch 1 projection exactly matched ptrace;
- persistent eBPF epoch 2 projection exactly matched ptrace;
- each eBPF session had event_count 7 / spawn 2 / exec 2 / exit 3 and passed every M8.7 health gate;
- BPF source remained unchanged; only transcript-only userspace evidence instrumentation was applied inside the runner and preserved in the artifact.
Authority remained unchanged:
full_surface_comparable=false,ebpf_pass_authorized=false,product_integration_authorized=false.E4 same-workload performance requalification is now authorized to be preregistered. Public eBPF authority remains prohibited.
E4 execution start — development-branch isolation
Execution is starting from the isolated branch
development/post-alpha4-behavioral-integrityat HEADc70a92694094e28c628f428cc5dd02f9e6064109.The E4 scientific protocol is already preregistered in
docs/milestones/POST_M9_EBPF_E4_SAME_WORKLOAD_VALUE_PROTOCOL.mdand the E4 harness/workflow already exist. No workload, sample count, health gate, threshold, classification rule, authority boundary, or harness semantics will be changed before measurement.One execution-plumbing-only change is authorized now: extend the E4 workflow push trigger from
mainto the isolated development branch so the preregistered experiment executes away from the published alpha.4 line. This trigger change does not modify the experiment itself.Frozen E4 rules remain: 3 warmups + 15 measured samples per mode, no outlier removal, no sample replacement, health/completeness before timing acceptance, 10% per-target threshold, and no public eBPF authority regardless of result.
Published
v0.1.0-alpha.4/ stable@v0.1remain untouched.E4 same-workload value requalification — CLOSED / BLOCKED BY INCOMPLETE EVIDENCE
Formal result committed on the isolated development branch:
docs/milestones/POST_M9_EBPF_E4_SAME_WORKLOAD_VALUE_RESULT.mdCommit:
337bc394291844d01496e146c590875b67f3aaf1
Run: https://github.com/AETHERXGLOBAL/execsurface/actions/runs/36490724891
Execution source:3c21521efd7b0f150d3785993916562c908f2f2cBoth preregistered targets (
junegunn/fzfandcasey/just) were blocked at ptrace warmup #1 by the existing alpha.4 fail-closedshared_fd_table_ambiguityguard. In both cases the target exited 0, butcomplete=false, so the protocol correctly refused to admit timing samples.No 15-sample measured set was formed. No failed sample was replaced. No rerun was used to seek a cleaner result. The 10% threshold and all health/authority rules remained unchanged.
Artifacts:
- FZF:
11001308264,sha256:611038a250936981d478d808c04e748b6389ed13c7602a76b085ab6a19fb69d1 - JUST:
11001373432,sha256:4c3e64ed009fb29e998a72de43b6881d4f3ded994dd541754f9b0000d0a242dd
Formal classification:
E4_BLOCKED_BY_INCOMPLETE_EVIDENCE.Interpretation: this does not establish an eBPF speed win or loss. It establishes that the current completeness contract makes the value comparison inadmissible on these concurrent workloads. The negative result is retained and becomes an input to Semantics v3/shared-FD authority work.
Authority remains unchanged: ptrace public reference retained; eBPF research-only; no eBPF learn/check/PASS, auto-selection, full-surface comparability, or baseline interchangeability.
Issue #85 can now close as completed scientific work with a negative/incomplete result. No public alpha.4 change is authorized.
- FZF:
Objective
Reopen the existing eBPF path as a research-only performance architecture gate for one reason only: real external ptrace workloads have now established a material stop-tax bottleneck, while the previous triggered eBPF value screen was killed before timing because the collector reported
incomplete_lifecycle/incomplete_capability.This issue does not authorize public eBPF exposure or change backend authority.
Evidence basis
docs/milestones/M8_8_PUBLIC_INTEGRATION_DECISION.mddocs/milestones/M9_1_EBPF_TRIGGER_VALUE_SCREEN_RESULT.mddocs/milestones/POST_M9_PTRACE_COST_ATTRIBUTION_RESULT.mddocs/milestones/POST_M9_PTRACE_OPTIMIZATION_BOUNDARY_DECISION.mdThe current ptrace reference has a measured causal all-syscall stop tax. Two bounded native micro-optimization candidates were rejected for integration under frozen value gates. A selective-stop seccomp route is not authorized as a transparent modification of the current public ptrace backend because it changes tracee process/security state.
Immutable authority boundary
Throughout this gate:
learnis NOT authorized;checkis NOT authorized;Fixed roles
Phase E1 — reproduce and attribute completeness failures
Before changing collector code:
mainsource SHA;incomplete_lifecycle;incomplete_capability;No timing value claim is admissible in E1.
Phase E2 — bounded completeness closure
Any code change must be justified by E1 attribution and must preserve fail-closed behavior.
Required properties include, where applicable:
Phase E3 — parity/health gate
Only after E2:
This gate still does not authorize public learn/check/PASS or baseline interchangeability.
Phase E4 — same-workload value requalification
Only if E1–E3 close successfully, preregister a value experiment before measurement on the same pinned real workloads that triggered the research path.
At minimum preserve:
A value claim is killed if eBPF completeness fails before accepted measured samples.
Prohibited shortcuts
incomplete_lifecycleorincomplete_capabilitydefinitions;First action
Read the current eBPF collector/session implementation and existing M8.3/M8.7 evidence, then produce an E1 failure-attribution record before any collector code change.