Repository navigation
P8 — External Validation, Independent Reproduction & Architecture Criticism #114
Description
Activity
P8-A0 closed; current external evidence gates preregistered
A0 decision:
P8_A0_EXTERNAL_VALIDATION_PACK_READY_BOUNDEDEvidence:
- workflow source
7ad1e6c139c4787f84e54261744a189e1835722a - run
36777065464— SUCCESS - artifact
11125794344 - digest
sha256:7d4021b4d93760f594a4faf26ad8ace2b1cdc2874f288fcd6e753d781483bce5 - decision commit
00e3751d3ad881add11cf4833a294cd02edaa741
Canonical P8 evidence files now include:
docs/external/P8_EXTERNAL_EVIDENCE_LEDGER.mddocs/development/P8_A1_INDEPENDENT_REPRODUCTION_PROTOCOL.mddocs/development/P8_A2_EXTERNAL_CRITICISM_PROTOCOL.mddocs/development/P8_A3_INTEROPERABILITY_REVIEW_PROTOCOL.mddocs/development/P8_A4_EXTERNAL_REAL_WORKLOAD_PROTOCOL.mddocs/development/P8_A5_EXTERNAL_VALIDATION_CLOSEOUT_PROTOCOL.md
A5 closeout minimum was frozen before current external results arrive: at least two distinct current qualifying external evidence records after A0, spanning (1) execution/use evidence from A1 or A4 and (2) challenge/interoperability evidence from A2 or A3. Historical
EXT-0001remains qualifying prior criticism but cannot substitute for the current two-record minimum.Current exact state:
P8_EXTERNAL_EVIDENCE_INSUFFICIENT — CURRENT EXTERNAL INPUT PENDINGThis is an external dependency, not a reason to replace independence with internal CI. No endorsement/adoption/independent-reproduction claim is authorized.
- workflow source
P8-A5 current external-evidence outreach — 2026-10-01
A5 remains preregistered as
P8_EXTERNAL_EVIDENCE_INSUFFICIENT — CURRENT EXTERNAL INPUT PENDING.Two narrowly scoped current requests were sent after the A0 freeze, targeting the two distinct evidence sides required by A5:
-
Track R — zero-assistance reproduction / execution-use side
- evaluator target: independent developer-security reviewer previously contacted before P8
- Gmail sent message id:
1a0f431e5c77822f - request: use public
v0.1.0-alpha.4and public docs only; preserve initial result before troubleshooting; failure/partial/no-fit are explicitly acceptable
-
Track I — standards/interoperability challenge side
- evaluator target: independent in-toto/supply-chain researcher previously contacted before P8
- Gmail sent message id:
1a0f4320f5d5fecc - request: challenge
P5_EXISTING_STANDARDS_COMPOSITION_SUFFICIENT_BOUNDED, specifically semantic mismatch, producer/verifier identity, authority/completeness representation, duplicate fields, replay/substitution, or a better existing interoperability path
Both requests point to the frozen
docs/external/P8_EXTERNAL_EVALUATOR_PACKET.md.Anti-drift boundary: outbound requests are not external evidence. No P8 closeout, endorsement, adoption, independent-reproduction, or interoperability-success claim is authorized until qualifying external technical results actually arrive and are entered in the P8 evidence ledger.
-
Alpha.5 internal release-readiness update — P8 remains OPEN
Alpha.5 has now completed the internal clean-candidate program without changing the P8 independence rule.
Frozen product source:
5067200452c174da6bc8d9d7ecf6957ee379f0a2Internal evidence now closed:
- fresh Alpha.5 A1 Semantics v3 reproof — PASS (
36898076576) - A2 proposition/backend authority — PASS (
36898309801) - A3 legitimate-variance anti-poisoning — PASS (
36898477456) - A4 attestation/provenance authority — PASS (
36898814464), with prior harness-only failure retained - A5 product/usability/package/reproducibility — PASS (
36899408076), with prior harness-only failure retained - A6 integrated destructive assault — PASS (
36899958419), with prior governance-locator failure retained - frozen-source independent red-team — PASS (
36900214264), including 500-run PATH-TOCTOU/shared-FD replay on Ubuntu 22.04 and 24.04 - release dry-run — PASS (
36900721300), with staged alpha.5 bundle/checksum/provenance and no public-ref mutation
Release decision:
ALPHA5_INTERNALLY_RELEASE_READY_BOUNDED — PUBLIC_RELEASE_BLOCKED_P8_EXTERNAL_EVIDENCE.This update does not count as external evidence and does not advance A1/A2/A3/A4 of P8. Current review still found no qualifying reply to the focused William Woodruff or Santiago Torres requests. Routing/referral responses remain non-qualifying by themselves under this issue's preregistered rules.
Public
v0.1.0-alpha.4and stablev0.1therefore remain unchanged. Full evidence ledger:docs/release/ALPHA5_RELEASE_DECISION.md.- fresh Alpha.5 A1 Semantics v3 reproof — PASS (
P8 external-validation execution has been expanded without changing the preregistered acceptance rules.
Public review hub: #118
Exact frozen product SHA under review:
5067200452c174da6bc8d9d7ecf6957ee379f0a2.The review request explicitly asks for qualifying evidence in the existing P8 classes: zero-assistance execution/reproduction or an external workload, and challenge/counterexample/interoperability feedback. Negative results are explicitly welcome.
Alpha.5 remains unpublished while P8 is open; this outreach does not self-certify the gate.
Live Alpha.5 release-boundary recheck — 2026-10-01
Current verification from the active release-candidate branch:
- branch
integration/post-alpha4-promotion-candidateremote HEAD remains67130f006799fbc630ad8cd7f26d9b9bf7073f62(Alpha5: remove redundant ABI-floor rehearsal after canonical v2 success); no newer commit was present at this recheck; - the HEAD-only change after the A7 closeout removes the redundant ABI-floor rehearsal workflow; it does not alter the frozen product source or erase historical run
36902483580; - final artifact-level A7 remains accepted as
ALPHA5_A7_RELEASE_ARTIFACT_DESTRUCTIVE_EQUIVALENCE_PASS — BOUNDED INTERNAL, accepted run36902486872; initial material failure36901600086remains retained negative evidence; - frozen Alpha.5 product source remains
5067200452c174da6bc8d9d7ecf6957ee379f0a2; - current public release remains
v0.1.0-alpha.4at source48e0b9a0707553349e75e97a9dfa096d13f9ab5d; - stable Action tag
v0.1still resolves to the alpha.4 source; - no
v0.1.0-alpha.5Git ref exists and no Alpha.5 GitHub Release is published.
P8 external-evidence recheck
Issue #114 remains OPEN. The preregistered A5 minimum is still binding: at least two distinct current qualifying external records after A0, spanning (1) A1/A4 execution-use evidence and (2) A2/A3 challenge/interoperability evidence.
A fresh Gmail recheck found no inbound reply from the currently targeted independent reviewers, including the focused William Woodruff and Santiago Torres tracks and the additional Alpha.5 review requests sent to Kees Cook, Paul Moore, Liz Rice, Josh Triplett, Marina Moore, Justin Cappos, Dan Lorenc, and Adam Korczynski.
The recent outbound Alpha.5 requests are recorded only as outreach/dependency expansion. They do not count as external evidence, validation, endorsement, reproduction success, interoperability success, or adoption.
Decision
ALPHA5_INTERNALLY_RELEASE_ARTIFACT_QUALIFIED — PUBLIC_RELEASE_BLOCKED_P8_EXTERNAL_EVIDENCENo release transaction is authorized now:
- do not create
v0.1.0-alpha.5; - do not publish Alpha.5 release assets or crates;
- do not move stable
v0.1; - do not reinterpret internal A1–A7/red-team evidence as external validation.
The next legitimate state transition is receipt of qualifying external technical evidence, followed by qualification under the frozen P8 protocol. Until then, the external dependency is the blocker.
- branch
Alpha.5 publication-governance update — owner override (2026-10-01)
P8 remains scientifically open and its evidence state is unchanged: current qualifying external evidence is still insufficient. The owner has explicitly authorized Alpha.5 publication without using P8 as a pre-publication blocker.
This does not close P8 and does not convert internal evidence into external validation. P8 will continue after publication and qualifying external reports will still be evaluated under the frozen P8 protocols.
Public-release wording must therefore remain bounded: Alpha.5 may state its internal qualification and post-publication verification results, but must not state or imply independent external validation until qualifying evidence actually exists.
Governance state:
P8_OPEN_POST_RELEASE — PREPUBLICATION_BLOCKER_WAIVED_BY_OWNER — NO_EXTERNAL_VALIDATION_CLAIMCurrent external-evidence qualification update — 2026-10-03
External contribution evidence is now recorded on
mainvia PR #139 / commit65ffaf848db5f03d523be66c18b058ca5f69e1d7.Current records added to
docs/external/P8_EXTERNAL_EVIDENCE_LEDGER.md:P8-EXT-0001— PandaHUN777 PR feat(examples): add deterministic hello-drift smoke test #126, merged external contribution;P8-EXT-0002— astrogilda PR docs: add deterministic Alpha.5 Python/pytest integration recipe #128, open external Python/pytest contribution with retained Alpha.5 execution results and later restricted-environment failures;P8-EXT-0003— PandaHUN777 PR docs: add GitHub Action consumer example #127, open external Action-example contribution with a retained maintainer-requested portability correction.
All three are genuine
EXTERNAL_CONTRIBUTIONrecords under A0. None has been laundered into A1/A4:- contributor-owned tests are not called independent A1 reproduction without the preregistered relationship/assistance separation;
- the PR docs: add deterministic Alpha.5 Python/pytest integration recipe #128 fixture was authored specifically for the ExecSurface example, so it is not an A4/P9.5 external real workload;
- the maintainer-found docs: add GitHub Action consumer example #127 portability defect is retained as an open contribution finding, not external architecture criticism.
P8 closeout state
Still OPEN / INSUFFICIENT FOR A5.
The remaining required current evidence pair is still:
- A1 zero-assistance reproduction/failure/independent partial attempt or A4 external real workload; and
- A2 architecture criticism/counterexample/no-fit or A3 interoperability guidance.
No external issue authored outside AETHER X currently exists in the repository search, so there is no additional independent issue report being omitted from this update.
New current external interoperability evidence identified and preserved.
Source:
probityai/agent-evidence-observer#41(merged 2026-10-02), which integrates the published ExecSurface Alpha.5 binary in an external repository and builds a bounded state/evidence companion around the retained raw trace.P8 classification:
INTEROPERABILITY_GUIDANCE/ qualifying A3 evidence, scientific processing OPEN. The external implementation identifies concrete expressivity gaps around typed collection health and successful write-result/byte-count semantics while explicitly retaining same-operator and incomplete-scope limits.This is not being relabeled as A1 zero-assistance reproduction, A4 real-workload evidence, production adoption, independent custody, endorsement, or complete validation.
Reproduction/design gate: #145
Ledger update PR: #146P8 A5 remains OPEN because a qualifying current A1/A4 execution-use record is still missing.
P8-A3 current interoperability evidence has now been fully processed.
- external source:
probityai/agent-evidence-observer#41; - ledger record:
P8-EXT-0004; - exact gap reproduced under P8-A3 — reproduce Probity interoperability gap before any schema change #145;
- research closeout merged in test(p8-a3): reproduce Probity gap and qualify typed report bridge #147;
- ledger closeout merged in docs(p8): record reproduced Probity interoperability evidence #148;
- selected bounded response:
REPORT_LAYER_TYPED_HEALTH_BRIDGE; - no raw-schema rewrite or public Semantics v3 promotion authorized.
P8 remains OPEN. The challenge/interoperability side of the current minimum now has a reproduced qualifying record, but the execution/use side still lacks a qualifying A1 zero-assistance reproduction/failure/independent partial attempt or A4 external real-workload report.
Follow-up #149 is research-only implementation and cannot substitute for that missing external execution/use evidence.
- external source:
Internal response to P8-EXT-0004 is now complete through an unreleased experimental implementation.
- P8-A3 — reproduce Probity interoperability gap before any schema change #145: external interoperability gap reproduced and bounded path selected
- P8-A3.1 — research-only typed evidence report bridge prototype #149 / test(p8-a3.1): prototype typed evidence report at internal boundary #150: typed-report prototype accepted
- P8-A3.2 — bounded public typed-evidence API design gate #151 / docs(p8-a3.2): select opt-in same-run typed evidence file #152: opt-in same-run API design selected
- P8-A3.3 — experimental same-run typed evidence output implementation #153 / feat(p8-a3.3): add experimental same-run typed evidence output #154: experimental implementation accepted and merged
- current main:
2fa91e73d35e74267b915de4f1b582efdc0fb612
Important boundary: none of this internal work satisfies the missing execution/use side of P8 A5.
The next legitimate P8 priority is therefore A1/A4 external execution/use evidence, not additional internal feature expansion. The experimental interface remains unreleased and must not be represented as external adoption or validation.
External evaluator outreach wave 1 — invitation only, not evidence
Date: 2026-10-04
Direct writes to third-party repository threads were attempted through the GitHub integration and returned
403 Resource not accessible by integration. No external comment is claimed.Fallback used: dedicated invitation issue #159 in this repository with individual GitHub mentions so the outreach remains transparent and non-spam.
Invited:
- @scadastrangelove — prior independent static critic / WATCHLIST reviewer; strongest current independence profile.
- @GuillaumeFalourd — external maintainer who previously merged the ExecSurface Action example; strong real-workload / GitHub Actions candidate.
- @astrogilda — Probity interoperability integrator whose work exposed the typed-health gap; technically strong but independence will be qualified conservatively because of prior ExecSurface interaction.
- @bb1nfosec — independent runtime supply-chain maintainer; BHEESHMA has an out-of-process strace/ptrace model, making this a strong adversarial-overlap candidate.
All invitations point to:
- packet
03cb5dd656d6090cd2cc5c5d40885334b06b344a; - product target
cea214244d7efe05fea2108a5f5d8ed6287d9a3f; - zero-assistance rule;
- evaluator-selected workload;
- initial-result preservation;
- explicit acceptance of negative/partial/NO_FIT outcomes.
Classification:
OUTBOUND_INVITATION_ONLY — NONQUALIFYINGNo P8/P9.5 gate is advanced until an actual external result is received and qualified.
Reacted by Guillaume Falourd
Current stable-v1 authority update — 2026-10-08
v1.0.0e70169b959f2163715090c371335fa6c5591e3e4v1.0.0; older Alpha source identities remain valid only for explicitly frozen historical evidence.The historical P8 program body below is retained as provenance. Where it names Alpha.5 as the current release, that wording is historical and is superseded by this current-authority block.
Status
OPEN —
P8_POST_RELEASE_EXTERNAL_VALIDATION_OPENCurrent public release:
v0.1.0-alpha.5Release source:
9e73b925d55557e33de1b0813995609aaefdc037Frozen product source used for Alpha.5 qualification:
5067200452c174da6bc8d9d7ecf6957ee379f0a2Public review/intake hub: #118
External engagement tracking: #94
Alpha.5 was published after an explicit owner decision to waive P8 only as a pre-publication requirement. That decision did not close P8, did not create external evidence, and must not be represented as independent validation, endorsement, adoption or approval.
Objective
Obtain and process technically meaningful evidence from people outside AETHER X without converting participation, routing, praise, organizational contact, outreach, or internal rehearsal into a false validation/adoption claim.
P8 accepts negative, partial, no-fit, interoperability, reproduction-failure and architecture-criticism outcomes as first-class evidence.
Fixed roles
Dynamic specialists
Governance
External criticism follows:
CLAIM -> EXTERNAL CRITICISM/RESULT -> QUALIFICATION -> TEST/REPRODUCTION -> EVIDENCE -> DECISIONNever
criticism -> defenseand neverparticipation/outreach -> endorsement.Evidence baseline
docs/external/OPENSSF_FEEDBACK_LEDGER.mdasEXT-0001; it influenced ptrace authority boundaries / hybrid architecture work;P8 gates
A0 — External-evidence qualification and review-pack freeze
Complete / retained. Qualification rules and evaluator material are frozen. Internal rehearsals may test pack integrity but cannot satisfy the external-evidence criterion.
A1 — Independent zero-assistance reproduction
Obtain at least one external attempt against the public Alpha.5 path without live assistance. Failure, partial completion and unsupported environment are retained exactly. Assisted follow-up cannot rewrite the original result.
A2 — Architecture criticism / counterexample
Capture at least one substantive current architecture criticism, counterexample, no-fit finding or explicit overlap/redundancy assessment. Qualifying criticism is more important than praise.
A3 — Standards/interoperability review
Seek technically specific feedback on the in-toto Runtime Trace / SCAI / SVR / SLSA / Sigstore composition. Do not claim standards endorsement.
A4 — External real-workload report
Seek at least one report using a workload not created by AETHER X. Record setup friction, evidence quality, false PASS/REVIEW/incompleteness and privacy observations separately.
A5 — P8 closeout
Close only from qualified external evidence. Internal tests can reproduce or falsify external findings but cannot manufacture independence.
Current closeout minimum
At least two distinct current qualifying external evidence records after A0, spanning:
Historical
EXT-0001remains valid historical criticism but does not substitute for the current two-record minimum.Qualifying external evidence classes
ZERO_ASSISTANCE_REPRODUCTIONREPRODUCTION_FAILUREARCHITECTURE_CRITICISMCOUNTEREXAMPLENO_FIT_OR_REDUNDANCYINTEROPERABILITY_GUIDANCEEXTERNAL_REAL_WORKLOAD_REPORTEXTERNAL_CONTRIBUTIONNon-qualifying by itself
Immutable boundaries
NO_FITintoFIT;SECURITY.md;Current action
Keep #118 open for independent Alpha.5 reproduction/challenge reports, qualify every substantive external result against this protocol, preserve the initial result before assistance, and close P8 only when the frozen current-evidence minimum is genuinely satisfied.