Skip to content

Post-alpha.4 — Competitive Architecture & Behavioral Integrity Advancement Program #100

Closed
Closed
Post-alpha.4 — Competitive Architecture & Behavioral Integrity Advancement Program#100
Labels
research

Description

@aetherxeg-source
aetherxeg-source
opened on Sep 28, 2026

Status

OPEN — PLANNING / EVIDENCE-GATED EXECUTION

Source of truth at program creation

Strategic objective

Raise ExecSurface from a strong public-alpha runtime-drift tool into a defensible runtime behavioral integrity and verification layer for software execution.

The program must NOT turn ExecSurface into a smaller Falco/Tetragon/EDR clone. The durable differentiation target is:

backend-agnostic evidence semantics + authority/completeness + canonical accepted surface + deterministic drift + explicit policy + attestable verification result

Collectors are replaceable evidence backends; they are not the product moat.

Fixed expert roles

  1. Innovation Scientist / Systems Architect (fixed) — seek non-obvious, high-leverage architecture; challenge incremental imitation.
  2. Anti-Drift / Scientific Integrity Reviewer (fixed) — block claim inflation, gate skipping, threshold weakening and competitor-driven feature drift.
  3. Independent Falsifier / Red Team (fixed) — construct false-PASS, false-REVIEW, incompleteness, concurrency, provenance and backend-equivalence counterexamples.

Dynamic specialist pool

  • Linux kernel / ptrace / eBPF / BPF-LSM engineer
  • Runtime evidence semantics / PL / formal methods specialist
  • CI/CD supply-chain security architect
  • in-toto / SLSA / Sigstore provenance specialist
  • Reproducibility / benchmarking / performance engineer
  • Filesystem/process/network identity specialist
  • Nondeterminism / build-systems specialist
  • Privacy / data-minimization security engineer
  • Developer-experience / GitHub Actions / GitLab integration engineer
  • OpenSSF ecosystem/interoperability lead

Specialists are selected per gate; fixed roles remain present throughout.

Competitive facts that constrain the plan

  • Falco/Tetragon/Tracee already own broad runtime observability/detection territory.
  • Tetragon has stronger kernel-authoritative hook options and explicit LSM/kernel-state paths.
  • StepSecurity Harden-Runner already performs CI process/file/network monitoring and network behavioral baselining with broad adoption.
  • cicd-sensor is a close emerging competitor: open-source eBPF CI/CD sensor, GitHub/GitLab support, amd64/arm64, runtime reports and runtime-trace attestation.
  • in-toto already defines https://in-toto.io/attestation/runtime-trace/v0.1; ExecSurface must interoperate rather than invent an isolated competing attestation format.

North-star architecture

                 Source / artifact / workflow identity
                            |
                            v
             Provenance binding (SLSA / Sigstore)
                            |
                            v
Declared command -> Evidence Backend Adapter Layer
                     |      |       |       |
                  ptrace  native   Tetragon external
                         BPF/LSM    adapter  traces
                     \      |       |      /
                      v     v       v     v
              Authority-graded typed evidence
                      + observer health/loss
                            |
                            v
                Canonical Execution Surface
                            |
               +------------+------------+
               |                         |
        Accepted surface            Current surface
               |                         |
               +------------+------------+
                            v
                 Deterministic semantic diff
                            |
                            v
                     Explicit policy
                            |
                            v
              PASS / REVIEW / BLOCK / ERROR
                            |
                            v
      Attestable verification result + runtime trace export

Program gates

P0 — Freeze product identity and competitive boundary

Deliverable: EXECUTION_INTEGRITY_NORTH_STAR.md.

Freeze:

  • ExecSurface is runtime behavioral integrity / verification, not EDR/threat detection.
  • no malware-rule race with Falco/Harden-Runner/cicd-sensor;
  • no claim that runtime monitoring, baselining or eBPF is novel;
  • differentiation is semantic/evidence composition and explicit authority/completeness.

Kill criterion: if external review shows no useful distinction from existing systems, record NO_FIT and reassess before feature expansion.

P1 — Finish Issue #85 E4 before any new backend promotion

Do not create a new kernel collector path until E4 closes.

Required:

  • preregister same-workload performance protocol;
  • direct / ptrace / persistent research eBPF comparisons;
  • completeness/health admissibility before timing;
  • preserve all failures;
  • no PASS/learn/check/public authority promotion from E4 alone.

Decision output must be one of:

  • VALUE_SIGNAL_ESTABLISHED_BOUNDED
  • NO_VALUE_SIGNAL
  • INCOMPLETE_FOR_VALUE_CLAIM

P2 — Execution Surface Semantics v3 (core moat)

Goal: make semantics backend-independent and stronger than raw event logs.

Define:

  • proposition-scoped evidence types;
  • object/effect identity rules;
  • authority level per proposition;
  • completeness/loss state per proposition;
  • causal lineage contract;
  • canonical equivalence rules;
  • explicit unsupported/ambiguous states;
  • baseline/diff compatibility rules across schema versions.

Critical innovation target: proof-carrying observation records — every canonical effect carries enough metadata to state what proposition is supported, by which observer capability, at what authority/completeness level.

No backend interchangeability claim until proposition-by-proposition equivalence is proved.

P3 — Nondeterminism & legitimate-variance model

Goal: reduce false REVIEW without hiding drift.

Build a separate variance model, never an auto-approved baseline mutation.

Classify variance sources:

  • temp roots;
  • generated paths;
  • caches;
  • randomized test directories;
  • parallel/interleaving-only differences;
  • timestamps/ephemeral identifiers where semantically irrelevant;
  • genuine behavior expansion/contraction.

Rules:

  • variance learning cannot silently authorize behavior;
  • canonicalization changes require adversarial fixtures;
  • preserve raw evidence and pre-normalized comparison;
  • user/policy approval remains separate.

Success metric: materially lower false-review rate on pinned real build/test workloads with zero newly introduced false-PASS counterexamples in the declared suite.

P4 — Backend adapter / authority architecture

Goal: separate semantic product from collection implementation.

Prioritize adapters before rebuilding mature collectors.

Candidate order:

  1. ptrace reference adapter (existing authority baseline);
  2. proposition-scoped native BPF-LSM/kernel-hook research adapter where ptrace authority is known weak;
  3. external trace import experiment from Tetragon/cicd-sensor where licensing/schema permits;
  4. audit/Falco-style import only if a concrete integrity use case exists.

No full-host agent or daemon unless a measured gate proves it is necessary.

P5 — Runtime attestation & provenance integration

Do NOT invent a competing ecosystem standard.

Deliver:

  • in-toto Runtime Trace export/import mapping;
  • ExecSurface verification-result attestation binding:
    • command identity;
    • source/artifact/workflow identity where available;
    • baseline digest;
    • current-surface digest;
    • observer/backend identity + capability/completeness;
    • policy digest;
    • verdict;
    • evidence digest;
  • SLSA provenance link where build provenance exists;
  • optional Sigstore/GitHub attestation signing/verification path.

Research question: determine whether existing in-toto verification predicates suffice or a narrowly scoped extension is needed. Prefer reuse.

P6 — Competitive benchmark & falsification harness

Create a reproducible comparison harness against the closest relevant classes:

  • Harden-Runner
  • cicd-sensor
  • Tetragon
  • Falco/Tracee only for overlapping propositions

Same pinned workloads and declared scenarios.

Measure separately:

  • setup/privilege friction;
  • observation coverage for declared propositions;
  • authority/completeness;
  • deterministic reproducibility;
  • false PASS;
  • false REVIEW / false incompleteness;
  • overhead;
  • privacy/data captured;
  • CI integration friction;
  • attestation/provenance output;
  • failure transparency.

No composite “winner score”. Publish factual matrix and retained failures only.

P7 — Platform/CI expansion only after semantic gates

Candidate order:

  1. Linux arm64 if backend/evidence parity can be proved;
  2. GitLab CI adapter using same semantics;
  3. self-hosted CI packaging;
  4. additional OS families only under separate evidence contracts.

Do not trade semantic confidence for superficial platform count.

P8 — External validation program

Continue OpenSSF ORBIT engagement under Issue #94.

Target evidence:

  • independent zero-assistance reproduction;
  • substantive architecture criticism;
  • no-fit/redundancy finding;
  • external counterexample;
  • integration/standards guidance;
  • independent real-workload report.

No endorsement/adoption claim unless exactly supported.

Explicit non-goals / anti-imitation rules

Do NOT prioritize:

  • threat-intelligence feeds;
  • malware signatures;
  • generic host EDR dashboards;
  • another broad kernel event collector merely for feature parity;
  • opaque ML anomaly scoring before deterministic semantics mature;
  • automatic baseline poisoning/approval;
  • payload/file-content/secret collection by default;
  • cross-platform expansion that weakens evidence contracts.

Release gate for next substantive alpha

A future alpha beyond current alpha.4 should require a bounded package of accepted evidence, not calendar pressure.

Minimum candidate gate:

  1. Issue Post-M9 — eBPF completeness closure & same-workload value requalification gate #85 E4 formally closed;
  2. semantics-v3 design accepted or explicitly deferred with rationale;
  3. at least one measurable false-review/nondeterminism improvement OR a stronger proposition-authority path;
  4. no regression in current fail-closed semantics;
  5. attestation/interoperability direction proven by executable prototype;
  6. zero-assistance public rehearsal PASS;
  7. independent red-team review survives bounded claims.

Competitive success definition

Success is NOT “more events than Tetragon” or “more detections than Harden-Runner”.

Success is establishing ExecSurface as the most rigorous practical layer in its chosen niche for:

accepted runtime behavior -> authority-aware evidence -> deterministic drift -> explicit verification -> attestable result

with collectors treated as interchangeable evidence sources only where equivalence is actually proved.

Activity

  1. aetherxeg-source commented on Sep 28, 2026

    @aetherxeg-source
    aetherxeg-source
    on Sep 28, 2026 – with ChatGPT Codex Connector
    ContributorAuthor

    Execution freeze — post-alpha.4 development isolation is now active

    The competitive-development plan is now operationally frozen, not just conceptual.

    Dedicated development branch

    development/post-alpha4-behavioral-integrity

    Created from public main at:
    54fe0609ca3c516cfc42603a9292c5ece68e407f

    Program/protocol commit on that branch:
    c70a92694094e28c628f428cc5dd02f9e6064109

    Canonical development protocol:
    docs/development/POST_ALPHA4_DEVELOPMENT_PROGRAM.md

    Public-release isolation

    The published product remains:

    • v0.1.0-alpha.4
    • immutable release source 48e0b9a0707553349e75e97a9dfa096d13f9ab5d
    • stable Action channel AETHERXGLOBAL/execsurface@v0.1
    • native ptrace public/default correctness-reference observer

    Post-alpha.4 runtime/observer/semantics experiments MUST NOT be developed directly on main, MUST NOT retag or rewrite alpha.4, and MUST NOT move the stable @v0.1 channel without a separately closed release/promotion gate.

    Documentation-only maintenance for the already-published product may continue on main when needed, but substantive next-generation development is isolated on the dedicated branch.

    Team/work-system freeze

    Every material gate retains three fixed roles:

    1. Innovation Scientist / Systems Architect
    2. Anti-Drift / Scientific Integrity Reviewer
    3. Independent Falsifier / Red Team

    Dynamic specialists are selected per gate from kernel/eBPF, formal semantics, CI/CD supply-chain security, provenance/attestation, benchmarking/reproducibility, filesystem/process/network identity, nondeterminism/build systems, privacy/data minimization, DX/platform integration and OpenSSF interoperability.

    Execution rule:
    QUESTION -> HYPOTHESIS -> PREDECLARED TEST -> ADVERSARIAL TEST -> EVIDENCE -> DECISION

    No metric/threshold relaxation, no force-push/history rewrite, no deletion/relabeling of negative evidence, and no reopening of closed milestones without a new falsifying result.

    Gate order

    The development branch follows the frozen sequence:

    • P1: close Issue Post-M9 — eBPF completeness closure & same-workload value requalification gate #85 E4;
    • P2: Execution Surface Semantics v3;
    • P3: legitimate variance / nondeterminism model;
    • P4: backend adapter / proposition-authority architecture;
    • P5: in-toto/SLSA/Sigstore interoperability;
    • P6: competitive falsification harness;
    • P7: platform/CI expansion only after semantic gates;
    • P8: external validation via OpenSSF/independent evidence.

    Promotion rule

    The development branch is NOT bulk-merged to main merely because work progresses. Promotion is by narrow reviewed PRs, each tied to an accepted gate, reproducible evidence, preserved failures, compatibility/migration analysis and independent red-team review.

    A future public alpha is authorized only after the release gate in the development protocol closes. Until then, v0.1.0-alpha.4 remains the public release and all post-alpha.4 advancement is research/development evidence only.

  2. aetherxeg-source commented on Sep 30, 2026

    @aetherxeg-source
    aetherxeg-source
    on Sep 30, 2026 – with ChatGPT Codex Connector
    ContributorAuthor

    P4 scientific closeout completed on development/post-alpha4-behavioral-integrity.

    Evidence-led outcomes:

    • P4_PTRACE_ADAPTER_CONTRACT_ESTABLISHED_BOUNDED
    • P4_EXTERNAL_IMPORT_PATH_ESTABLISHED_BOUNDED
    • P4_BACKEND_EQUIVALENCE_NOT_ESTABLISHED
    • no second backend justified by the three tested P4-B gaps

    P4-C accepted source: 7db008ad64aad0782b7438ea8f225739ab942fc5
    Run: 36739233309 / Job: 109968900093
    Evidence artifact: 11110450417
    Artifact digest: sha256:3aaaf91a7ea3f6dc28636bd864be7a227b3129545a39d885f305f30f7bcda043
    Frozen P4-C corpus: 14/14 PASS; P4-B cross reproof 12/12; B1/B2/B3 live 10/10 each; Semantics v3 7/7; public M11 6/6.

    Closeout ledger: docs/development/P4_CLOSEOUT.md
    P4-C result ledger: docs/development/P4_C_EXTERNAL_TRACE_IMPORT_RESULT.md

    Public alpha.4/stable channels remain immutable; no public integration or release promotion is authorized.

    P5 Runtime Attestation / Provenance Interoperability is now the active scientific phase. Initial standards review is reuse-first: in-toto Runtime Trace v0.1, SCAI v0.3, SVR v0.2, SLSA v1 provenance (current SLSA spec v1.2), and GitHub/Sigstore attestations. No new predicate will be introduced unless a preregistered representation-gap test proves existing predicates insufficient.

  3. aetherxeg-source commented on Sep 30, 2026

    @aetherxeg-source
    aetherxeg-source
    on Sep 30, 2026 – with ChatGPT Codex Connector
    ContributorAuthor

    P5 execution update — 2026-09-30

    P5-A2 is scientifically closed as P5_VERIFICATION_ATTESTATION_BINDING_ESTABLISHED_BOUNDED (research/development only; no public integration/release authorization).

    Accepted source: 63b9a9fcde16d9dbb66aa484e73a1e2338afd6cb
    Decision record commit: 8ef427a57d085f6cc6d45a0beef9599c5c4f002f
    Full gate run: 36750284527 — PASS
    Evidence artifact: 11113484839 (p5-a2-36750284527-1)
    Artifact SHA-256: 45f786ed36ca399ee35d595a5a38997a517edef088ff13764c8831b2da180984

    Same-run evidence passed: original A2 14/14; verifier-identity addendum 4/4; A1 12/12; A0 18/18; critical P4 authority/live regressions; Semantics v3; public M11. Historical identity-regression and formatting failures remain retained in the decision record.

    Next frozen program step: P5-A3 SLSA provenance binding. A3 will use an exact commit/blob-pinned upstream SLSA v1 provenance statement or generated fixture and will not infer SLSA level from provenance presence.

  4. aetherxeg-source commented on Sep 30, 2026

    @aetherxeg-source
    aetherxeg-source
    on Sep 30, 2026 – with ChatGPT Codex Connector
    ContributorAuthor

    P5-A3 closeout — 2026-09-30

    Decision: P5_A3_SLSA_PROVENANCE_BINDING_PASS_BOUNDED (research/development only).

    Accepted source: d0ee3d0d1bd5a491ddc5796be08458026e1cbd8a
    Decision record: docs/development/P5_A3_SLSA_PROVENANCE_BINDING_DECISION.md at commit 19624d64de370a90323f14d87cbaf59c3322108f
    Full gate run: 36753014493 — PASS
    Evidence artifact: 11114644663 (p5-a3-36753014493-1)
    Artifact SHA-256: 050205a25cde61b66a443cd35dabded60ab909e91d487146c0efb2bdb4a4ed67

    Pinned upstream SLSA fixture: slsa-framework/slsa-verifier@30d0be3bbab553fc51557377baba2f7572dfc212, path verifiers/internal/gcb/testdata/v1.0-gcloud-container-github-single.json, Git blob 7102e40887a758e01184ac79ca10cb34b7281627.

    Same-run evidence passed: A3 12/12; A2 14/14; A2 verifier-ID addendum 4/4; A1 12/12; A0 18/18; critical P4 authority/live gates; Semantics v3; public M11; immutable alpha.4/stable boundaries.

    Retained failures include static/compile harness failures and run 36752814588 (0/12 due fixture path resolution before any assertion), explicitly classified and preserved in the decision record.

    Next frozen step: P5-A4 signed GitHub/Sigstore attestation prototype. A valid signature will not be allowed to upgrade ExecSurface authority/completeness/verdict.

  5. aetherxeg-source commented on Sep 30, 2026

    @aetherxeg-source
    aetherxeg-source
    on Sep 30, 2026 – with ChatGPT Codex Connector
    ContributorAuthor

    P5 closeout completed — bounded research decision recorded.

    Decision: P5_EXISTING_STANDARDS_COMPOSITION_SUFFICIENT_BOUNDED
    Retained A5 decision: P5_A5_CROSS_ATTESTATION_GRAPH_PASS_BOUNDED

    Closeout source: c3caa857581d6370008fe30779bd5fd084b28751
    Closeout run: 36760093433 — SUCCESS
    Closeout artifact: 11118009105 (p5-closeout-36760093433-1)
    Artifact digest: sha256:94312e440d6873bb3577ada92e3e0f13b52d103ccf63db962c8d367a8c52a815

    Corrected A5 evidence retained:

    • run 36758531046, 12/12 PASS
    • artifact 11117568130
    • digest sha256:d1ba8f88a42b0b3f23880445f6063735cef99b306861241ed2e04cdeee57d343

    Historical scientific counterexample evidence also retained unchanged:

    • run 36755962689, 8/12
    • artifact 11116268894
    • digest sha256:97aa4db53e27355d8b2d62e86d6f1ba5c2ee6d7d6e10607b965e4730fa878314

    Scientific boundary: no demonstrated need for a new custom predicate/schema in the bounded P5 experiment. This is not a universal interoperability claim and does not authorize public integration or release promotion.

    Decision record: docs/development/P5_CLOSEOUT_DECISION.md
    Next phase: P6 Competitive Falsification, proposition-level factual comparison only; no scalar winner score.

  6. aetherxeg-source commented on Oct 1, 2026

    @aetherxeg-source
    aetherxeg-source
    on Oct 1, 2026 – with ChatGPT Codex Connector
    ContributorAuthor

    Post-alpha.4 promotion sync — A1 Semantics v3 repaired candidate requalified

    A1 Semantics v3 promotion candidate has completed repair, independent falsification, and candidate-branch requalification.

    Historical broken source 5079a990b924d8ccd7ac6414f8a9a2571b54e240 and its material-gap evidence remain retained.
    Repaired source: 0200f09557906118dd0e96f8a5a73aa4f5c4a9cc.
    Independent review: run 36853136002, F01-F12 12/12 PASS.
    Actual candidate gate: run 36853799391 at 3522cc87b5b23a42e8009e779549e5dcca95da5b — SUCCESS across Semantics, P3, P4, P5, M11 and M12 on Ubuntu 22.04/24.04.
    Artifact 11157785131, digest sha256:76a5a8e2a1f0ca01e8d291fda409749b57feda9fcc892bb5a6a6cfdf391885b3.

    Current classification:
    POST_ALPHA4_PROMOTION_A1_SEMANTICS_V3_REQUALIFIED_BOUNDED_CANDIDATE

    Boundary unchanged: internal next-gate eligibility only; no public promotion, no main merge, no stable-tag movement, and no P8 closure. Public alpha.4 remains 48e0b9a0707553349e75e97a9dfa096d13f9ab5d.

  7. aetherxeg-source
    added
    research
    on Oct 6, 2026

  8. aetherxeg-source commented on Oct 8, 2026

    @aetherxeg-source
    aetherxeg-source
    on Oct 8, 2026 – with ChatGPT Codex Connector
    ContributorAuthor

    Closing this post-Alpha.4 parent advancement program as a completed historical program.

    Its original product-transition objective has been superseded by later evidence-gated programs and the published stable v1.0.0 line. Closing the parent does not close or relabel the remaining independent research/evidence tracks.

    Still-open specialized tracks retain their own authority, including:

    Historical P2–P7 decisions, negative evidence, candidate qualifications and no-fit/defer boundaries remain retained. This is repository-state hygiene, not a rewrite of the research record.

  9. aetherxeg-source
    closed this as completedon Oct 8, 2026
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Labels

    research

    Type

    No type

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions