Files
git.stella-ops.org/docs-archived/product/advisories/20260221 - Four novel, testable moat hypotheses.md

6.0 KiB
Raw Permalink Blame History

Here are four fast, defensible moat experiments you can add to StellaOps and validate in 12 sprints—each with a crisp pass/fail and public references.


1) Functionlevel semantic fingerprints (“semhash”)

Why: resilient artifact identity across rebuilds/optimizations; raises codereuse evasion cost. Signals to measure:

  • ≥80% withinversion function match across gcc/clang and -O0/-O2/-Os;
  • <1% crossproject false positives vs ~10kfunction corpus;
  • Robust to symbolstripping & minor reordering. 2sprint plan:
  • A: Extract perfunction IL via DWARF, normalize, embed/hash; prototype store & query. (Dwarfstd)
  • B: CI job builds 3 variants, computes semhash diffs; publish report. Singlecard experiment: CI asserts ≥80% functions cluster to same semhash; attach diffoscope snapshot for mismatches. (Diffoscope) Grounding: KEENHash (functionaware hashing); DWARF v5 spec; diffoscope latest (v312, 20260206). (arXiv)

2) Deterministic shortrun behavior attestations (sandbox traces)

Why: converts “it executed” into a cryptographically verifiable exploitability signal; complements VEX. Signals:

  • 99% identical syscall/observable trace on repeated runs in hermetic harness;

  • Variance across different inputs;
  • Capture <10s, replay <10s. 2sprint plan:
  • A: Record a containerized microhandler under Firecracker (or gVisor) using rr; wrap trace as DSSE; sign with cosign. (Amazon Web Services, Inc.)
  • B: Verifier replays trace (rr) and checks DSSE + Rekor pointer. (RR Project) Singlecard experiment: Build+run a <1s handler, emit signed trace, store Rekor pointer, run verifier → PASS only if replay+verify succeed. (GitHub)

3) Duallog “twinproof” stitching (multilog anchoring + witness)

Why: attacker must tamper with two independent logs; stronger story for procurement/legal. Signals:

  • Same DSSE digest appears in Rekorv2 and a second appendonly log (e.g., signed Git tag);
  • Consistency/witness checks detect divergence;
  • Measurable increased attack cost vs singlelog. 2sprint plan:
  • A: Write attestation to Rekor v2 and to signedtag Git “proofs” repo; record both pointers in DSSE. (Sigstore Blog)
  • B: Verifier fetches Rekor inclusion proof + Git tag sig; PASS only if both validate. (Sigstore Blog) Singlecard experiment: Produce one DSSE, anchor to both logs, run verifier → PASS iff both proofs verify. (Sigstore Blog)

4) Attestable runtime canary beacons

Why: lowvolume evidence that a specific artifact actually ran in a real env—without shipping raw telemetry. Signals:

  • Signed beacon ⟨artifact_id, nonce, timestamp⟩ verified against cosign key + Rekor pointer;
  • 90% beacon verification rate in staged infra;

  • Origin/IP/arrival cadence provide internal execution evidence. 2sprint plan:
  • A: Embed a oneshot beacon emitter (Go) at entrypoint; post DSSE to a small collector over mTLS; sign + anchor. (GitHub)
  • B: Collector verifies sig + Rekor, stores events; expose query API; (optionally align with OTel signals). (Canarytokens) Singlecard experiment: Run the binary once in staging → collector shows verified DSSE + Rekor pointer. (GitHub)

Where this slots into StellaOps

  • Evidence Locker: store semhashes, traces, dualanchors, and beacons as firstclass DSSE records.
  • Attestor: add “sandboxtrace.verify()” and “twinproof.verify()” checks to your policy engine.
  • AdvisoryAI: surface investorfriendly KPIs: semhash stability %, tracereplay PASS rate, dualanchor PASS rate, beacon verification %.
  • Release Orchestrator: make these jobs optional gates per environment.

Acceptance criteria (quick)

  • Semhash: ≥80% stable across two compiler flags; <1% FP vs 10k corpus. (arXiv)
  • Sandbox traces: rr replay PASS + DSSE verify + Rekor pointer in CI. (RR Project)
  • Twinproof: verifier fails if either Rekor or Git proof missing. (Sigstore Blog)
  • Beacons: ≥90% verified beacons from staged runs. (Canarytokens)

If you want, I can draft the four CI job cards (Makefile targets + sample DSSE predicates + policy checks) sized for a twosprint push.