Veridict

The verdict that survived verification. | Türkçe

Veridict is a protocol and reference implementation for auditing AI with AI when humans no longer can. An append-only, hash-chained evidence ledger records every step of an audit: falsifiable claims, machine verifiers, a blind heterogeneous jury, third-party watchers, and a signed certificate that anyone can verify offline — no trust in the auditor required.

The founding premise: as AI output outgrows human review capacity, the human role is not code reviewer but risk owner. Veridict is built around that transformation — the machine owns the verdict of intelligence, the human owns the verdict of responsibility.

In 30 seconds

An agent says it did the work — deployed, paid, refactored, passed. "It says so" is a claim. Veridict turns it into proof: a signed certificate whose content hash, signature and timestamp anyone can recompute from the file alone — no ledger, no network, no account, no trust in the auditor.

pip install veridict-standard

# an agent issues a signed receipt for completed work
export VERIDICT_HOME=$PWD/.veridict
veridict receipt issue --achievement "deployed api v2 to staging" \
  --actor agent-7 --evidence "gh run 8812 passed"

# the other side verifies it — independently, from the file alone
veridict verify .veridict/receipts/<cert_id>.json
# → {"valid": true, "signature_valid": true, "content_hash_valid": true, ...}

# tampering breaks it: one bit in the body breaks the signature,
# one digit in cert_id breaks the content hash; revocation is an
# append-only ledger entry, never a rewrite

Two document kinds, one verifier:

Receipts (veridict receipt) — signed proof-of-done for one achievement: the smallest unit a counterparty actually needs. Exposed to agents over MCP (mcp/): issue / verify / revoke / list.
Audit certificates (veridict audit) — a full adjudication over a machine-verifiable evidence ladder (tests, static analysis, a blind heterogeneous jury, watchers), replayable offline against the ledger.

Both are append-only, hash-chained, and fail closed: no evidence → INCONCLUSIVE, never a clean bill.

Why an audit ledger

No silent passes. No evidence → INCONCLUSIVE, flagged — never a clean bill.
Tiers, not vibes. Evidence has a fixed rank order: what a machine can check directly (W1a) beats what can be reproduced statistically (W1b), and both beat what a jury of models believes (W2/W3) — the ladder never lets doctrine overturn machine evidence, and W3 alone never verifies.
Identity is binding. Every entry's hash binds its author; watcher manifests are signed, W1a is structurally impossible for them; a calibration ledger discounts producers whose claims were contradicted.
Verify, don't trust. Certificates replay offline against the ledger. examples/spec_verifier.py is a verifier written from the standard alone (zero imports of this codebase) that reaches the same verdicts from the published test vectors.

Quickstart

pip install veridict-standard
# PyPI name is `veridict-standard` (the bare `veridict` on PyPI is a
# DIFFERENT project — not ours). Same package from the repo, if preferred:
#   pip install -e . && pip install pytest

# first run: who am I, what can I do
veridict --version && veridict --help

# the system audits itself: ledger → claims → jury → certificate → offline verify
python scripts/dogfood.py

# verify a certificate offline (the property third parties care about)
# ledger + cert ship as release assets: gh release download --latest
veridict verify --ledger dogfood_ledger.jsonl --cert dogfood_cert.json

# or verify ANY certificate from the file alone — no ledger needed
# (the cert embeds the signer's public key and a signed timestamp)
veridict verify dogfood_cert.json

# issue a signed proof-of-done receipt for one achievement, and list them
export VERIDICT_HOME=$PWD/.veridict
veridict receipt issue --achievement "cut release v1.1" --actor agent-7
veridict receipt list

# your own repo, three commands: see "Audit your own repo" below

# regenerate the standard's conformance test vectors (deterministic)
python scripts/build_test_vectors.py

# measure GATE latency against the design budget (§6.5: p95 < 30 min)
python scripts/measure_latency.py

Receipts (evidence over claims)

The self-audit badge above is regenerated by CI on every push from the actual dogfood certificate (never hand-drawn) — and it links to that certificate, which you can verify offline yourself: fetch the ledger and certificate from the latest release (gh release download or the assets list), then run veridict verify --ledger dogfood_ledger.jsonl --cert dogfood_cert.json. A failing audit turns it amber or red — worst-verdict-wins.

Check	Result
Test suite	1.873 passed, 3 skipped rc=0 (G49-ölçümü 2026-10-22, `pytest tests/`, 1.876 collected, 817s; test_dogfood-ikilisi-önceki-koşumda-flake-idi-bu-koşumda-yeşil, `test_dogfood`-ikilisi-bu-makinede-önceden-de-KIRMIZI ( 2026-10-22-stash-ölçümü), LLM-olmayan-flake; skips are the real-LLM/MCP/soak suites and the two Tamga-parity tests, which skip where the mesh sibling is absent) — both invocation styles, Python 3.12–3.14 in CI
Self-audit	valid certificate, risk `low`, GATE not blocked
Offline replay	`veridict verify` rc 0 on the dogfood certificate
Standalone verification	receipt/cert verified from the file alone — content hash + signature + signed timestamp; new suites cover issue/verify/revoke/list, tamper, custody and the MCP protocol surface
Canary (scripted jury)	23 catches / 3 honest misses / 0 false positives across 26 defect classes — measures a hand-authored refutation table, not model capability; labeled as such
Canary (real LLM, lower bound)	2/25 classes caught, 0 false positives, by local qwen2.5:3b + llama3.2:3b — the audit's end-to-end catch rate; the earlier 23/25 figure was inflated by a metric bug that counted coverage-meta-claim refusals as findings; sheets in `docs/notes/`
Judgment sonde (model, decoupled from the ladder)	42/48 correct refutations of defective artifacts by the same 3B jurors (23/24 + 19/24), but 2/4 false refutations on clean code — the model sees most of the defects; §5.3 rule 2 keeps those refutations from turning a W1a-supported verdict, and the false-refutation rate is the measured justification for that rule
Tamper soak	1500 mutated ledgers, 5 seeds → 100% detected, 0 silent passes
Spec parity	reference verifier ≡ spec-only verifier on 8 failure modes

CI runs the suite, the dogfood self-audit, and the offline replay on every push — the receipts are regenerated, not narrated. The 1500-ledger tamper soak runs nightly (the push run exercises a 150-ledger slice).

Register a watcher in 3 commands

veridict registry init --registry r.jsonl --key-out r.key.json
veridict registry register --registry r.jsonl --manifest examples/manifests/example-security.json --key-file r.key.json
veridict registry index --registry r.jsonl --out index.json

Then audit with the external registry as authority: veridict audit ... --registry r.jsonl — revoked watchers are refused (§6.6). See CONTRIBUTING.md for writing your own.

Audit your own repo in 3 commands

The fastest way to see Veridict work on your code: write a one-file task manifest naming your checkout and the claims you want evidence about, run the audit, verify the certificate offline — then link the pair from your README. A runnable demo lives in examples/run_audit.py.

# 1) a task manifest (see examples/ and TaskManifest in veridict/schemas.py)
cat > task.json <<'EOF'
{"task_id": "my-audit", "artifact_path": "/path/to/your/repo",
 "actor_identity": "you", "intent_lines": ["DOCTRINE: modules are idiomatic python"],
 "criticality": [], "has_existing_tests": true, "pytest_args": []}
EOF

# 2) audit → ledger + certificate
veridict audit --task task.json --mode HYBRID \
    --ledger my-ledger.jsonl --cert-out my-cert.json

# 2b) or WATCH a live audit as it writes (streaming transport, §10.3):
veridict watch --ledger my-ledger.jsonl --forever

# 3) verify offline (anyone can; no trust in the auditor required)
veridict verify --ledger my-ledger.jsonl --cert my-cert.json

Want the reference implementation to run the audit for you and publish the certificate? Open an Audit my repo issue — the certificate that comes back is evidence, not endorsement (§13): anyone can replay it offline against the published ledger, and a REFUTED claim says exactly which evidence refuted it.

Audit on every push (GitHub Action)

Run Veridict inside your own CI without installing anything locally — two equivalent styles:

# style 1 — the composite action (root action.yml; appears on the
# GitHub Marketplace as "veridict audit")
jobs:
  audit:
    runs-on: ubuntu-latest
    steps:
      - uses: goun7/veridict@v1
        with:
          intent: "DOCTRINE: modules are idiomatic python"
          mode: HYBRID          # GATE / WATCH / CERTIFICATE also available
          fail-on-refuted: false # flip to true once you trust the audit

# style 2 — the reusable workflow (same engine, same inputs)
  audit-wf:
    uses: goun7/veridict/.github/workflows/veridict-audit.yml@v1
    with:
      intent: "DOCTRINE: modules are idiomatic python"
      mode: HYBRID

The run uploads veridict-audit artifacts (ledger + certificate) — download once, verify forever: veridict verify --ledger veridict-ledger.jsonl --cert veridict-cert.json. See the workflow file for all inputs — notably anchor: rekor (pin the checkpoint to the public transparency log; the verify step checks the sidecar automatically) and actor-family (declare the authoring model's family and same-family jury opinions are excluded — the self-preference guard).

By default the jury runs on scripted stubs (loudly marked in the report — stubs never pretend to be judges). To bring a real model jury, the action reads the same environment as the CLI — step-level env: flows into composite steps:

      - uses: goun7/veridict@v1
        env:
          VERIDICT_JURY_URL: https://api.openai.com/v1      # any OpenAI-compat
          VERIDICT_JURY_KEY: ${{ secrets.JURY_KEY }}
          VERIDICT_JURY_MODEL: gpt-4o
          VERIDICT_JURY_URL2: https://api.anthropic.com/... # a 2nd FAMILY
          VERIDICT_JURY_KEY2: ${{ secrets.JURY_KEY2 }}      # (required: juries
          VERIDICT_JURY_MODEL2: claude-sonnet-4-5           #  need ≥2 families)
        with:
          intent: "DOCTRINE: ..."

Receipts for agent work (proof-of-done)

An audit certificate is the heavy instrument — evidence ladder, jury, watchers. Most of the time a counterparty needs the light one: did the agent do the thing it said it did, and can I check it myself? That is a receipt.

export VERIDICT_HOME=$PWD/.veridict     # workspace: ledger + issuer key

# issue: achievement + grounds -> signed, self-contained receipt
veridict receipt issue --achievement "paid invoice 1042" \
  --actor agent-pay --evidence "bank confirmation tx 77f3"

# verify standalone (file alone) or fully (with the issuer's ledger)
veridict verify $PWD/.veridict/receipts/<cert_id>.json
veridict receipt verify --cert .veridict/receipts/<cert_id>.json \
  --ledger .veridict/ledger.jsonl

# revoke (append-only: the receipt file is never modified) and list
veridict receipt revoke --cert .veridict/receipts/<cert_id>.json \
  --reason "chargeback case 91"
veridict receipt list --active-only

veridict verify dispatches on certificate_type: a receipt proves chain integrity, issuance and revocation over a ledger, an audit certificate adds the claim replay — the top-level command handles both. Revoke accepts the receipt file (--cert) or a bare --cert-id.

Every receipt carries a content_digest recomputable from its own fields, an Ed25519 signature over the canonical body verified against the public key embedded in the receipt, and an issued_at inside the signed body. The issuer's private key stays local (mode 0600, never emitted); only its public half ships. Revocation appends a receipt.revoked entry to the same hash-chained ledger that issued it — the signature stays intact and the disavowal stays auditable. See mcp/README.md for the agent-facing surface and docs/arastirma/ for the academic map of the proof gap this closes.

From certificate to settlement

A certificate proves what happened. It does not by itself authorize payment — a settlement layer still has to decide whether this verified work qualifies, and that decision has to fail closed on its own. payment — a settlement layer still has to decide whether this verified work qualifies, and that decision has to fail closed on its own. veridict settle is that boundary:

# certificate already verified offline (veridict verify)
veridict settle --cert dogfood_cert.json --out claim.json   # derive
veridict settle --cert dogfood_cert.json --claim claim.json  # reconcile

The claim is derived only from certificate fields — accepted claims, jury families, risk level — under a policy that is itself part of the claim's digest, so the caller cannot swap in a more permissive policy after seeing the verdict. Reconciliation never trusts the presented claim: it recomputes it from the certificate, so inflating accepted_claims (or flipping valid, or presenting a claim minted against a different certificate) is refused. Exit code 0 means the claim holds; non-zero means it does not — a pipeline that ignores the report and pays on any rc has a bug, and the tool will not help it.

If your workflow keeps a dossier (a machine-readable decision record) rather than a bare certificate, veridict resolve is the same boundary for that shape: it records a human decision on an already-issued dossier into the ledger — which dossier, which of the four decisions, who decided, and the risk note they accepted under. A resolution for a dossier that was never issued is refused, so no human authority can be fabricated.

veridict resolve --ledger ledger.jsonl --dossier doss-2026-10 \
                 --decision accepted --decided-by oncall@example.com \
                 --note "tutor reviewed the diff"

This is deliberately a claim generator, not a payment executor. It answers "was provably-done work done", not "should money move". Keeping those separate is what stops a dropped or replayed settlement message from becoming a silent release.

Interoperability

Positioning vs runtime governance (e.g. microsoft/agent-governance- toolkit): those constrain agents before/during actions; Veridict adjudicates after the fact into a certificate a counterparty verifies without trusting vendor, auditor, or runtime. Directions of flow differ — a deployment can run both, and the runtime's own action logs are exactly the evidence class an audit anchors on.

Two ways a Veridict verdict reaches tooling that never heard of us:

External anchoring — veridict audit … --anchor rekor (or in CI, the anchor: rekor action input; or veridict anchor publish) pins the certificate's checkpoint to the Sigstore Rekor public-good transparency log. The anchor proves existence at time T; authority stays with the certificate's own key. Anyone holding cert + ledger
- anchor sidecar checks it fully offline: veridict verify --ledger L --cert C --anchor A (pinned Rekor key, no network, no trust in us). The v1 release's dogfood certificate is anchored live — see docs/receipts-anchor-v1-dogfood.json.
Mesh-local anchoring — veridict audit … --anchor tamga --anchor-ledger mesh.jsonl pins the same checkpoint binding (same bound_fields, same digest) into a Tamga-grammar hash-chained ledger — the mesh's own permanent proof-anchor layer, no third party and no network at all. Both transports commit to identical bytes, so a relying party can check either surface. The record follows Tamga's chain rules byte-for-byte (h = sha256(prev + jcs(rec)), 1-based seq, optional ed25519 node-cosign), which means Tamga's own ledger-verify accepts a ledger Veridict wrote — verified by a parity test that shells out to Tamga's verifier rather than asserting agreement. veridict verify … --anchor A dispatches on the sidecar's anchor_version, so both anchors verify through one command.
SLSA export — veridict export --format vsa projects a certificate onto a SLSA v1.2 Verification Summary Attestation (in-toto Statement) so existing supply-chain policy engines can consume the verdict. The VSA is a lossy projection: it names the certificate by digest (inputAttestations) and embeds the full binding as a spec-sanctioned extension field; the certificate remains the authoritative object. Optional --sign KEYFILE wraps it in a DSSE envelope under the issuing key.
SPDX 3.0.1 export — veridict export --format spdx emits an SPDX 3.0.1 AI-profile JSON-LD document, validated in CI against the pinned official schema. The audited artifact rides as an ai_AIPackage (identity, digest, supplier); every audit semantic — claims, evidence, jury composition, risk, Rekor anchor — rides on core Annotation/Relationship/ExternalRef elements. Deliberate limit: the AI profile models AI systems, not audit verdicts, so no ai_* field is made to carry a meaning the spec does not give it (our risk_level never occupies ai_safetyRiskAssessment — there is a test locking that refusal in). The certificationReport externalRef binds the authoritative certificate by canonical digest.

Public site

The standard and the design document are readable (and linkable) at https://goun7.github.io/veridict/ — deployed from main on every push. The launch article — "Agents already found their forum. We built the better one." — is on dev.to.

Documents

Standard (normative draft): docs/specs/2026-09-10-veridict-standard-v1.0.md — §14.2 carries errata; errata proposals are a first-class issue template
Standard v1.1 delta (draft): docs/specs/2026-09-12-veridict-standard-v1.1-delta.md — ratifies the WATCH transport latency sentence (D10); read on top of v1.0
Design document (founding paper): docs/specs/2026-09-09-veridict-design.md
Architecture: ARCHITECTURE.md — module map to standard sections
Commercial model: docs/commercial-model.md — open-core model, growth levers, anti-corruption guardrails
Landing page prep: docs/landing.md — 30-second pitch, feature map, tiered pricing ($29–99/mo, the core never paywalled)
Academic research: docs/arastirma/README.md — the proof gap mapped to six 2025–2026 papers (real links), W3C VC v2.0 / Data Integrity alignment, and the EAS on-chain model
MCP server (receipts): mcp/README.md — stdio install for Claude Desktop / Cursor / Cline; tools issue / verify / revoke / list; registry definition in mcp/mcp.json
Contributing / Governance / Security: CONTRIBUTING.md · GOVERNANCE.md · SECURITY.md
Building your own verifier (independent of this codebase, issue #1): docs/verifier-onboarding.md — closure kit: trusted-artifact table, byte-level traps list, one-command parity harness, proofs-as-behavioral-spec

Status & roadmap

Formal core (Sep 2026): every structural layer of the core is machine-checked in Lean (v4.34, kernel-only). The ladder (proofs/ladder/Ladder.lean) proves seven invariants over evidence lists of arbitrary length — fail-safe-empty, no-silent-pass, W1a-decisiveness, doctrine-can-never-topple, escalation conditions, split visibility, meta-budget — and a generated oracle (TruthTable.lean, digest-tied to the bounded receipt) re-verifies the model against all 28,080 sampled cases on every CI run; its first build caught a real model-vs-code divergence and the model moved. The ledger chaining (proofs/ledger/Chain.lean) proves six theorems over chains of arbitrary length — writer/reader agreement, prefix-closure, payload-edit and honest-remint tamper detection (the one collision hypothesis is stated, not smuggled), genesis-rooting. The anchor sidecar (proofs/anchor/Anchor.lean) proves nine theorems about certificate⇄CT cross-verification — its headline (A4): with the wrong binding, valid ECDSA signatures cannot rescue a sidecar; structural rejection is unconditional. ECDSA itself stays honest TCB: no structural proof discharges a computational assumption, so it enters the model as a disclosed opaque boolean rather than being silently "proven". Trust placement (kernel decide; disclosed native_decide for the oracle lane only) and the measured axiom footprints (C1 and I1/I7 axiom-free; the rest [propext]/Quot.sound-maximal) are disclosed in the file headers and re-printed by proofs.yml on every run — per house doctrine: receipts, not narrative.

Phase 1 (core) and Phase 2 (watcher layer) are implemented with end-to-end receipts in the dogfood run. Phase 3 (platform + standard) substrate is in place: spec draft, conformance kit (C1–C13), marketplace index, test vectors, spec-only verifier. The remaining exit criteria are external by nature:

an independent verifier implemented from the spec by someone else,
a first external production deployment,
≥10 active watcher manifests on the marketplace.

Roadmap decisions are tracked in the issue tracker — the post-launch expansion plan is Roadmap v2 (#10) (launch-hygiene fixes first, then standard maturity, federation, and formal verification); the commercial model is documented in docs/commercial-model.md.

Akademik zemin (2024-2026)

Delilix tarzı bir not: aşağıdaki her madde, bu README'ye eklenmeden önce HTTP 200 ile fetch edildi ve başlık+yazar birebir karşılaştırıldı — FP0 (uydurma yok). Veridict'in dayandığı alan — DSSE/in-toto tarzı imzalı attestation, yazılım tedarik zinciri güvenliği, doğrulanabilir iddialar (verifiable claims), SLSA build provenance, ajan harness'in yazdığı kaydın güvenilirliği ve imzalı manifest denetimi — üzerine 2024-2026 yayınları. Tüm bağlantılar canlı arXiv sayfalarıdır.

[1] DSSE İmzalı, Değişmez ve Yeniden Oynatılabilir Ajan Kanıtı — NovaFabric: Tamper-Evident, Replayable Evidence for Autonomous AI Agent Runs — Ardebili, arXiv 2026. Otonom bir ajan çalışmasını, hiçbir ajan mantığını değiştirmeden, DSSE imzası + RFC 3161 zaman damgası + Merkle log ile mühürlü, taşınabilir ve yeniden oynatılabilir bir "Run Capsule" olarak kaydeder. Veridict'in hash-chained ledger + çevrimdışı doğrulanabilir sertifika yaklaşımının aynı ailesinden, doğrudan DSSE uygulaması. arXiv:2609.12582
[2] Doğrulanabilir Eğitim/Yayın İddiaları için Attestation Geçidi — Attesting LLM Pipelines: Enforcing Verifiable Training and Release Claims — Tan et al., arXiv 2026. Üçüncü parti ağırlıklar, adapter'lar ve dependency'lerden oluşan LLM tedarik zincirinde "data/code lineage, build environment, security scanning" gibi iddiaların kriptografik olarak artifact'e bağlanmamasını bir boşluk olarak tanımlar ve attestation-aware bir promotion gate önerir. Veridict'in "iddia → imzalı, makine-doğrulanabilir kanıt" dönüşümünün ML pipeline genellemesi. arXiv:2603.28988
[3] CI Hatları için Kanıt Tabanlı Provenance Protokolü — An Evidence-driven Protocol for Trustworthy CI Pipelines — Castillo et al., arXiv 2026. Deterministic Build Systems + Trusted Execution Environments kombinasyonuyla CI artifact'leri için kriptografik doğrulanabilir bütünlük, özgünlük ve attestation garantisi sunar; her tüketicinin build'i yeniden çalıştırmasının doğurduğu doğrulama darboğazını çözer. SLSA build provenance hedefinin kanıt-odaklı bir gerçeklemesi. arXiv:2605.21089
[4] SLSA/in-toto Ötesinde Otonom Tedarik Zinciri Savunması — Agentic AI for Autonomous Defense in Software Supply Chain Security: Beyond Provenance to Vulnerability Mitigation — Syed et al., arXiv 2025. Mevcut SLSA, SBOM ve in-toto çerçevelerinin provenance ve izlenebilirlik sağladığını ama üretime gömülü zafiyetleri aktif olarak tespit edip gideremediğini belirtir; LLM tabanlı otonom savunmayla tamamlamayı önerir. Veridict'in kanıt katmanının (provenance) ötesine geçiş ihtiyacını tanımlayan referans. arXiv:2512.23480
[5] Kod Asistanları Provenance Sinyallerini Okuyor mu? — Do AI Coding Assistants Check Before They Install? A Pre-Registered Demand-Side Audit of Trust Signals in the Research Software Supply Chain — Shan, arXiv 2026. SBOM, imzalı release'ler, build provenance attestations ve resmi kanal beyanları gibi makine-okunabilir güven sinyallerinin, paket seçip kuran AI kod asistanları tarafından okunup okunmadığını ölçen önceden kayıtlı bir deney. İmzalı kanıtın üretildiğiyle tüketildiği arasındaki açığı gösterir. arXiv:2609.07754
[6] LLM Ajanlarında Yürütüm Provenance'ı ve Kanıt İzleme Araştırması — From Agent Traces to Trust: A Survey of Evidence Tracing and Execution Provenance in LLM Agents — Wang et al., arXiv 2026. Ajan izlerini (trace) salt gözlemden, süreç düzeyinde hesap verebilirliğin temeli olarak yürütüm provenance'ına yükseltir: her iddianın arkasındaki kanıt, araç çağrılarının gerekçesi ve belleğin kararlara etkisi. Veridict'in evidence ladder'ının akademik zeminini oluşturur. arXiv:2606.04990
[7] Skor Değil, İddia Doğrulaması — Verify Claims, Not Scores: Evidence-Based Verification of Modular Agents — Alzahrani, arXiv 2026. Ajanı toplam bir görev skoruyla yargılamak yerine, her sonucu arkasındaki kanıtla birlikte kaydeden ve destekli / desteklenmiyor / çözümlenemedi / değerlendirilmedi olmak üzere dört verdicttan biriyle etiketleyen bir doğrulama denetimi önerir. Veridict'in fail-closed INCONCLUSIVE verdict felsefesiyle aynı özü taşır. arXiv:2610.01348
[8] Yapısal Ajan Eylemleri için Kaynaklarası Bütünlük Sertifikası — Certified Multi-Source Integrity for Structured Agent Actions — Pandey et al., arXiv 2026. Ajanların ayrıcalıklı, geri alınamaz yapısal eylemlerini (ör. fatura ödeme) bozabilecek belge/araç çıktısı kirlenmesi ve prompt injection'a karşı, bir corruption budget altında eylemin ne zaman güvenle sertifikalanabileceğini tanımlar. Veridict'in imzalı makine-verifier'larıyla aynı "sertifikala, sonra serbest bırak" modeli. arXiv:2609.34245
[9] Harness'in Yazdığı Kayda Auditor Güvenebilir mi? — Hearsay: Can an Auditor Trust the Record a Deployed Agent Harness Writes? — Dai et al., cs.CR, Eylül 2026. On altı dağıtılmış framework'ün hiçbiri "evidentiary" (orada olmamış bir okuyucunun yazana güvenmeden denetleyebileceği) bir kayıt tam olarak yazmıyor. Makalenin ölçtüğü şey ve Veridict'in tezinin deneysel gerekçesi: bir harness'in kendi kaydı üstündeki hash chain, harness'in çalışırken işlediği 28 omission/fabrication'in hepsini geçirir (yani sessizce onaylar); buna karşın harness dışında tutulan ve kayıtla her iki yönde okunan bir append-only log aynı 28'inin de hepsini yakalar. Çözüm "ilk yazana daha güçlü bir mühür değil, ikinci bir yazar"dır. Bu, Veridict'in neden hash-chain ile yetinmeyip ikinci yazarlar eklediğidir: watcher manifest'leri, dışsal anchor (Rekor / Tamga) ve standarttan bağımsız yazılmış verifier — hepsi "kanıtın yazarı kim" sorusunun farklı bir yanıtı. arXiv:2609.32495
[10] MCP Çağrıları için İmzalı Manifest ve Transparency Log — Verifiable Manifest Signing and Transparency Enforcement for Secure MCP-Based LLM Pipelines — Jamshidi et al., cs.CR, Ocak 2026 (v2 Haziran 2026). MCP baseline'ının tool-use manifest'lerini kriptografik olarak doğrulamadığını, tazelik (freshness) denetlemediğini ve bağımsız denetime açmadığını bir boşluk olarak tanımlar; her manifest'i policy-validate + freshness-check + dijital imza + execution öncesi doğrulama + tamper-evident audit evidence'a bağlayan bir uygulama katmanı önerir, kabul edilen çağrıları Merkle transparency log'a yazar ve fail-closed authorization uygular. Veridict'in imzalı watcher manifest'lerinin (registry, §6.6; iptal edilen watcher reddedilir) ve agent'a bakan MCP receipt yüzeyinin (mcp/) akademik zeminidir. arXiv:2601.23132

Navigation Menu

Repository navigation

About

Repository files navigation

Veridict

In 30 seconds

Why an audit ledger

Quickstart

Receipts (evidence over claims)

Register a watcher in 3 commands

Audit your own repo in 3 commands

Audit on every push (GitHub Action)

Receipts for agent work (proof-of-done)

From certificate to settlement

Interoperability

Public site

Documents

Status & roadmap

Akademik zemin (2024-2026)

About

Releases

Packages

Contributors

Languages

Name		Name	Last commit message	Last commit date
.github		.github
adapters		adapters
corpus		corpus
docs		docs
examples		examples
mcp		mcp
proofs		proofs
scripts		scripts
tests		tests
veridict		veridict
watchers		watchers
.gitignore		.gitignore
ARCHITECTURE.md		ARCHITECTURE.md
CHANGELOG.md		CHANGELOG.md
CONTRIBUTING.md		CONTRIBUTING.md
GOVERNANCE.md		GOVERNANCE.md
LICENSE		LICENSE
README.md		README.md
README.tr.md		README.tr.md
SECURITY.md		SECURITY.md
Veridict		Veridict
action.yml		action.yml
dogfood_policy.json		dogfood_policy.json
pyproject.toml		pyproject.toml

Latest commit

History

Folders and files

Repository files navigation

Veridict

In 30 seconds

Why an audit ledger

Quickstart

Receipts (evidence over claims)

Register a watcher in 3 commands

Audit your own repo in 3 commands

Audit on every push (GitHub Action)

Receipts for agent work (proof-of-done)

From certificate to settlement

Interoperability

Public site

Documents

Status & roadmap

Akademik zemin (2024-2026)

About

Topics

Resources

Contributing

Security policy

Stars

Watchers

Forks

Releases

Packages

Contributors

Languages