ABI Claims Validator
Agent-side procedure to validate documentation, collateral, and public claims
against the repository source of truth (Cargo.toml, src/, contract tests,
docs/contracts/external-claims-audit.mdx). Use before publishing any external
artifact. There is no /abi-claims-validator binary or slash command — follow
the steps below (or dispatch the external-claims-auditor subagent).
For editing docs to stay claim-honest, prefer abi-doc-claims-sync.
Procedure (agent actions)
scan
- Open the target tree (default
docs/, plus README/AGENTS/CHANGELOG/walkthrough as needed).
- Search for claim patterns (table below) with ripgrep or equivalent.
- For each hit, cross-reference source/tests/
docs/contracts/external-claims-audit.mdx.
- In
--strict mode, treat any unproven number or deployment claim as a fail.
Detects:
- Performance numbers (QPS, latency, throughput, accuracy)
- Deployment claims (Kubernetes, H100, multi-node)
- Implementation language claims (Swift, Python, TensorFlow, PyTorch)
- Security claims (AES-256, RBAC, certifications)
- Distributed claims (sharding, multi-host)
- Benchmark comparisons (SQuAD, CodeSearchNet, GPT)
- Energy/GPU speedup claims
check-claim
Given one claim string, return SUPPORTED | UNSUPPORTED | PARTIAL plus the
evidence path (source, test, or audit section). Prefer executable evidence over prose.
audit
Produce a short markdown report of flagged claims vs evidence. The durable
ledger lives in docs/contracts/external-claims-audit.mdx — update that file
when the audit changes reality, do not invent a parallel truth source.
Source of Truth Hierarchy
- Executable config:
Cargo.toml, Cargo.toml.zon, rust-toolchain.toml
- Source implementation:
src/, tests/golden/
- Contract tests:
tests/golden/ (surface, mcp_tools, feature_modules, plugin_registry, public_docs)
- Explicit claim boundary:
docs/contracts/external-claims-audit.mdx
- North-star mapping:
docs/spec/wdbx-north-star.mdx §2 (Current/Partial/Proposed)
- Prose docs:
README.md, docs/*.mdx — lowest priority, must reconcile upward
Common Claim Patterns to Flag
| Pattern | Likely Status | Check Against |
|---|
| `\d+\s*(QPS | req/s | throughput)` |
\d+\s*ms\s*latency | ❌ Unsupported | ../wdbx/crates/abi-wdbx/tests/ artifact |
| `distributed | sharding | multi-host |
| `AES | RBAC | encryption |
| `Swift | Python | TensorFlow |
| `H100 | A100 | InfiniBand |
| `SQuAD | CodeSearchNet | GPT |
| `energy | kWh | efficiency |
Replacement Wording
From docs/contracts/external-claims-audit.mdx §Reusable Delta:
Current ABI repo evidence supports a stable-Rust local AI orchestration framework with deterministic Abbey/Aviva/Abi profile routing, an in-process WDBX vector/key-value/block store, segment checkpoint plus WAL persistence with runtime recovery, a compatibility JSONL snapshot mirror, epoch reclamation helpers, snapshot-persisted temporal/causal graph records, MCP hybrid WDBX query ranking, HNSW-style cosine search, SHA-256-linked conversation blocks, 3D spatial search, feature-off stubs, CLI/MCP contract coverage, explicit connector live-mode boundaries, GPU capability reporting with CPU fallback, and a tested WDBX consensus RPC transport with shared-secret and optional peer-allowlist controls. The repo does not currently prove distributed sharding, production multi-host deployment, AES/RBAC WDBX storage, Swift/Python/TensorFlow implementation claims, Kubernetes/H100 deployment claims, regulatory certifications, QPS/latency/accuracy targets, GPU speedup figures, energy-efficiency metrics, or SQuAD/CodeSearchNet/GPT comparative scores.
Integration
Pre-publish gate (agent runs the scan procedure after code gates):
./tools/check.sh
./tools/check.sh
# then: scan docs/ + collateral per Procedure above (--strict)
No feature gates — procedural validation against source.