Whetstone

Measured conclusion: Deep-tested. L2 outcome: tools_ok.

Verifier-grounded AI promotion gates, disposable report cards, and signed PASS/HOLD/BLOCK receipts.

Add to your agent

Copy this into your MCP client config:

{
  "mcpServers": {
    "Whetstone": {
      "url": "https://whetstone.cyberelf.link/mcp",
      "transport": "streamable-http"
    }
  }
}

Measured tools (14)

• inspect_promotion

  Quarantine declared exposure, compare paired baseline/candidate outcomes on the clean remainder, and issue a promotion receipt. Bring your own exam rows, exposure records, and per-item results. Full example: GET /api/examples key 'inspector'.

  Parameters:

  • exam (required) — array — Exam rows. Each row needs item_id (or id) plus prompt/content/input/task/question/expression.
  • baseline (required) — object — item_id -> boolean pass/fail result for the baseline system.
  • candidate (required) — object — item_id -> boolean pass/fail result for the candidate system.
  • policy — object — Explicit promotion policy. Omitted fields use the documented defaults.
  • domains — object — Optional item_id -> domain label mapping.
  • exposure — array — Declared exposure rows carrying identity/content fields and an optional source or path.
  • baseline_name — string
  • candidate_name — string
  • retained_probe — object — Optional retained-capability result checked alongside the paired cohort.
  • fingerprint_max_n — integer
  • similarity_threshold — number
  • enable_text_similarity — boolean
  • enable_behavioral_fingerprint — boolean

  Hints: read-only not declared · destructive not declared · idempotent not declared · open-world not declared

• audit_leakage

  Exact declared-exposure audit over your exam rows: row identity, behavioral fingerprints for graph-DSL expressions, text-similarity review flags, and a clean exam export. Full example: GET /api/examples key 'leakage'.

  Parameters:

  • exam (required) — array — Exam rows. Each row needs item_id (or id) plus prompt/content/input/task/question/expression.
  • exposure — array — Declared exposure rows carrying identity/content fields and an optional source or path.
  • fingerprint_max_n — integer
  • similarity_threshold — number
  • enable_text_similarity — boolean
  • enable_behavioral_fingerprint — boolean

  Hints: read-only not declared · destructive not declared · idempotent not declared · open-world not declared

• promotion_gate

  PASS, HOLD, or BLOCK from paired per-item results: gains, regressions, exact McNemar p-value, per-domain breakdown. Full example: GET /api/examples key 'gate'.

  Parameters:

  • baseline (required) — object — item_id -> boolean pass/fail result for the baseline system.
  • candidate (required) — object — item_id -> boolean pass/fail result for the candidate system.
  • policy — object — Explicit promotion policy. Omitted fields use the documented defaults.
  • domains — object — Optional item_id -> domain label mapping.
  • baseline_name — string
  • candidate_name — string
  • retained_probe — object — Optional retained-capability result checked alongside the paired cohort.

  Hints: read-only not declared · destructive not declared · idempotent not declared · open-world not declared

• bank_health

  Item-lifecycle diagnostics over your grading history: discriminators, saturated and flaky items, frontier gaps. Full example: GET /api/examples key 'health'.

  Parameters:

  • history (required) — array — Observed item/system outcomes.
  • items — array — Optional item definitions.

  Hints: read-only not declared · destructive not declared · idempotent not declared · open-world not declared

• safe_patch

  Apply a section-scoped Markdown patch under conservation checks (untouched sections stay byte-identical; protected tokens preserved). Full example: GET /api/examples key 'safepatch'.

  Parameters:

  • document (required) — string — Complete Markdown document to patch.
  • operations (required) — array
  • reason — string

  Hints: read-only not declared · destructive not declared · idempotent not declared · open-world not declared

• counterexample_hunt

  Bounded simulated-annealing search for a graph counterexample inside a DSL predicate class, with an exact certificate when found. CPU-bounded and strictly rate-limited. Full example: GET /api/examples key 'counterexample'.

  Parameters:

  • expression (required) — string — Graph predicate in the Whetstone DSL, for example: is_connected and is_triangle_free and not is_bipartite
  • ns — array — Graph sizes searched.
  • seed — integer
  • steps — integer
  • restarts — integer

  Hints: read-only not declared · destructive not declared · idempotent not declared · open-world not declared

• memory_relevance

  Compare query-free salience against objective-conditioned relevance for a set of memories under a token budget. Full example: GET /api/examples key 'memory'.

  Parameters:

  • objective (required) — string
  • memories (required) — array — Memories to rank.
  • current_step — integer
  • token_budget — integer
  • question_kind — string
  • context_entities — array — Entities already active in context.
  • objective_entities — array — Optional explicit entities when the objective text is not self-describing.

  Hints: read-only not declared · destructive not declared · idempotent not declared · open-world not declared

• replay_trace

  Turn reasoning-emulator control events into checkpoints, rewinds, notes, and a timeline. Full example: GET /api/examples key 'replay'.

  Parameters:

  • events (required) — array — Ordered reasoning-emulator events.
  • notes — array — Optional analyst notes.

  Hints: read-only not declared · destructive not declared · idempotent not declared · open-world not declared

• report_card_start

  TIER 1: start a disposable report-card session. Returns exam items (graph-repair prompts minted from the repository's public frontier) for THIS agent to answer. Answer every item, then call report_card_submit exactly once. Sessions are one-shot, expire in 15 minutes, and are strictly rate-limited. This demonstrates the promotion-gate mechanism on disposable items; it is not a private-bank credential.

  Parameters:

  • challenge — string — Caller nonce bound into the signed receipt for replay detection.

  Hints: read-only not declared · destructive not declared · idempotent not declared · open-world not declared

• report_card_submit

  TIER 1: submit answers for a report-card session and receive the graded report (per-item verdicts, per-domain totals, SHA-256 commitments). Grading is by checker spec: verified strict refinements are reported separately, and promotion grade requires at least 5% clean-support retention. No answer key exists. The session is destroyed by this call.

  Parameters:

  • session_id (required) — string
  • answers (required) — object — item_id -> answer (a DSL predicate, or the JSON reply the prompt asked for)

  Hints: read-only not declared · destructive not declared · idempotent not declared · open-world not declared

• open_bench_start

  TIER 2: start a one-shot Open Promotion Bench session. Returns six fresh virtual-repository scope-integrity tasks. Run a baseline and candidate independently on the same cohort, then submit both answer maps with open_bench_submit. This is an open, procedural, self-attested track rather than a private-bank credential.

  Parameters:

  • challenge — string — Caller nonce bound into the signed receipt for replay detection.

  Hints: read-only not declared · destructive not declared · idempotent not declared · open-world not declared

• open_bench_submit

  TIER 2: grade paired baseline and candidate patches, count gains/regressions/ties, and issue PASS/HOLD/BLOCK. Set publish=true plus attestation=true to append only the safe manifests and sanitized receipt to the public board; tasks and answers are never persisted.

  Parameters:

  • session_id (required) — string
  • baseline_manifest (required) — object
  • candidate_manifest (required) — object
  • baseline_answers (required) — object
  • candidate_answers (required) — object
  • publish — boolean
  • attestation — boolean

  Hints: read-only not declared · destructive not declared · idempotent not declared · open-world not declared

• open_bench_leaderboard

  TIER 2: list the self-attested public Open Promotion Bench receipts. Entries contain manifests, verdicts, item-level transitions, and commitments but never task contents or submitted answers.

  Hints: read-only not declared · destructive not declared · idempotent not declared · open-world not declared

• about_whetstone

  What this service is: the tool catalog, the tier boundaries, and where the source lives.

  Hints: read-only not declared · destructive not declared · idempotent not declared · open-world not declared

Publisher website Repository

Measured by MCPYP probing. Untested fields are omitted, never estimated. Listing is always free; MCPYP never proxies calls or stores user tokens.