Skip to main content
Glama

Verify Performance

verify_performance
Destructive

Verify declared performance requirements by comparing measurement uncertainty bands to limits and returning pass, fail, or indeterminate verdicts with recorded provenance.

Instructions

Prove (or fail to prove) every requirement declared with declare_performance — the step that turns "a solver printed 0.29" into a claim with a band and a provenance.

A verdict has THREE states. pass and fail each require the measurement's whole uncertainty band to sit on one side of the limit; a band that straddles it is indeterminate, meaning "escalate", not "probably fine". A correlation reading Cd = 0.28 ± 10 % against a limit of 0.30 spans 0.252–0.308 and has NOT shown the part passes — collapsing that to a pass is how a spec silently goes unmet.

tier picks the evidence:

  • 'screen' — each requirement's cheap estimator only. Milliseconds, no solver.

  • 'solver' — the real solve for every requirement.

  • 'auto' (default) — screen first, escalate only what the screen could not decide or what declares fidelity_floor: 'solver'. This is the ladder that keeps a design loop cheap: cheap measurements eliminate candidates, solves confirm survivors.

Trust is part of the measurement, not a footnote: trust: {converged: true} or a band_max_pct cap makes an unconverged (or insufficiently mesh-converged) solve come back indeterminate with the reason, never pass.

Solver-tier measurements are asynchronous, so this returns EITHER the finished verdict (screen-only, or everything already decided) or {job_id, status, pending, results} — poll job_result for the completed verdict. Never raises on a failing requirement; a failure is a row.

Every verdict is also RECORDED on the part, stamped with a geometry signature of the shape it measured (#261). That record is what merge_assembly, substitutability_check and component_contract_check consult, since a gate has to answer synchronously and this may not have: an in-flight solve records rows the gates read as unverified, and editing the part invalidates the signature so they read stale — never a pass on either path.

Returns {handle, tier, ok, n_requirements, passed, failed, indeterminate, escalate, results: [{name, tier, metric, state, measured, limit, band_pct, worst_case, best_case, margin, margin_pct, detail, trust_reasons?, screen?, job_id?}]}.

Tool call template

Replace each placeholder with a value matching the input schema. Only required parameters are included; add optional parameters as needed.

{
  "id": 1,
  "jsonrpc": "2.0",
  "method": "tools/call",
  "params": {
    "arguments": {
      "handle": "<handle>"
    },
    "name": "verify_performance"
  }
}

Input Schema

NameRequiredDescriptionDefault
tierNoauto
handleYes
workspaceNoRun this call in this named workspace

Schema Changelog

Changes observed during successful MCP inspections.

  1. Changed1 schema field changedv0.5.7
    • addedInput schema / properties / workspace
      Added value: +{
      +  "default": "",
      +  "description": "Run this call in this named workspace",
      +  "title": "Workspace",
      +  "type": "string"
      +}
  2. First observed

TDQS

A4.8/5.0
Behavior5/5

Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?

Annotations only say destructive/not-read-only; the description explains what that mutation actually is — every verdict is recorded on the part with a geometry signature, and edits invalidate it to `stale`. It also discloses the asynchronous dual return shape, the poll-with-job_result requirement, that trust options downgrade to `indeterminate`, and that a failing requirement never raises.

Agents need to know what a tool does to the world before calling it. Descriptions should go beyond structured annotations to explain consequences.

Conciseness4/5

Is the description appropriately sized, front-loaded, and free of redundancy?

Front-loaded with the one-line purpose, then bulleted tier semantics, then behavioral caveats — the ordering is deliberate and the prose is dense rather than padded. It is on the long side, and the final literal return-signature block is verbose, but with no output schema it earns most of its space.

Shorter descriptions cost fewer tokens and are easier for agents to parse. Every sentence should earn its place.

Completeness5/5

Given the tool's complexity, does the description cover enough for an agent to succeed on first attempt?

Covers the async split, the meaning of each verdict state, the trust/escalation semantics, and the side effect of recording a signed verdict — everything an agent needs given there is no output schema. The explicit return field list compensates for the missing output schema.

Complex tools with many parameters or behaviors need more documentation. Simple tools need less. This dimension scales expectations accordingly.

Parameters4/5

Does the description clarify parameter syntax, constraints, interactions, or defaults beyond what the schema provides?

Schema coverage is only 33% and `tier` has no enum in the schema, so the description carries the load: it defines `'screen'`, `'solver'`, and `'auto'` with their cost/behavior tradeoffs. `handle` is only referenced obliquely as 'the part' and `workspace` is left to its schema description, so it is not fully exhaustive.

Input schemas describe structure but not intent. Descriptions should explain non-obvious parameter relationships and valid value ranges.

Purpose5/5

Does the description clearly state what the tool does and how it differs from similar tools?

States a specific verb and resource ('prove (or fail to prove) every requirement declared with declare_performance') and names the sibling that produces its inputs, so the agent knows the exact place of this tool in the workflow. The three-state verdict framing further pins down what 'verify' means here.

Agents choose between tools based on descriptions. A clear purpose with a specific verb and resource helps agents select the right tool.

Usage Guidelines5/5

Does the description explain when to use this tool, when not to, or what alternatives exist?

Explicitly spells out the three `tier` modes and when each applies, including a rationale for the default ('auto' ladder that keeps a design loop cheap). It also names job_result as the required follow-up for the async path and explains that gates (merge_assembly, substitutability_check, component_contract_check) read the recorded verdict instead of calling this synchronously.

Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.

Deploy Server

Other Tools