Is the mathematical claim established?
Inspect the identity, rule, canonical form, negative cases, and supported operator family.
Review kit
Evaluate one bounded workflow with the right artifact.
Choose the review packet that matches the question: local execution, scoped verification, reproducibility, scientific-AI record quality, or benchmark credibility.
Review questionDoes this evidence materially change your confidence in the computation?
Review packet matrix
Start from the artifact, not a broad product claim.
The strongest first review is narrow enough to replay and important enough to reveal whether OperatorWorks changes the quality of technical judgment.
| Review path | Best for | Primary artifact | What the reviewer should confirm |
|---|---|---|---|
| Workbench local review | Researchers, labs, technical evaluators | Notebook, managed environment, derivation trace, export path | The workflow installs, launches, executes, preserves order, exposes assumptions, and produces a reviewable output locally. |
| Verify workflow review | Quantum, scientific, and high-assurance R&D | Scoped claim, assumptions, outcome, certificate-style record, replay context | The result is pass, fail, inconclusive, or unsupported for reasons the reviewer can inspect. |
| Evidence Bundle review | Research handoff, enterprise review, procurement | Manifest, inputs, trace, outputs, environment, warnings, hashes, notes | The artifact is complete enough to understand, identify, and replay without reconstructing the workflow from screenshots. |
| Scientific-AI record review | AI-for-science teams and technical-data buyers | Record schema, provenance, derivation, canonical result, difficulty, uniqueness and verification metadata | The record is useful, non-duplicative, traceable, and better suited to evaluation or training than a plausible but unverified example. |
| Benchmark review | Technical buyers, independent reviewers, partners | Task definition, baseline, environment, expected output, measured result, limitations | The comparison is fair, reproducible, relevant to a real workflow, and adequate for the precise claim being made. |
Reviewer scorecard
Six questions separate evidence from theater.
A strong packet should let a skeptical technical reviewer answer these questions directly.
Are the conditions explicit?
Check conventions, ordering rules, domains, index conditions, and any facts required by the transformation.
Can the artifact be reproduced?
Use the recorded environment, versions, inputs, deterministic steps, and replay guidance.
Is the trail complete?
Look for manifests, hashes, warnings, outputs, certificates, and links between the claim and its evidence.
Are unsupported cases visible?
Confirm that inconclusive and unsupported states remain distinct from verified results.
Does the language match the proof?
Ensure the packet does not imply production, customer, cloud, API, or superiority status beyond the evidence.
Evidence Bundle anatomy
One packet connects the claim to its context.
The exact schema may evolve, but the review standard remains: preserve enough information for a qualified reviewer to understand what happened and what remains unproven.
01
Identity — workflow ID, product version, run context, and stable artifact identifiers.
02
Scientific context — input expression, assumptions, operator families, conventions, and claim.
03
Computation — transformation trace, outputs, warnings, and explicit verification outcome.
04
Reproducibility — environment, dependency context, hashes, replay guidance, and reviewer notes.
What to send
- A redacted description of one narrow workflow
- The reviewer audience and artifact type you need
- The failure mode: order, assumptions, equivalence, provenance, or replay
- The cost of a silent error or weak handoff
- The boundary of information you are authorized to share
What to keep out
- Unredacted proprietary or confidential equations
- Customer data, credentials, regulated records, or export-controlled content
- Requests that assume a hosted service or broad deployed operation exists
- Claims that depend on undisclosed benchmarks or unreviewed artifacts
- Broad “prove everything” requests that cannot be scoped or replayed
Controlled review
Bring one narrow workflow and one clear decision.
Review posture: review packets are selected, bounded artifacts. They are not a substitute for independent technical judgment and do not imply broad production readiness, public benchmark leadership, hosted availability, customer adoption, or support outside the stated workflow boundary.