Crypto Research Pipeline

Crypto Research Pipeline

Coordinate a research project as a sequence of evidence gates. Preserve uncertainty and negative results. A later stage never silently clears an earlier limitation.

Scope

Use this skill to orchestrate these installed skills when relevant:

• crypto-idea-evaluator
• crypto-literature-search
• crypto-paper-reader
• crypto-related-work-synthesis
• crypto-novelty-check
• crypto-theorem-review
• zkp-security-review
• lattice-parameter-audit
• crypto-experiment-reproducibility
• crypto-paper-writing
• crypto-paper-review

This skill manages research state, routing, rollback, and human sign-off. It does not guarantee novelty, proof correctness, security, reproducibility, venue acceptance, or publication. It does not assign authorship, contact third parties, submit a manuscript, or make ethics decisions for the responsible researchers.

Required Inputs

Collect or explicitly mark missing:

• the idea, manuscript, repository, proof, protocol, or artifact in scope;
• research objective and intended contribution;
• domain, threat/security model, assumptions, and target setting;
• known constraints on time, compute, data, disclosure, and collaboration;
• intended venue and policy snapshot, if venue readiness matters;
• ethics, dual-use, privacy, conflict, and disclosure boundaries.

Do not turn a one-sentence idea into asserted results. Record hypotheses, missing evidence, and kill criteria first.

Success Criteria

A successful run:

• maintains the research state defined in research-state.md;
• gives every stage a status of pass, pass-with-limitations, fail, or blocked;
• preserves source, claim, proof, parameter, experiment, and artifact provenance;
• propagates blockers and rolls back dependent claims when evidence changes;
• records decisions, owners, risks, and the next falsifiable action;
• invokes human gates for ethics, authorship, disclosures, final claims, and submission;
• declares submission readiness only when all blocking gates are cleared.

Failure Modes

• If a required skill or capability is unavailable, use its documented output contract if feasible; otherwise mark the stage unavailable and name the missing evidence.
• If strong prior art materially matches the contribution, pause the original narrative and recommend a pivot, narrower claim, or stop decision.
• If a theorem, reduction, security property, parameter estimate, or experiment fails, reopen dependent stages and retract unsupported downstream language.
• If browsing, computation, proof tooling, source access, or artifact execution is unavailable, do not simulate the missing verification.
• If ethics, disclosure, privacy, or dual-use concerns are unresolved, pause affected work for an accountable human decision.
• If venue rules may be stale, refresh them from the venue's official source before claiming compliance.
• Never fabricate a citation, theorem, proof step, benchmark, result, artifact execution, reviewer outcome, or novelty conclusion.

Workflow

1. Initialize and triage. Create the research-state ledger. Identify responsible humans, ethics/disclosure constraints, and the exact research question.
2. Evaluate the idea. Invoke crypto-idea-evaluator. Define the contribution hypothesis, feasibility risks, discriminating tests, and kill/pivot criteria.
3. Map the literature. Invoke crypto-literature-search; use crypto-paper-reader on closest and foundational works. Preserve exact bibliographic identifiers and verification state.
4. Synthesize and test novelty. Invoke crypto-related-work-synthesis and then crypto-novelty-check. Separate verified novelty from bounded-search residual risk.
5. Develop and audit theory. Route formal claims to crypto-theorem-review; route ZKP security claims to zkp-security-review. Keep definitions, games, assumptions, reductions, and protocol properties distinct.
6. Audit parameters and empirical design. Use lattice-parameter-audit for lattice instantiations and crypto-experiment-reproducibility for benchmarks or artifacts. Separate asymptotic, symbolic, estimated, measured, and independently reproduced claims.
7. Build the evidence package. Freeze inputs, scripts, environments, seeds, outputs, limitations, and claim-to-evidence links. Retain negative and null results.
8. Write from verified state. Invoke crypto-paper-writing. Draft only claims permitted by the ledger and label conjectures, estimates, and incomplete work.
9. Run adversarial internal review. Invoke crypto-paper-review; route every blocking finding back to the earliest responsible stage. Re-review after material changes.
10. Check readiness and stop for sign-off. Apply top-venue-gates.md. Require responsible humans to approve ethics, authorship, acknowledgments, disclosures, final claims, venue compliance, and any external submission.

Use stage-gates.md for routing and rollback. Do not skip a gate because a deadline is near.

Output Contract

Return:

1. Project state — current stage, scope, version, and overall status.
2. Stage ledger — status, evidence, limitations, blockers, and rollback targets for every stage.
3. Claim ledger — each important claim, type, dependencies, evidence, confidence, and permitted wording.
4. Evidence inventory — verified sources, proof objects, parameters, experiments, artifacts, and policy snapshots.
5. Risk register — novelty, correctness, security, parameter, reproducibility, ethics, disclosure, and venue risks.
6. Decision log — pivot/stop/continue decisions, rationale, owner, and date.
7. Readiness verdict — not ready, conditionally ready, or ready for human submission review, with unresolved items.
8. Next action — the smallest evidence-producing step, its success/failure criterion, and the responsible human or skill.