Crypto Literature Search

Crypto Literature Search

Build a reproducible, version-aware evidence corpus. Discovery, bibliographic verification, claim support, coverage, and novelty are separate conclusions.

Scope

In scope:

• targeted, exploratory, systematic, or update searches for cryptography/ZKP/lattice research;
• prior-art and closest-work mapping;
• citation identity, version, and claim-support verification;
• related-work evidence tables and bounded novelty assessments;
• query/source logs, screening, deduplication, citation chaining, and coverage reporting.

Out of scope:

• summarizing only one supplied paper when no discovery is requested;
• auditing whether a proof is valid;
• formatting an already verified bibliography without discovery/verification;
• guaranteeing exhaustive global coverage, global novelty, or absence of unpublished work;
• publishing, contacting authors, or changing an external bibliography unless separately requested.

Required Inputs

• research question or candidate contribution decomposable into searchable claims;
• relevant primitives/problems, properties/security notions, setup/model, assumptions, parameter regime, and claimed differences;
• date, language, venue/source, publication-state, or corpus boundaries when required;
• desired deliverable: search log, verified bibliography, evidence map, related-work synthesis, or bounded novelty assessment;
• available source-access capabilities and search budget.

If the candidate claim is too vague to screen or compare, ask for the minimum contribution description. Otherwise state reasonable boundaries and continue.

Success Criteria

• Sources, exact queries, dates, filters, result counts, screening decisions, and stopping rule are reproducible.
• Work identity is separated from version identity; ePrint/preprint, conference, journal, and revision relationships remain visible.
• Every included citation has a verification status; every extracted claim points to an inspected version/location or is marked unverified.
• Search snippets and secondary summaries are used only for discovery unless independently verified against the exact source.
• Closest prior work is compared claim by claim across model, assumptions, security notion, parameters, functionality, proof basis, and compatible cost measures.
• Coverage gaps, inaccessible sources, excluded populations, and unresolved conflicts are explicit.
• Any negative or novelty conclusion is bounded by documented sources, queries, dates, and stopping conditions.

Failure Modes

• No search/network access: design the search plan and process supplied sources only; label discovery, citation chaining, and novelty coverage not executed.
• Inaccessible full text: retain bibliographic discovery status but do not treat title, abstract, snippet, or third-party summary as support for theorem/security details.
• Conflicting metadata/versions: preserve all records, reconcile authoritative identifiers when possible, and cite the exact version used.
• Over-broad or ambiguous question: decompose alternative interpretations and avoid one universal conclusion.
• Search budget exhausted before saturation: return the partial corpus, frontier, cutoff, and next queries/sources.
• Incompatible cost evidence: refuse a ranking until cost model, parameter regime, hardware/software, setup, and amortization are aligned or qualified.

Never invent an author, title, venue, year, identifier, quotation, theorem number, search result, query run, citation relationship, or novelty claim.

Workflow

1. Freeze the question, contribution claims, boundaries, deliverable, available capabilities, and budget.
2. Apply search-protocol.md to decompose concepts, generate query families, discover broadly, screen, chain citations, and stop under a documented rule.
3. Record every query, work/version, screening decision, and claim using corpus-schema.md.
4. Apply source-verification.md before treating bibliographic identity or a paper's content as verified evidence.
5. For prior-art or novelty work, apply novelty-assessment.md and compare closest work claim by claim.
6. Separate discovered, screened, identity-verified, full-text-inspected, claim-verified, and excluded records.
7. Return the corpus, evidence map, closest-work matrix, coverage limits, and bounded conclusion.
8. When evaluating this skill itself, use evaluation-scenarios.md; do not claim those scenarios ran unless independent fresh-context outputs were inspected.

Output Contract

Return, in order:

1. question, decomposed claims, and search boundaries;
2. sources, exact queries, dates, filters, and stopping rule;
3. inclusion/exclusion criteria and decision log;
4. verified corpus and version graph;
5. claim-evidence table with verification status;
6. closest-work comparison when applicable;
7. citation conflicts, inaccessible sources, and coverage gaps;
8. bounded conclusion and explicit novelty limitations;
9. reproducible next-search frontier when coverage is partial.