skills/crypto-literature-evidence/references/researcher-studies.md

# Evidence for research-method studies

Use this reference when asked to learn reusable methods from a researcher's work, a collaboration, or a selected corpus. The result should identify mechanisms and their limits, not construct a persona or infer a private discovery process. Apply the main skill's version and locator checks; this reference adds the attribution questions specific to these studies.

## Attribute the observation before generalizing it

Keep support for a technical observation separate from support for attributing a general habit to an individual. A clearly explained mechanism in a joint paper can strongly support the former while leaving the latter unknown.

| Evidence category | Supports | Does not establish |
| --- | --- | --- |
| Direct expression | What a person explicitly says in a first-person account, solo exposition, or clearly attributed explanation | A universal practice or a complete discovery history |
| Single-author technical work | The listed author's exposition and explicitly claimed technical contribution | Exclusive origin of the underlying ideas or every acknowledged contribution behind it |
| Joint-work mechanism | The construction, definition, proof organization, or methodological statement in a coauthored paper | Which author originated the idea or wrote the passage |
| Reception | What a colleague, interviewer, or award citation says about influence or significance | Proof correctness, individual authorship, or a causal explanation of success |
| Inference or proposed adaptation | The analyst's synthesis or suggested research move | A source statement, author endorsement, or validated improvement |

Credit the full author team for a joint-paper mechanism unless a primary account supports more specific attribution. Author order, seniority, and advisor relationships do not allocate individual ideas or prose. Distinguish a source's own words from an interviewer's paraphrase and a later account's recollection. A discovery narrative requires evidence; an order chosen to explain a proof is not evidence of the order of discovery.

### Optional attribution checks

Use these cases when a synthesis assigns an idea's origin or a discovery episode to an individual. The source-reported roles are narrower than authorship of the complete result.

- **Exposition versus origin.** Mihir Bellare's sole-authored *Practice-Oriented Provable-Security*, September 1998 version, credits ideas developed jointly with Phillip Rogaway in its acknowledgments (p. 12). Preserve that joint credit when discussing the ideas behind the exposition. Access: title page and acknowledgment passage checked September 21, 2026; no whole-paper audit. [Author PDF](https://cseweb.ucsd.edu/~mihir/papers/isw-pops.pdf#page=12).
- **An explicitly assigned discovery episode.** Bellare and Oded Goldreich's *On Defining Proofs of Knowledge*, August 26, 1992 version, §7.1, p. 15, attributes recognizing the inadequacy of earlier formulations to Goldreich after Leonid Levin requested a modular presentation of graph nonisomorphism. Do not retell this as Bellare's personal discovery. Access: title page, §7.1's opening attribution passage, and acknowledgments checked September 21, 2026; no full proof audit. [Author PDF](https://cseweb.ucsd.edu/~mihir/papers/pok.pdf#page=15).

## Record access depth

For each source used in a material claim, record what was actually inspected alongside the exact version and locator. Useful descriptions include full text, selected sections with their names or numbers, abstract only, metadata only, transcript only, audiovisual playback, or inaccessible. Opening a file or locating a book is not equivalent to reading it. Record a summary's reported access separately from your own inspection.

Match each conclusion to that depth. Metadata can support bibliographic discovery; it cannot supply an unseen mechanism or private research habit. An abstract can support what the abstract claims, while detailed model or proof obligations may remain unverified. Transcript timestamps are navigation aids unless checked against playback. Report missing sources and selective coverage when they constrain the synthesis.

## Extract a conditional research move

For each proposed lesson, capture only the fields needed to make it usable:

- **Target and obstacle:** the sought capability or guarantee and the specific obstruction identified in the source.
- **Mechanism and obligation:** what representation, definition, protocol role, or proof step changes; what must be shown; and which parts are intuition, source-stated results, or cited dependencies.
- **Attribution and access:** source/version, exact locator, evidence category, and inspected material.
- **Applicability and counterevidence:** when the move helps, a condition under which it fails or is a poor choice, and evidence that narrows the lesson.
- **Proposed transfer:** what is your adaptation to the user's problem and which new obligations remain unresolved.

Scale this record to the task; do not require a table or every field for a single observation. Name the concrete action in a proof, such as which variable a hybrid permits replacing and under which premise, rather than extracting a lesson like “use clever hybrids.”

Before promoting an observation into reusable guidance, look for relevant contrary cases or alternative explanations. Broaden a selective corpus across time, problems, or collaborations when a claimed general pattern warrants it; multiple papers from one collaboration are related evidence, not independent votes. Successful published results do not reveal the denominator of unsuccessful attempts. If the evidence cannot distinguish a general method from a local technique, keep the conclusion local.

Do not invent a historical failed attempt to illustrate a lesson. A hypothetical counterexample or toy obstruction is useful when explicitly labelled and mathematically justified. Preserve evidence that different problems favor different choices rather than forcing one research template.

## Preserve provenance and validation separately

A source's account of who suggested an idea, who revised the prose, and who checked an argument supports distinct process claims. It does not independently certify the mathematics. When adapting a method, separate its historical attribution from the status of the new lemma or construction. Neither the researcher's reputation nor an AI-use disclosure discharges a proof obligation.

Deliver the supported observation, its limited attribution, the proposed adaptation, and the decisive uncertainty. Extracting a plausible method is not evidence that adding it to an assistant improves research quality; such a claim requires a separate evaluation against the existing workflow.

Provenance: adapted from §§2, 5, and 8 of the user-supplied *Learning from Amit Sahai: A Cryptography Research Distillate* (research cutoff September 20, 2026). These are proposed evidence-handling practices, not attributed personal habits or independently validated research improvements.

The single-author distinction also incorporates §1.2 of *Learning from Brent Waters: A Cryptography Research Distillate* (research cutoff September 21, 2026). Stronger individual attribution does not make a source stronger evidence for a particular technical claim; inspect the actual definition or argument.

The optional attribution checks were prompted by §2.2 of the user-supplied *Learning from Mihir Bellare: A Cryptography Research Distillate*; their factual statements were checked against the primary passages linked above.