agentleFS
wrencanflywrencanfly/research-loop-skills/skills/research-hypothesize/SKILL.md

research-hypothesize

Skill49 starsChanged 8 days agoMITView on GitHub

Form competing explanations and design discriminating tests for an unresolved research question. Use when evidence leaves alternatives open or an experiment needs a falsifiable plan.

The file

We split this skill into the rules it gives an agent

3 rules. About 0 are common, 3 are distinctive, and 0 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈3

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 0

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Use the question, evidence map or existing knowledge, available data and tools, and work budget.

    line 12
  2. Distinctive:

    Write the decision rule before seeing the test result.

    line 19
  3. Distinctive:

    Create a short test plan containing:

    line 26

≈ “Common” is an estimate: at least 1 in 5 of the 364,980 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Research Hypothesize

Choose a test for the uncertainty it can resolve. Generate ideas when useful, but do not require novelty or a fixed number of hypotheses for every investigation.

Inputs

Use the question, evidence map or existing knowledge, available data and tools, and work budget. Identify assumptions that lack evidence before building a test around them.

Procedure

  1. State the leading proposition and at least one credible alternative when alternatives exist. Include a null or baseline explanation for comparative empirical work. For qualitative synthesis, competing interpretations may be more appropriate than a statistical null.
  2. For each alternative, state an observable implication and what would weaken it. Distinguish a testable claim from a preference, definition, or speculative possibility.
  3. Choose the least costly feasible test that separates the alternatives. Consider the value of the answer, expected information gain, input availability, validity, and cost. Avoid arbitrary precision in scoring these judgments.
  4. Write the decision rule before seeing the test result. Specify the comparison or baseline, primary outcome, evaluation population or data split, uncertainty treatment, and any threshold used to choose among alternatives. Say what an inconclusive outcome means.
  5. Design checks for the main threats to validity. Examples include confounding, data leakage, measurement error, selection bias, dataset overlap, sensitivity to assumptions, or nonrepresentative sources. Choose relevant checks rather than accumulating a universal checklist.
  6. Protect the evaluation. Keep a holdout untouched until the designated check; do not tune repeatedly to it. If prior exploration influenced the hypothesis, label the next analysis exploratory or obtain fresh confirmation data. Log changes to the planned method and why they were necessary.
  7. Specify a bounded execution and stop rule. If a required experiment cannot be performed, deliver a test proposal with its dependencies. A simulation can test model behavior; it does not substitute for unobserved real-world outcomes.

Output

Create a short test plan containing:

  • Claim and credible alternative or baseline.
  • Distinguishing prediction and disconfirming outcome.
  • Required inputs, provenance, and access constraints.
  • Method, primary outcome, decision rule, and uncertainty treatment.
  • Validity checks, execution budget, and stopping rule.
  • Whether the test is exploratory or confirmatory.

In evidence-synthesis mode, the test may be a targeted search for a critical missing study, an applicability comparison, or a reconciliation of conflicting definitions. Do not invent an experiment when the user's question only calls for a grounded synthesis.

Completion check

The plan permits an unfavorable or inconclusive result, and either would affect the next decision. Its success criterion is not merely producing a plausible explanation. State when no feasible test can distinguish the alternatives with current resources.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.