agentleFS
wangyi0403wangyi0403/dl-research-workflow/.agents/skills/result-to-claim/SKILL.md

result-to-claim

Skill4 starsChanged 42 days agoMITView on GitHub

Assess which completed experiments support, weaken, reject, or leave claims unresolved; separate performance, mechanism, and generalization evidence.

4 other projects use a version of this.

Adopted by

Most recently taken up by

The file

We split this skill into the rules it gives an agent

6 rules. 0 are common, 5 are distinctive, and 1 is very specific to this project.

  • 0

    Common

    Most of the 4 edited versions we read keep these.

  • 5

    Distinctive

    This author's own choices. Worth a look.

  • 1

    Very specific

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

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Use completed, traceable experiment records to set each Claim ID's permissible wording.

    line 8
  2. Very specific:

    Write the decision, supporting Experiment IDs, confidence/uncertainty, and next action into docs/08_analysis.md.

    line 14
  3. Distinctive:

    Do not call a mechanism established unless an intervention, ablation, counterfactual, condition shift, or similarly discriminating test …

    line 19
  4. Distinctive:

    Do not call a result generalizable from one site, dataset, condition, or seed.

    line 20
  5. Distinctive:

    When protocol validity is doubtful, route to design repair or rerun.

    line 21
  6. Distinctive:

    Do not silently edit the manuscript or rerun costly experiments.

    line 22

“Common” is measured: kept by at least 6 in 10 of the 4 edited versions of this file we could read. “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

Result-to-Claim Gate

Use completed, traceable experiment records to set each Claim ID's permissible wording. This is an evidence gate, not a way to turn positive metrics into a stronger story.

  1. Start from docs/00_start.md and the Claim IDs under review. Read their entries in 03, planned comparisons in 05, source index in 06, and linked structured results and registry rows. Consult 07, logs and configurations for relevant deviations or provenance gaps; do not load every run. Record absent or incompatible sources rather than filling gaps from memory. Rebuild reported tables with their recorded aggregation command; keep numeric outputs generated from structured data, with 06 as the schema/source index.
  2. For every Claim ID, identify the planned comparison, analysis unit, data split, metric, uncertainty treatment, baseline, and failure or stopping condition.
  3. Classify the result as promote, weaken, reject, or collect-more-evidence. State the exact supported result, the unsupported extrapolation, and the smallest discriminating next check.
  4. For engineering research, decide separately whether the result supports: (a) the target engineering task under its stated operating constraints; (b) the proposed measurement, physical, or inference mechanism; and (c) any transfer claim. A task improvement alone does not establish (b) or (c).
  5. Write the decision, supporting Experiment IDs, confidence/uncertainty, and next action into docs/08_analysis.md. Preserve failed and null results. Update the same Claim ID in docs/03_idea_report.md: promote → promoted, weaken → weakened, reject → rejected, collect-more-evidence → needs-evidence. Analysis retains the decision evidence; do not maintain a second current-status ledger.

Decision rules

  • A reported number is admissible only when its experiment ID, configuration, data version/split, seed treatment, and output path are traceable.
  • Do not call a mechanism established unless an intervention, ablation, counterfactual, condition shift, or similarly discriminating test supports it.
  • Do not call a result generalizable from one site, dataset, condition, or seed. State the actual scope.
  • When protocol validity is doubtful, route to design repair or rerun; when evidence is insufficient, weaken the claim; when the core mechanism fails, route to pivot or reject.
  • Do not silently edit the manuscript or rerun costly experiments. Report the decision and required authorization.

Report the Claim ID table, gate decision, unresolved conflicts, and the next smallest verifiable action.

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.