result-to-claim

Evaluating experimental results to verify research claims requires collecting data from wandb and experiment logs, then judging whether the evidence supports your intended hypothesis. This Skill uses Codex to analyze those sources and determine if you should pivot, supplement, or confirm your claims.

Yes, you can use wandb logs alongside EXPERIMENT_TRACKER.md and EXPERIMENT_LOG.md for result evaluation. The Skill collects and normalizes data from these multiple sources to build a complete results picture before judging claim support.

The best way to decide whether to pivot or confirm after an experiment is to use Codex to judge your collected results against your intended claims. It routes your next actions by determining if you need to pivot, supplement with more data, or confirm findings.

Routing experimental judgments to plan follow-up work involves parsing and normalizing Codex output into actionable verdicts. This Skill automatically routes those verdicts to post-mortems, additional ablation experiments, or manuscript planning steps based on the decision.

You need to run a result-to-claim evaluation immediately after your experiments finish. Running this process at the end of the experimental cycle allows Codex to judge the final results and route the next actions for your research workflow.