# Researchstack Experiment Ops

> Experimental-process management skill for research projects. Use when Codex should structure runbooks, logging, checkpointing, seed control, result triage, failure analysis, experiment queues, and evidence collection so that the day-to-day experimental workflow stays publication-grade instead of devolving into ad hoc trial-and-error.

- Skill: `lqf0624/researchstack-experiment-ops-2` (Agent Skill, multi-file: 6 files)
- Install (CLI): `npx skillmds@latest add lqf0624/researchstack-experiment-ops-2`
- Raw SKILL.md: https://api.skillmd.com/api/skills/lqf0624/researchstack-experiment-ops-2/raw
- Safety review: pending
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: Research & Search
- Author: lqf0624 (https://skillmd.com/u/lqf0624)
- Updated: 2026-09-17
- Page: https://skillmd.com/skills/lqf0624/researchstack-experiment-ops-2

---


<!-- AUTO-GENERATED for codex from researchstack v0.3.0. Edit source SKILL.md files, then regenerate from the source checkout. -->

# Experiment Ops

Read [../researchstack/references/workflow.md](../researchstack/references/workflow.md) and [../researchstack/references/evidence-contracts.md](../researchstack/references/evidence-contracts.md).

Use the bundled templates:

- [assets/runbook-template.md](assets/runbook-template.md) for the canonical per-campaign runbook.
- [assets/result-triage-template.md](assets/result-triage-template.md) for classifying wins, failures, and suspect runs.
- [assets/claim-evidence-ledger.md](assets/claim-evidence-ledger.md) to decide which numbers are allowed into the paper.

Operate like a disciplined lab manager.

For ongoing experiments, enforce:

1. Clear run naming and provenance.
2. Explicit seed and environment capture.
3. Separation between exploratory runs and paper-claim runs.
4. Result logs that explain failures, not only wins.
5. Promotion rules for when a result is allowed into the paper.
6. Explicit states for observation, mechanism evidence, controlled result, robust result, and paper-ready claim.

Produce practical artifacts such as:

- experiment queue,
- runbook template,
- result triage checklist,
- failure taxonomy,
- promotion criteria for plots and tables.

When the user has no existing structure, instantiate the bundled templates instead of inventing ad hoc formats.

Be strict about process failures:

- overwritten outputs,
- undocumented config changes,
- inconsistent warmup or stopping criteria,
- comparing runs from mismatched code versions,
- repeated cherry-picking without negative-result accounting.
- comparing narrowed timings that omit required dynamic work against full end-to-end baselines.