# Research To Decision

> Run bounded technical or product research and turn it into a decision-ready repository artifact. Use when a spec, architecture choice, dependency, standard, feasibility gate, or implementation plan depends on external evidence or a controlled experiment, especially when research is delegated to another agent or chat.

- Skill: `dills122/research-to-decision` (Agent Skill, multi-file: 2 files)
- Install (CLI): `npx skillmds@latest add dills122/research-to-decision`
- Raw SKILL.md: https://api.skillmd.com/api/skills/dills122/research-to-decision/raw
- Safety review: pending
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: AI & ML
- Author: dills122 (https://skillmd.com/u/dills122)
- Updated: 2026-09-17
- Page: https://skillmd.com/skills/dills122/research-to-decision

---


# Research To Decision

Answer one decision question with retained evidence. Keep research, inference, and the final decision distinct.

## Frame The Packet

State before searching or experimenting:

- question and decision owner;
- why the answer matters now;
- authorized scope and prohibited actions;
- preferred source hierarchy;
- comparison or success criteria;
- stop condition and expected repository artifact.

Do not broaden a research task into implementation authority.

## Gather Evidence

Prefer primary and current sources. For repository experiments, record the exact commit, environment, inputs, commands, and artifact paths. Separate each conclusion into:

- **Documented fact:** directly supported by a cited source;
- **Observation:** reproduced in the named environment;
- **Inference:** reasoned from facts or observations;
- **Unknown:** material uncertainty that remains.

Avoid long source dumps. Retain the smallest excerpt or result needed to make the conclusion auditable.

## Compare Against The Decision Criteria

Evaluate credible options consistently. Record constraints, trade-offs, failure modes, reversibility, and what evidence would change the recommendation. A failed candidate is not proof that every alternative works.

## Retain The Result

Write the result into the repository's existing research, feasibility, or decision location. Include:

- scoped status;
- executive conclusion;
- method and source index;
- evidence with fact/observation/inference labels;
- option comparison;
- confidence, limitations, and unresolved questions;
- recommendation and required decision owner;
- implementation consequences and next gate.

If the conclusion changes architecture, security posture, public behavior, or an expensive-to-reverse choice, create or update the governing ADR or spec after approval. Chat history alone is not retained evidence.