# Factor Hypothesis Extractor

> Convert an audited quantitative-finance paper or evidence packet into explicit, testable factor hypotheses without sourcing data or running a backtest. Use when reviewed academic claims must be translated into signal definitions, timing assumptions, test parameters, and falsification criteria before market-data ingestion.

- Skill: `knuckles-team/factor-hypothesis-extractor` (Agent Skill)
- Install (CLI): `npx skillmds@latest add knuckles-team/factor-hypothesis-extractor`
- Raw SKILL.md: https://api.skillmd.com/api/skills/knuckles-team/factor-hypothesis-extractor/raw
- Safety review: pending
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: Coding & Dev Tools
- License: MIT
- Author: Knuckles-Team (https://skillmd.com/u/knuckles-team)
- Updated: 2026-09-17
- Page: https://skillmd.com/skills/knuckles-team/factor-hypothesis-extractor

---


# Factor Hypothesis Extractor

Translate reviewed evidence into a test contract. Preserve the paper's stated
population, period, lag, construction, and limitations; do not turn an association
into a causal or investable claim.

## Required evidence

Use the citation-audited paper or evidence packet, stable claim/source identifiers,
the intended asset universe, and any user-supplied market, timing, or cost constraints.
If the source omits a necessary definition, mark it unresolved instead of choosing a
convenient convention.

## Output contract

For each distinct hypothesis, return:

- A unique hypothesis ID and the exact source claim IDs supporting it.
- Signal inputs, transformations, formula or pseudocode, direction, expected
  relationship, and economic rationale as stated by the source.
- Information timestamp, publication or availability lag, lookback, prediction
  horizon, rebalance frequency, and holding period.
- Eligible universe, exclusions, grouping or neutralization, missing-value policy,
  weighting assumptions, benchmark, and required transaction-cost model.
- Primary metric, robustness slices, null hypothesis, falsification threshold, and
  conditions under which the result should be rejected.
- Leakage, look-ahead, survivorship, selection, multiple-testing, capacity, and
  replication risks supported by evidence or clearly labeled as reviewer concerns.
- A readiness status of `READY`, `NEEDS_CLARIFICATION`, or `NOT_TESTABLE`, with every
  unresolved input listed.

Do not retrieve market data, implement the signal, run a backtest, rank securities,
recommend an investment, or place an order. The output is a research specification,
not evidence that the factor works outside the cited study.