@agentbiz/quant-research

@agentbiz/quant-research turns a declared research instrument into a checked codebook, recode artifact, coverage result, and analysis specification before anyone opens SPSS, AMOS, SmartPLS, or lavaan.

Large language models do not reliably keep Likert bounds, reverse-keyed items, construct codes, or SEM path lists consistent. This library is the typed contract those agents should call before anyone opens SPSS, AMOS, SmartPLS, or lavaan.

It compiles JSON in and emits dictionaries, coverage maps, recode artifacts, SPSS syntax, lavaan specs, AMOS path lists, and SmartPLS indicator maps. It does not estimate models, compute fit indices, write Excel or Google Sheets, or call a model provider.

Live status (2026-08-24): public alpha (0.1.0-alpha.5 on GitHub, 0 stars, 0 forks). No external user, downstream repository, or pilot is verified. npm alpha resolves 0.1.0-alpha.5 while latest remains 0.1.0-alpha.4. This release is a deterministic contract and emitter library. It does not estimate SEM, compute fit indices, write Excel or Google Sheets, call a model provider, or claim adoption, downloads, or dependent repositories.

If it caught one inconsistent recode before analysis, star it. That is the only growth signal this repo tracks.

Install from npm

Install the public alpha from the npm registry:

npm install @agentbiz/quant-research@alpha

The package targets Node.js 20 and 22. Use the GitHub repository for release notes, boundary documents, examples, and issue reporting.

Verify from source

To reproduce the committed release locally, use the verified GitHub tag:

git clone --branch v0.1.0-alpha.5 https://github.com/daichunghy/quant-research.git
cd quant-research
npm ci
npm run verify

Support and first-run questions: .github/SUPPORT.md or GitHub Discussions.

This is Github 4. It is not PatchGate, contribkit, or OpenSheet-AI. See docs/BOUNDARY.md. For citation and clean-room reproduction, see CITATION.cff and docs/REPRODUCIBILITY.md.

The shortest route to a concrete artifact is the first-use walkthrough.

Install (local)

npm install
npm run verify

node dist/cli.js compile examples/tam-instrument.json
node dist/cli.js codebook examples/tam-instrument.json
node dist/cli.js recode examples/tam-instrument.json examples/tam-dataset.json
node dist/cli.js gap examples/tam-instrument.json examples/tam-observed.json
node dist/cli.js audit examples/tam-instrument.json examples/tam-dataset.json
node dist/cli.js emit-lavaan examples/tam-bundle.json
node dist/cli.js emit-spss examples/tam-bundle.json
node dist/cli.js csv examples/tam-dataset.json

MCP stdio: node dist/mcp.js (agentbiz-quant-mcp). Tools are executable through executeTool, not documentation-only.

Agent skill: copy skills/quant-research/SKILL.md into the target agent's skill directory, or use the installed package with AGENTBIZ_QUANT_RESEARCH_ROOT pointing at a clone. The skill is portable and does not assume the maintainer's filesystem path.

Library

import {
  compileInstrument,
  compileTopicCard,
  compileMeasurement,
  emitLavaanSyntax,
} from "@agentbiz/quant-research";

const instrument = compileInstrument(instrumentJson);
const topicCard = compileTopicCard(topicCardJson, instrument.artifact.instrument);
const measurement = compileMeasurement(measurementJson, instrument.artifact.instrument, topicCard.artifact);

const syntax = emitLavaanSyntax({
  instrument: instrument.artifact.instrument,
  topicCard: topicCard.artifact,
  measurement: measurement.artifact,
});

Agent tool JSON Schemas live on @agentbiz/quant-research/tools. They describe functions. They do not call Vercel, OpenAI, or Anthropic.

What v1 covers

Contract	Job
agentbiz.instrument.v1	Scale bank with reverse flags, Likert bounds, demonstration vs cited provenance
agentbiz.topic-card.v1	Question, claim class, hypotheses bound to instrument codes
agentbiz.gap-map.v1	Coverage: covered / partial / missing / mismatch / unexpected
agentbiz.dataset.v1	Tabular snapshot audit: bounds, missing, duplicates, straight-lining (report only)
agentbiz.recode.v1	New columns only: reverse-score, missing sentinels, straight-line flag
agentbiz.measurement.v1	Declared CB-SEM / PLS-SEM / reliability spec
emitters	SPSS syntax, lavaan text, AMOS JSON paths, SmartPLS indicator map

The TAM example items are demonstration items. They are not published scales.

Non-goals

• Cronbach’s alpha, AVE, HTMT, EFA, CFA, SEM, PLS, p-values, or bootstrap
• Deleting items or cases to chase a threshold
• .xlsx / Google Sheets writes (that is OpenSheet-AI)
• Model calls or a Claude/Codex plugin in this package
• Empty @agentbiz/sales-crm / finance-hr packages

The alpha is locally verified by npm run verify, including schema reconciliation, package-content checks, secret scanning, and CLI smoke tests. Local verification is not evidence of external users, registry adoption, ethical approval, or correctly specified research.

Who this is for

• Survey researchers and students whose instruments go into SPSS, lavaan, AMOS, or SmartPLS and need bounds, reverse-keyed items, and coverage checked before analysis.
• Research assistants who maintain scale banks and codebooks across survey waves.
• Agent builders who need a typed contract layer under a research copilot.
• Not a fit if you want fit indices, estimation, or item deletion — see Non-goals.

If it caught one inconsistent recode for you, star the repository. It helps other researchers find the contracts.

Release history: CHANGELOG.md.

License

Apache-2.0. See LICENSE.