# Signal Foundry

> 公開情報を企業単位で正規化し、観測・差分・シグナルとして蓄積するプロダクト。

Signal Foundry prepares company data before agents spend tokens searching the web. The shipped v0 surface is company-list first: agents can resolve companies, combine region / industry / employee count / website / listing filters, save lists, add columns, and export CSV artifacts with corporate numbers, locations, reasons, and evidence through CLI / Skills / REST. Output is structured JSON with coverage, warnings, credit boundaries, and recovery hints.

Use the CLI first when operating from Claude Code or Codex. The browser UI is primarily a read surface for saved Lists, rows, Columns, evidence, credits, and Runs.

## Docs

- [Full LLM context](https://signal-foundry.app/llms-full.txt): complete public route index for docs, changelog, blog, and core pages
- [Getting started](https://signal-foundry.app/docs/getting-started/getting-started): product shape and the first path
- [Install sf CLI](https://signal-foundry.app/docs/getting-started/install-sf-cli): Homebrew, verified GitHub Release artifact, and winget status for the CLI
- [Quick start](https://signal-foundry.app/docs/getting-started/quick-start): install -> auth -> first read-only company lookup
- [Data provenance](https://signal-foundry.app/docs/getting-started/data-provenance): source families, evidence states, coverage gaps, and user-provided source boundaries
- [Job workflows](https://signal-foundry.app/docs/workflows/workflows): choose query-start, import-start, company-start, theme-start, or seed-company
- [Sales list prep](https://signal-foundry.app/docs/workflows/sales-list-prep): list create -> refine -> save -> column add -> filter -> export
- [Import workflow](https://signal-foundry.app/docs/workflows/import-workflow): CSV -> resolve -> rows -> select -> saved List
- [Company research workflow](https://signal-foundry.app/docs/workflows/company-research): resolve company_id -> profile -> observations -> filings -> compare
- [Market research](https://signal-foundry.app/docs/workflows/market-research): cross-company observation search and evidence review
- [Competitor research](https://signal-foundry.app/docs/workflows/competitor-research): seed company -> candidates -> same-shape comparison
- [Essential commands](https://signal-foundry.app/docs/cli/essential-commands): canonical command sequence for agents
- [API overview](https://signal-foundry.app/docs/api/api): auth, rate limits, endpoint groups, errors, and CLI equivalents
- [List API overview](https://signal-foundry.app/docs/api/lists): create, refine, save, columns, filter, and export sequence
- [List Columns API](https://signal-foundry.app/docs/api/lists-columns): estimate and run credit-capped Column writes on saved Lists
- [List Export API](https://signal-foundry.app/docs/api/lists-export): export a saved List without spending credits
- [Commands & flags](https://signal-foundry.app/docs/cli/commands-flags): CLI reference index and required flags
- [CLI updates](https://signal-foundry.app/docs/cli/cli-updates): version check, OS package manager update, skill reinstall
- [Claude Code workflow](https://signal-foundry.app/docs/workflows/claude-code-workflow): agent skill setup and execution loop for Claude Code
- [Authentication](https://signal-foundry.app/docs/authentication/overview): API key model
- [Credit schedule](https://signal-foundry.app/docs/billing/credit-schedule): operation-level free/paid boundary, max-credits, idempotency, and retry rules
- [Access plans](https://signal-foundry.app/docs/billing/access-plans): team workspace Free / Pro / Credit Pack grants
- [Troubleshooting](https://signal-foundry.app/docs/troubleshooting/troubleshooting): recover from auth, not found, low-hit, and preview URL problems

## API

- [Companies search](https://signal-foundry.app/api/signal-foundry/companies): search by code, name, segment
- [Company profile](https://signal-foundry.app/api/signal-foundry/companies/{companyId}/profile): canonical profile
- [Company observations](https://signal-foundry.app/api/signal-foundry/companies/{companyId}/observations): timeline of observations
- [Company filings](https://signal-foundry.app/api/signal-foundry/companies/{companyId}/filings): list of public filings
- [Filing detail](https://signal-foundry.app/api/signal-foundry/companies/{companyId}/filings/{filingId}): single filing detail
- [Filing compare](https://signal-foundry.app/api/signal-foundry/companies/{companyId}/filings/{filingId}/compare): compare two filings
- [List create](https://signal-foundry.app/api/signal-foundry/lists): create a free List preview from a query
- [List detail](https://signal-foundry.app/api/signal-foundry/lists/{listId}): read preview or saved List rows
- [List refine](https://signal-foundry.app/api/signal-foundry/lists/{listId}/refine): narrow a List preview before saving
- [List save](https://signal-foundry.app/api/signal-foundry/lists/{listId}/save): create a saved List with a credit cap
- [List column estimate/run](https://signal-foundry.app/api/signal-foundry/lists/{listId}/columns/{columnKey}): add bounded columns to a saved List
- [List export](https://signal-foundry.app/api/signal-foundry/lists/{listId}/export): export a saved List
- [Import create](https://signal-foundry.app/api/signal-foundry/imports): import CSV or external rows
- [Import rows](https://signal-foundry.app/api/signal-foundry/imports/{importId}/rows): review matched, ambiguous, and not-found rows
- [Credits balance](https://signal-foundry.app/api/signal-foundry/credits/balance): available / remaining / reserved credits
- [Credits summary](https://signal-foundry.app/api/signal-foundry/credits): ledger summary and recent credit events

## Credit Contract

- Team workspace is the billing and credit scope. A one-person workspace is still a team workspace for API keys, Lists, credits, and billing.
- Read endpoints, List preview/refine, Import row review, export, and feedback consume 0 product credits.
- Credit-consuming writes require execute + maxCredits. In CLI, use --max-credits <n> --json.
- list save consumes Basic credit for saved rows.
- import resolve/select consumes Basic credit for matched or newly selected rows.
- website and employee columns consume column credit only for found rows. not_found, no_data, and skipped are not billable.
- If max_credits_required, max_credits_exceeded, credit_balance_insufficient, unsupported, weak, or needs_human appears, stop and follow the docs recovery path instead of reporting silent success.

## Provenance Contract

- Always preserve Company, List, Column, Import, Workflow, Run, evidence, and credit fields when summarizing results.
- Use data capabilities and source_coverage before treating low hits as real market absence.
- Import create and row review are free. canonical company resolution and explicit row selection are the credit boundary.
- EDINET text sections and financial facts are evidence sources; do not present unsupported coverage as 0 results.
- Public docs describe CLI / REST as shipped. OpenAPI generation and public MCP are deferred until endpoint tags and schemas are fixed.

## Agent Routing

- Start with: sf version --json --check-update
- Then run: sf auth show --json
- For source/grain/field discovery, run sf data catalog --json before choosing CLI/API entrypoints
- For capability/filter/pricing status, run sf data capabilities --json
- For aggregate data count/coverage/freshness, run sf data health --json. Do not treat it as query relevance proof
- For company-list or sales-list work, start with sf list create "<query>" --json and refine the preview before saving
- For List previews, read candidate_set.candidates before rows[]. Use evidence_handles for verification and inspect rows[].evidence only when needed
- Before any credit-consuming write, check sf credits balance --json and the Credit schedule docs
- Use sf list create -> refine -> save -> column add -> filter -> export for saved List workflows
- Use sf import csv -> resolve -> rows -> select -> save for CSV/import-start workflows
- Use sf company search before profile / observations / filings for one-company research
- Treat unsupported or weak source coverage as needs_human; do not silently turn it into a 0-result answer
- For CLI updates, use Homebrew or the verified GitHub Release artifact. Use winget only after the Windows package appears in the public catalog. Then reinstall the signal-foundry-agent skill with --force and run sf agent eval --json
- The v0 docs stay on the current Next docs surface; Mintlify migration, OpenAPI public spec, and public MCP are deferred

## Optional

- [Blog](https://signal-foundry.app/blog): updates
- [Changelog](https://signal-foundry.app/changelog): release notes
- [Pricing](https://signal-foundry.app/pricing): access plans
- [FAQ](https://signal-foundry.app/faq): common questions
- [Contact](https://signal-foundry.app/contact): support