GovCon FAR/DFARS Clause & Flow-Down Extractor

What it does (for agents & pipelines): give it US government contract / RFP text (or a public URL/PDF) and it returns structured JSON of every FAR/DFARS clause, each clause's official title and subcontractor flow-down status, plus delivery milestones, penalties and option periods — every field backed by a verbatim source quote. Machine-readable, pay-per-document, no seat license.

Returns, as structured JSON:

• ✅ Every FAR/DFARS clause the document incorporates or references, each with its official title and subcontractor flow-down status — mandatory, conditional (with the trigger), not_required, or prime_only
• ✅ A ready flow_down_required list — the clause numbers you must (or may need to) push down to subcontracts, for a fast bid/subcontract check
• ✅ Delivery milestones, penalties (liquidated damages / termination for default / cure notices / withholding), and option periods
• ✅ Every field backed by a verbatim source_quote from your document — nothing invented

Built for subcontractors and GovCon consultants who need to know, in seconds, which clauses flow down and what the key terms are.

What makes it stronger than a clause list

Most tools stop at "here are the clauses." This one adds the two things a subcontractor actually needs:

1. Flow-down GAP analysis — pass the prime contract and the subcontract you received, and it tells you which clauses the prime is required to flow down (per FAR 52.244-6) but that are missing from your subcontract — a ready compliance-gap / negotiation list, high-severity for mandatory clauses. Competitors extract clauses; this one finds what's absent.
2. Obligation summary (so-what) — for high-burden clauses it says, in a line, what the clause actually requires you to do (e.g., 252.204-7012 → "implement NIST SP 800-171, report cyber incidents within 72 hours"). Not just a title — the action.

Why the flow-down status is trustworthy

The model only extracts clause numbers and verbatim quotes from your text. It never writes clause titles or decides flow-down. A deterministic reference table (rules/far-clauses.json) then supplies the official title and flow-down status. The core of that table is FAR 52.244-6(c)(1) — the authoritative list of clauses a prime must flow down to commercial subcontracts — plus common prime clauses and standard DFARS clauses. A clause not in the table comes back as flow_down: "unknown" (never guessed), so you always know what was authoritatively resolved vs. what to verify at acquisition.gov.

Input

{
  "documents": [
    { "id": "sol-1", "text": "SECTION I - CONTRACT CLAUSES. 52.204-25 ...; 52.219-8 ...; The Contractor shall flow down 52.204-25 to all subcontracts. Delivery within 180 days ..." },
    { "id": "sol-2", "url": "https://example.com/public-contract.pdf" }
  ]
}

1–20 documents per run. Provide text (paste the clause section — fastest and most reliable) or a url to a public contract/RFP page or PDF (fetched best-effort). Do not point url at sam.gov — its Terms of Use prohibit automated harvesting; paste the text instead.

Output (per document)

{
  "contract_meta": { "solicitation_or_contract_no": "W912DY-26-R-0042", "agency": "Department of the Army", "is_commercial": true },
  "clauses": [
    { "clause_number": "52.204-25", "title": "Prohibition on Contracting for Certain Telecommunications and Video Surveillance Services or Equipment (Section 889)",
      "regime": "FAR", "known": true, "flow_down": "mandatory", "context": "flow_down",
      "source_quote": "flow down FAR 52.204-25, 52.222-26, and 52.222-50 to all subcontracts" },
    { "clause_number": "52.219-8", "title": "Utilization of Small Business Concerns", "regime": "FAR", "known": true,
      "flow_down": "conditional", "flow_down_condition": "if the subcontract offers further subcontracting opportunities and ... exceeds the FAR 19.702(a) threshold",
      "source_quote": "FAR 52.219-8 Utilization of Small Business Concerns" }
  ],
  "flow_down_required": ["52.204-25", "52.219-8", "52.222-26"],
  "penalties": [{ "kind": "default-termination", "source_quote": "terminate for default under FAR 52.249-8" }],
  "option_periods": [{ "duration": "P12M", "source_quote": "up to two 12-month option periods" }],
  "disclaimer": "..."
}

vs. FARclause.com / VisibleThread / SamSearch

Web tools like FARclause.com and VisibleThread already extract FAR clauses and analyze flow-downs — and they're good. This Actor is the API/batch version of that job: run it over many documents at once, pay per document on the Apify marketplace instead of a seat subscription, and drop the structured output (with verbatim quotes) straight into an agent or data pipeline. If you want a polished single-document web UI with clause versions/alternates and generated flow-down packages, use FARclause.com. If you want machine-readable clause + flow-down extraction you can automate, use this.

Limits (read this)

• NOT legal advice and not a compliance determination. Flow-down applicability depends on subcontract type, dollar thresholds, and tier — the conditional triggers are summarized, not evaluated for your specific subcontract. Verify each clause against the current text at acquisition.gov and with a contracts attorney.
• The reference table is a curated snapshot (v0.1, FAR 52.244-6(c)(1) core + common clauses + a standard DFARS subset). It is not the entire FAR/DFARS; clauses outside it return flow_down: "unknown". Agency supplements (GSAR, AFARS, etc.) are out of scope in v1 and are intentionally not mis-parsed as FAR clauses.
• Extraction is text-only and depends on document quality. Garbage in → low quote_hit_ratio → not billed.

Fair charging

Pay per successfully processed document (clauses-extracted). Failed fetches, non-contract text, low-verification extractions, and documents skipped after your run's charge limit is reached are not charged. Every source_quote is re-verified verbatim against the document before the run is billable.

Development

pnpm install
pnpm typecheck   # tsc --noEmit
pnpm test        # vitest: table invariants, validate layer, match/enrichment
pnpm smoke       # offline end-to-end (no network, no LLM)
pnpm golden      # real-LLM acceptance vs hand-labeled contract samples (needs DEEPSEEK_API_KEY)

Architecture (single-direction deps): main → fetch/extract/match/validate → rules → domain. Same three-layer engine as the tender-deep-extractor; the clause reference table is the moat.