- Ingest, compile, structural parsing, reference resolution
- Adversarial-mechanics classification and ambiguity detection
- Vulnerability scoring via missing-blocker analysis
- Evidence-linked ROUTE output for expert review
Compile first.
Reason second.
DAC turns high-stakes credit agreements into machine-computable behavioral models so CLO managers and credit teams can surface adversarial liability-management routes — and see what the document actually permits before capital is at risk.
Γ ⊢ φFrom the premises Γ, the conclusion φ follows.
- Must / Obligation
- May / Permission
- Cannot / Prohibition
- Contrary-to-duty
- Defeasible implication
- Does not prove
A formal system with a company attached
Not an AI wrapper on PDF text. ClauseFlow compiles the rule system first, then reasons over structure with deterministic validation gates.
Ingest
Credit agreements enter as documents. Stable identifiers attach to clauses before any model is asked a question.
Compile
The compile engine builds a clause tree and Agreement Behavior Graph: parties, permissions, duties, exceptions, thresholds, events, dependencies, evidence.
Reason
Frontier models reason over the compiled graph — not raw PDF prose — and emit ROUTE objects with outcomes, enabling provisions, and missing protections.
Validate
Deterministic gates check grounding, structural invariants, reference resolution, and chain completeness. The model is never the source of truth.
From the four corners, this route follows
The turnstile is not decoration. In logic, Γ ⊢ φ means the conclusion is entailed by the premises. That is exactly the product: from the agreement as written, which adversarial paths are open, and which protections fail to follow.
Color only marks logical state — assert, defeat, discharge, structure. Marketing cannot drift into gradient-and-glow startup costume because the system will not allow a color that does not mean something.
Portfolio query and cross-document aggregation — the layer that makes institutional memory and coverage licensing real for a CLO book. Per-document depth is shipping; portfolio scale is the next load-bearing product surface.
Findings as sequents
Every finding carries a one-line formal statement above the prose. The formal line is something no reorg deck and no generic LLM wrapper can print honestly.
Γ ⊢ prime(MinorityLenders) ∧ Γ ⊬ P(block(MinorityLenders))
Missing-blocker analysis: an unrestricted subsidiary path can release liens without a protection that follows from the four corners.
Γ_orig ⊢ pathway(Serta-style uptier)
Backtested on six landmark LME situations using origination-era text. The structural pathway later became economically significant — finished historical evidence, not a production recall claim.
output ⊢ reviewable first pass
Every claim stays checkable against the compiled graph and source text. The product is built to be reviewed, not believed.
Built for the people who underwrite the book
CLO credit & portfolio teams
Beachhead users and economic buyers. Portfolio-scale structural questions across hundreds of credits are the demand signal already observed.
General Counsel as gate
Security, document permissioning, isolation, retention, and data-flow answers are part of the offering — not afterthoughts for procurement.
Credit PMs more broadly
Same core product expands to loan funds, BDCs, and insurance credit books once the CLO production load is real.
Request a fixed-fee portfolio diagnostic
Fifty to one hundred credits. Same order of money as a per-deal pilot, structured to demonstrate portfolio coverage rather than teach transactional pricing. Founder-led; security package ready for GC review.