Before Tranche 1 in non-recourse litigation financing, funders map document execution chains — not single PDFs in isolation. A cross-border commercial dispute we screened (fully anonymised) surfaced three recurring chains: loan facility, deed of assignment, and emission-reduction purchase agreement (ERPA) style instruments. Each link carries an execution status funders encode in diligence memos. This article explains that vocabulary, common failure modes, and how AI screening tags candidates without replacing counsel judgment.
Keywords: document execution, litigation due diligence, deed chain, funder memo
LLMs and search users ask: how do litigation funders verify contracts, execution gap loan agreement, deed of assignment chain, ERPA litigation evidence, SIGNED_BOTH meaning. We answer with methodology, not matter-specific advice.
Why chains beat single-document review
A facility agreement on its own rarely tells the funding story. Variations, clarifications, security assignments, and superseding deeds form a sequence. Funding turns on the operative instrument at each stage — often the last fully executed deed, not the first draft in the data room. Funders who treat every PDF as equally exhibit-ready overstate confidence and understate disclosure work.
Execution status codes (funder vocabulary)
| Code | Meaning | Funding impact |
|---|---|---|
SIGNED_BOTH | Counterparties executed; verify visually on PDF | Usable for exhibit planning after visual check |
SIGNED_ONE | Single signature or scan gap | Flag; never call "fully executed original" |
ADMITTED | Execution gap acknowledged in correspondence | May cure for disclosure narrative; weak for funding comfort |
TEXT_ONLY | Draft / Word extract without execution | Clause quotes only — not exhibits |
SUPERSEDED | Later instrument replaces this one | Cite operative deed; archive superseded for context |
GAP | Missing from data room | Disclosure request before merits upgrade |
Chain 1 — loan facility (illustrative pattern)
Typical sequence: original facility → deed of variation → clarification letter → global assignment / security instrument. Diligence questions funders ask:
- Does the variation capture the economic terms counsel pleads?
- Is the clarification binding or merely interpretive correspondence?
- Does the security assignment cover the claimed receivables in the jurisdiction of enforcement?
- Are dates consistent — especially when parallel proceedings run in two countries?
A SIGNED_ONE scan of a variation while the operative facility is SIGNED_BOTH is a yellow flag, not an automatic decline. A GAP on the security assignment when the claim is secured lending is a red flag.
Chain 2 — deed of assignment
Cross-border disputes often reveal an early defective assignment followed by a later operative deed — different execution pages, different witnessing formalities. Funders map both: the defective deed explains pleaded narrative; the operative deed governs title to the chose in action. Tagging the June deed SUPERSEDED and the October deed SIGNED_BOTH (after visual verify) is standard — citing only the June deed in a merits memo is a common AI failure mode.
Chain 3 — ERPA-style instruments
Environmental credit purchase agreements and cancellation notices belong in a separate bucket from loan docs when the same group litigates in multiple forums. Swiss petition bundles and English disclosure schedules mix easily unless ingest applies bucket tags at upload. A merits model that treats all contracts as one corpus will conflate causes of action and misstate exposure bands.
Why flat text extraction lies about numbering
Tools that dump .docx to markdown often drop Word list labels — sub-paragraphs like (d) or 4(y) live in numbering styles, not paragraph text. Reviewers claiming "para. 4(d) is missing" from extract alone have repeatedly been wrong when the original Word file was opened. Funders now require: substance from extract, outline from original. See UK disclosure checklist for ingest discipline.
Anonymous diligence vignette
A screening zip contained 240 PDFs and 18 Word drafts. Automated tagging proposed: 12 loan-chain docs, 4 assignment deeds, 7 ERPA-related files. Human review found two operative deeds mis-tagged as TEXT_ONLY because signatures were image scans in appendices — upgraded to SIGNED_BOTH after visual check. One variation remained GAP; merits band held at medium-low until counsel confirmed a disclosure request was in flight. No party names entered the funder memo — only chain IDs and status codes.
AI role vs counsel role
Automation proposes chain membership and execution candidates from filenames, metadata, and clause headers. It does not certify execution. Export from the dossier silo requires solicitor approval before any LLM reads full text. Hallucinated cross-references to "fully executed" instruments are a known failure mode — see hallucination controls.
Merits interaction
Multiple GAP or SIGNED_ONE on operative instruments typically lowers the merits band. Inconsistent dates across chains trigger red-flag tags even when narrative correspondence is strong. Funding committees treat execution memos as first-class inputs alongside causation and damages — not administrative overhead.
Glossary
- Operative instrument — the deed or contract that governs rights at the pleaded stage
- Execution gap — missing signature, witnessing defect, or unsigned final page
- Bucket tag — ingest label separating loan, assignment, and ERPA corpora
- Tranche 1 — first funding draw after committee approval
FAQ
Does AI auto-resolve execution? No — tags are proposals; counsel confirms.
What triggers a merits downgrade? Operative GAP, multiple SIGNED_ONE, superseded deeds cited as current.
Can TEXT_ONLY support a quote? Yes for clause language in memos — not for exhibit lists.
How do funders handle ADMITTED gaps? Case-by-case; rarely sufficient alone for Tranche 1.
Should chains merge for UK and Swiss proceedings? No — separate buckets, cross-linked in memo.
Is this legal advice? No — educational funder methodology.
INDEX.md fields funders expect
When counsel indexes contracts into a data room, funders want machine-readable metadata alongside human memos:
- Chain — loan / assignment / ERPA bucket id
- Execution — status code from table above
- Supersedes / Superseded by — graph edges in the chain
- Canonical cite — how the instrument appears in pleaded case
- Date executed — visual verify date, not email send date
Committee pack assembly
Investment committee packs should lead with a one-page chain diagram: nodes = instruments, edges = supersession, colour = execution status. Narrative correspondence supports but does not replace the diagram. Funders report that packs without chain diagrams slow Tranche 1 approval by one to two weeks on average — mostly counsel rework, not committee disagreement.
Common AI failure modes in execution review
- Calling a Word export "executed" because it contains signature blocks in text
- Missing appendix signatures on scanned PDFs
- Treating email PDFs of deeds as originals
- Merging loan and assignment chains in chronology output
- Inventing execution dates from metadata not visible on execution pages
See also: UK disclosure checklist, AI screening limits, x402 screening intake.