Mapping a Reg E Dispute to a Reasoning Graph
“Should we credit this dispute?” is a compound question, and Regulation E has already done the decomposing: whether the claim is a covered error, whether notice was timely, what the investigation clock requires and when provisional credit is owed, and how liability tiers apply — each element with its own evidence and its own deadline. This post walks the transcription: statute elements become graph nodes, each scoped to its own evidence; the statutory clocks become a backward-chained schedule; and the coverage gate refuses any determination that left an element unanswered. The payoff is an artifact examiners can read in the regulation’s own structure — because the graph’s parts are the rule’s elements.
The compound question
A cardholder calls about a charge she says she never made. One question arrives — do we make her whole? — but nothing about it is one question. Under Regulation E’s error-resolution framework (12 CFR §1005.11), the determination decomposes into elements that are answered from different evidence, on different clocks, with different consequences for getting them wrong. A reviewer who answers in one motion has still answered all of them — silently, unauditably, and with any error infecting the whole. This is the pattern our decomposition strategy exists for, and Reg E is the cleanest worked example in consumer finance, because the regulation itself is the decomposition.
The elements, as nodes
Transcribing the framework produces a graph whose first act is naming its parts — and the parts are the regulation’s:
| Element | The question | Its own evidence |
|---|---|---|
| Coverage | Is this a covered “error” — an unauthorized EFT, an incorrect transfer, a computational mistake — on a covered account? | Transaction records, account type, claim description |
| Notice | Was notice received within the window the rule measures from transmission of the periodic statement? | Statement dates, contact logs, claim date |
| Authorization | Does the evidence support that the transfer was unauthorized — or that access was furnished? | Device and location data, authentication records, prior activity |
| Clocks | Which investigation timeline applies, and is provisional credit owed while it runs? | Claim date, account age, transaction type |
| Liability | Which consumer-liability tier applies, given when the loss was reported? | Report timing relative to learning of loss and to statements |
| Determination | Assembled from the above: credit, deny, or partial — with the required notices | All of the above, recomposed |
Two properties make this more than an outline. Each node is scoped: the authorization step sees authentication records and device data, not the notice dates — so “probably unauthorized” cannot lean on “filed promptly, seems honest.” And each answer is evidenced: the node’s output cites the records it rests on, which is what lets a later challenge land on one element and leave the others standing.
The clocks are a graph of their own
Reg E’s timing rules interact: the investigation window depends on facts about the account and transaction; the longer windows are available only if provisional credit issues on time; the notices have their own deadlines measured from completion. That structure is exactly what our working-backward strategy formalizes — fix each obligation’s date, chain the prerequisites in reverse, and surface the implied internal deadlines while they are still ahead. In practice the clock subgraph runs alongside the elements: a determination that is substantively right and procedurally late is not right, and the graph treats the schedule as part of the determination rather than as project management around it.
The gate, and what the examiner reads
The coverage gate at the end enforces the contract the decomposition made at the start: every named element answered, every answer citing its own evidence, the recomposed determination addressing the claim as filed — and the whole record carrying its clock history. What an examiner receives is not a narrative reconstructed after the fact but the determination in the regulation’s own shape: element by element, the same way an exam workpaper would test it. The reviewer’s attestation sits on top, with anything they questioned preserved in the record — the human making the determination, the system making the determination defensible.
One honest boundary: transcribing the framework does not decide the hard cases. Whether furnished access defeats an unauthorized-use claim on a given fact pattern is judgment, and the graph’s contribution is to isolate that judgment in a named, evidenced, challengeable step — not to dissolve it. The rule supplies the structure; the structure disciplines the judgment; the judgment remains visible. That division of labor is the whole design.