How to read a brief

What it is¶

The thesis brief is the plain-English exhibit rendered from a thesis record. It is what a carrier reads after a run, alongside the thesis JSON and the receipt. Every sentence on it comes from a template record, every paragraph points back into the record, and nothing on it is free text .

The acceptance test for the page is a target brief the expert writes first, for the Morales worked example, as it should read. The renderer must match it sentence by sentence, and where the template differs, the template changes . The sections below are in the order the target defines.

Determination¶

The brief opens with the referral decision. The enum is refer, investigate or no referral, with conditions listed, and hold carried as a separate field. Seven reason codes explain the decision, including insured at fault and no first-party payment. On Morales the determination is refer with conditions: two routine conditions, no material ones.

Read the conditions. A material condition (a statute of limitations bar, sovereign immunity, act of god, spoliation) changes what an adjuster does next. A routine verification (waiver of subrogation) does not.

What it rests on¶

The primary theory, with the facts it cited. On Morales, respondeat superior is confirmed against the platform company by a demand letter, and the at-fault third-party driver is confirmed with intercompany arbitration as the channel because both carriers are members. The channel is a carrier-membership fact seeded at tenant install, not a state fact.

Inputs considered¶

Every field each theory and gate read, one row per field, with one of three words: Stated with its value and speaker, Not stated, or Not supplied. Decisive fields are marked. This is the completeness ledger rendered for a reader, and it is the section that makes the brief checkable against the payload: the receipt returned before the run must predict it exactly .

Not stated and Not supplied are never swapped. Not stated means the carrier reviewed the field and found nothing; a rule may act on that. Not supplied means the carrier said nothing about it; no rule may. The Morales sample receipt reads 29 stated, 7 checked and not present, and 23 not supplied, and those 23 include the waiver and the five fleet negligence facts, which is why the waiver stays indeterminate and fleet negligence stays open .

Other theories¶

Each theory not on the thesis appears by status with the missing attribute named. An indeterminate theory names the fact that would resolve it; an excluded theory names the fact that excluded it. On Morales, fleet negligence is indeterminate because none of the five narrative fields it can read was supplied; negligent entrustment is excluded because no owner other than the driver exists; product liability is excluded.

The sentence matters. "No vehicle owner is on the record" was ruled false on Morales, because the driver owns the vehicle; the exclusion is that no other owner exists. A brief that names the wrong reason for the right outcome fails the target.

What could defeat it¶

Every gate and defense the corpus holds, with its outcome. A cleared gate reads "does not bar recovery", never "did not apply", because a gate that was evaluated and cleared is a different fact from one that was not evaluated. On Morales the comparative fault gate clears on a police determination and a citation against the other driver, and the waiver is indeterminate. A first-party payment gate that renders "did not apply" on a claim with a stated payment is a meaning error, not a cosmetic one.

Money¶

Money is rendered as heads, never as a single figure. Each head is stated, computed from stated inputs, pending (rate and limit only), excluded, or indeterminate. Nothing is estimated. On Morales the base is 18,400 with a 500 deductible shown, the rental head is pending because the policy states 40 per day and a 30-day limit only, and the medical reserve of 28,000 to 35,000 is excluded from the subrogation base .

Two figures carry the recovery: the recoverable base and the legally recoverable amount. The second stays pending until a stated fault percentage or a liability determination supplies the adjustment. There is no confidence percentage on any figure, and every figure must be traceable to the source payload; a seed figure with no source in the document is a seed defect.

Deadlines¶

Only deadlines for causes on the thesis appear, each with its anchor shown. On Morales that is one row: the Texas negligence statute of limitations, two years from the date of loss 2025-07-22, falling 2027-07-22. A deadline with no anchor date is never computed; the obligations section carries a sentence saying deadlines will compute once the anchor is available and that this is not to be read as no deadline. A row for a cause with no theory on the thesis is a resolver defect .

What would change it¶

Derived from the indeterminates: the facts that, if supplied, would move a theory, clear a gate or resolve a head. It tells the adjuster what to ask for.

Provenance¶

The payload hash, the spec versions, the ruleset hash and the corpus snapshot version, plus the curation status of each jurisdiction section read. Where a state section is drafted rather than verified, the status prints on the provenance line. The record and the render carry the same run_id, and both come from the same seed run .

Reading a Property brief¶

A Property brief has the same sections; six things read differently .

Vocabulary. Building, business personal property, and business income and extra expense, never the data's dwelling, contents and ALE. Never "employer", "vehicle", "repair estimate" or "above the deductible". Theories by plain name: contractor negligence, product liability, utility negligence, tenant negligence, municipal negligence, neighbouring property.
Two waiver lines. The contract waiver and the lease waiver each show their own status: stated false reads "none on record"; not stated reads "reviewed, nothing recorded"; omitted reads "not supplied. Check the policy." or "not supplied. Check the lease."; stated true reads "on record" with the scope result. Page one and page two of a case file must agree, and there is no third generic waiver row.
Fault rule. Usually not_applicable, because the carrier format carries no Property fault facts. Where it is printed, the label comes from the state's fault_system record.
Act of God. One row. A natural event with no party cause bars; with a party cause it reads as a condition to confirm the natural event was not the sole cause.
Money. Heads shown separately and summed into the base; no deductible figure, and the line ends "Confirm the deductible."
Deadlines and what is not assessed. The property-damage limitation row only, reading "within the limitation period" or "expired on" a date. A contractor theory adds "Construction time limits (statute of repose): not assessed." A government target adds that the notice-of-claim deadline is not assessed and must be checked immediately.

The receipt¶

The receipt precedes the brief. It is produced by the harness receipt mode from the same payload, runs no tool, and lists by tier what was stated, what was checked and not present, and what was not supplied, with which theory or gate needs each field, what falls outside the profile, and whether each jurisdiction section is published and each deadline computable. It carries no determination words. A receipt rendered from a hand-made fixture is on the list of mistakes not to repeat; it comes from a scenario-card run .

The glossary as a record¶

The glossary the carrier reads is a governed record, seeded as 39 glossary entries in reference_data and rendered on every brief. It is versioned with the templates in the corpus snapshot and covered by the drift check. Its definition of Stated was found to describe span-verified extraction, which is wrong on the structured path, and is logged for correction .

Every paragraph carries two attributes¶

Every rendered paragraph and table row carries data-trace, a pointer that resolves into the thesis record, and data-status, the outcome key, with a CSS rule per key. Any sentence resolves to the record field and rule outcome behind it, which is what makes the brief an exhibit rather than prose.

The rendering rules¶

Every sentence comes from a template record keyed by an outcome key; 46 sentence templates are seeded as reference_data beside the glossary. No engine basis, reason, text or verify string is ever printed. No identifier, enum literal, snake_case key, bracketed tag, .0 float, empty parenthesis or Python repr appears in prose. Labels come from the field guide record, never from title-casing a key. Status words come only from ledger templates. The renderer reads only the thesis record; nothing reads the trace. Twelve assertions enforce these on every render .

The sentences must read as an adjuster would write them. A cleared gate is "does not bar recovery". An exclusion names the fact, not the record.

The false-affirmative bar¶

The false-affirmative class is the only defect category that loses a proof of concept outright; everything else produces a recoverable gap. Three were found in one audit: recommended actions closing with no escalation required when every upstream section had errored; a coverage verdict of covered reached with zero exclusions evaluated because an empty catalogue read as no exclusions applying; and no indicators fired printed after the detector was handed empty content.

The bar for any page is therefore: a brief full of honest gaps passes, and a confident brief with one fabrication does not. Could-not-evaluate and considered-and-rejected must never render the same. A figure that contradicts its own qualifier is read as the figure, so a reserve of zero beside zero percent completeness says the claim is worth nothing, where "not established, 0 of 5 heads resolved" says what is true.

Open items¶

The Morales brief is accepted in all sections except the six What Could Defeat It rows, the negligent entrustment and fleet negligence sentences and one pronoun, with replacement sentences issued; drift output and snapshot version are required before acceptance.
The engine render of the carrier sample payload has not yet been diffed sentence by sentence against the target.
Glossary entry for Stated needs correction for the structured path.
The 52 remaining seeds render under the same templates after Morales is accepted, followed by a UX pass on four renders.
The Property wording rules above are enforced by SL-FIX-07P; the 58-file Property case-file run of 2026-10-08 still showed fault-rule label, waiver row, gate wording and deductible defects.