Valuation and strategy: capabilities and limits

The model is authored from the record

• Every line in the claim tree names where it came from

  Each edge in the claim graph carries its derivation, down to the field it was read from, and a basis saying what it does not assert. Nothing on that read calls a model: the adversarial lanes are stored artifacts, read back rather than written again.

• Two lists per element, and a denominator

  An element's proof state is what establishes and what defeats it, each with its verbatim quote and page, against how many elements the forum defines. The two are never netted into a score. A model's reading stays a model's, the attorney's disagreement beside it.

• An empty comparable pool is not a finding

  No comparable cases found is a claim about the world, printed where the forum could not be derived. Each absence is named: no forum key, none in this forum with the number examined, or a pool the tenant and verdict-facts gates emptied after counting.

The number, and the width around it

• You cannot get a point estimate

  A currency figure not bound to a second, different figure raises. It parses magnitude words, so the same number written two ways is caught, and it requires the ends to differ, after a real matter produced an identical low, middle and high.

• Five authors, and the median model kept whole

  Extended reasoning cannot be pinned to a temperature, so five authors run and the median member's model and simulation publish whole, not blended. A run where four of five failed once published a median reading like a healthy one, so every snapshot carries the count.

• An interval, never a point

  Shipped

  The honest interval can only widen the simulated band, never narrow it, and the method that produced it travels with the number.

• Thin data widens the range and can never tighten it

  Outstanding discovery widens it: the half width divides by the square root of the share of served requests returned, while a fully answered record is left untouched. A thin comparable group pools toward the collective, and a fact pattern far from the corpus widens.

What bounds the dollar

• A floor under the cost the model authored

  What the matter has cost is read from billed invoices, work in progress at matter rates, and expenses. It raises an authored figure below it and never lowers one above. On contingency, three empty ledgers are an absence of record, not a cost of nothing.

• A policy limit counts only whole

  A ceiling exists only when every active claim carries a limit on a known basis. One unknown limit yields no ceiling and reports how many, because a fabricated ceiling under real coverage is worse than none. What is known is a floor, never a ceiling.

• The clock that discounts every dollar

  Present value falls with the months to resolution, authored by a model behind a posture regex that had never read a docket or scheduling order. A trial date on the record now floors it, and only floors it: a longer timeline is left alone.

Where the case actually turns

• One kill shot per matter

  Shipped

  Four surfaces once answered what single thing decides this case and disagreed on the same matter. Now there is one answer, and it is stored.

• Ranked by what moves the outcome

  Factors are ordered by outcome-determinativeness from figures already held: the simulation's variance attribution, how many claims an issue gates, and the clause defeats the record carries. No model call does the ranking. On a small-dollar matter discretionary doctrines are demoted with the reason written down.

• Adversarial war-gaming

  Shipped

  The opponent wins each pivotal issue in turn, under shared random draws so the comparison is not simulation noise.

• The Red Brief

  Shipped

  The case against your own position, written from the same record.

The other side, and the bench

• The distribution judges the offer, never the reverse

  Every demand and offer is an immutable row priced against the band live that day; a correction appends rather than edits. None of it reaches the sampler: simulated outcomes are identical with the ledger and without it. An offer is reported against the finished distribution.

• A walkaway, a zone, and a curve

  Off the stored snapshot, the engine derives the certain payment worth as much as trying the case, the mirror for the other side, the zone between, and the chance trying beats resolving at each rung. It flips for a defendant; the portal never sees it.

• What a docket records, and what it cannot

  Motion rulings are read deterministically out of the court's operative sentence. An ambiguous, split, procedural or second-hand opinion is skipped, never resolved to a coin flip. A rate carries its sample size, the level of bench, and an interval wide while the record is thin.

Review, and going on record

• Two disciplines, and a review that expires with its version

  Two assessments, technical and attorney, are recorded separately against one exact valuation version, append-only, audit row first. A moved valuation leaves an existing review reading as an earlier version, and where freshness cannot be established the standing reads unknown. A queue lists the unassessed.

• The larger error bar, measured

  Sampling error on the quantiles is the smaller of the two by orders of magnitude. The larger is re-authoring: the same matter run again with nothing new moves the band. A transition with no record event between readings is reported as model noise.

• Provable precedence

  Shipped

  A prediction is committed to an append-only ledger before the outcome is known, which is what makes a later score mean anything.

Limits

• This is an estimate, not a measurement. Nothing here has been scored against a resolved outcome.
• The band is calibrated and widened by disagreement. That is not a claim of coverage at a stated level.
• The default outcome model averages over liability, so only the mean is sound and the published limitation sentence travels with any percentile.
• The public calibration scoreboard ships switched off, and it has nothing to publish yet.
• The docket corpus records how written motion opinions came out. It is not a read on how a named judge will rule, and written opinions are not the universe of motions decided.

What BRON refuses

• Publish a calibration result.

  The ledger holds no scorable outcome, publication refuses below ten resolved matters, and the public scoreboard ships switched off. It states the insufficiency instead of filling the space with a number.

• Narrow the band because the calibration set is thin.

  A thin set returns an infinite width before any clamp, so it degrades into a wider honest interval rather than a falsely tight one.

• Call a valuation fresh because nothing flagged it.

  Absence of a signal is not evidence of health. A snapshot whose provenance cannot be established reads unknown rather than fresh, and the engine version is stamped on it, so figures produced by superseded code read stale instead of being served quietly.

• Hide a cap that bound the outcome.

  Where a coverage ceiling applied, the share of simulated worlds it bound and the mean amount it removed are both reported. A cap applied without saying so is its own dishonesty.

• Let a what-if move the published band.

  A counterfactual clones the model, perturbs one lever and reports a delta against the baseline under the same draws. It persists nothing, and it is never folded into the interval on screen.

• Feed the other side's offer into the simulation.

  An offer measures the opponent's estimate and their appetite, not the matter. Anchoring on it would let one lowball drag the whole distribution down and hand the firm back a number that was partly their opening move wearing our label.

• Produce a judicial read for a court it cannot map.

  A forum that cannot be reduced to a docket-corpus court identifier returns nothing rather than a corpus-wide federal rate under a state court's name. The live book is state matters, so nothing is what it returns today. The corpus is queried on neutral terms and never on the outcome phrase, because sampling on the thing you are measuring returns a corpus that agrees with you.

• Figure every matter's damages the same way.

  A liquidated debt, an injury schedule, a back-pay claim and a trade-secret royalty are different arithmetic under different law. One generic rule once let a small invoice claim anchor to a far larger verdict, and statutory caps are described rather than hardcoded.

• Re-author a pinned valuation on its own.

  A pinned matter returns the frozen model, and if that model cannot be read it raises rather than quietly authoring a fresh number.

Why would I put a number a model wrote in front of a client?

Claims, elements, costs, coverage and comparable outcomes are authored into one model, simulated, and published as an interval that carries what widened it.

Inputs

• The contested element ledger and established facts
• Comparable outcomes and researched authority
• Billing, coverage and damages records
• Served discovery, deadlines and the docket corpus

Outputs

• Settlement posture and the negotiation ledger
• The Red Brief and the war game
• The client conversation
• The committed prediction ledger