Skip to content
Arrange a meeting

Product reference

Valuation and strategy: how it works

Compare settlement and trial from the saved authored model, costs and uncertainty. Then use war games and the Red Brief to challenge the position before counsel recommends the next move.

On this page

Return to valuation and strategy

Availability

Limited availabilityAn estimate. Its ranges have not been checked against resolved matters.
Supported work
Compare settlement and trial economics from the saved model, then challenge pivotal assumptions with war games and the Red Brief.
Start with
Settlement-versus-trial analysis needs a persisted authored valuation snapshot. Its negotiation ranges, best and worst alternatives, offer ladder and applicable fee-shift context use that same basis, with date and freshness.
Key limit
This is an estimate, not a measurement. No range has been scored against a resolved outcome.

Some capabilities have their own availability or configuration requirements, shown below.

Before you start

  • Settlement-versus-trial analysis needs a persisted authored valuation snapshot. Its negotiation ranges, best and worst alternatives, offer ladder and applicable fee-shift context use that same basis, with date and freshness.
  • Risk-aversion, certainty-equivalent and related acceptance comparisons appear only when the corresponding option is enabled and the saved model carries that information.
  • War games need an authored case model and an explicit run. The fact-to-element matrix supplies inspectable supporting and adverse context, but does not directly replace the model’s probability or publish a new headline range.
  • A matter with its claims, elements and the facts recorded for them. Each of the five estimates is drafted from those.
  • Billing, coverage and damages records on the matter.
  • A known policy limit on every active claim, if a limit should cap the range. One unknown limit means no cap.
  • Served discovery recorded on the matter. Requests that have not come back widen the range.
  • Discovery widening counts served requests and answers under a disclosed equal-request assumption. Drafted, unserved requests do not widen the band; no served requests is unknown posture, not completed discovery.
  • The preferred simulation engine is the Python service. An unavailable or refused service uses an in-process sampler and records the fallback; inspect the reported runtime and available diagnostics before relying on the result.

How it works

Inspect settlement economics from the saved authored model: claims, costs, coverage and uncertainty. Compare negotiation alternatives and recorded offers, then use war games and the Red Brief to challenge pivotal assumptions. Counsel chooses the recommendation; client authority, acceptance and closing are separate acts.

What does settlement mean against trial costs and uncertainty, and which assumptions should counsel challenge?

Settlement and trial economics

  • Derive negotiation work from the saved model

    The settlement analysis reads the saved valuation snapshot to prepare negotiation ranges, favorable and adverse trial alternatives, an offer ladder and applicable offer-of-judgment or fee-shift context. It carries the snapshot’s date and freshness; it does not silently author new dollar figures.

  • Keep optional risk preferences conditional

    When enabled and present on the saved model, risk-aversion analysis adds a certainty-equivalent and acceptance comparison. It is not an unconditional feature of every valuation. Missing inputs or stale figures need review before counsel uses them in negotiations.

  • Carry the recommendation to the right decision

    Internal negotiation posture stays attorney work product. Counsel’s recommendation, the client’s authority range and acceptance of specified terms remain distinct records. An accepted settlement can lead to a reviewed closing statement, but no recommendation or statement itself moves money.

From disputed fact to strategy

  • Use war games to choose the next evidence question

    An explicit war-game run reads the current authored model and recorded facts, then tests the consequence of the opponent winning a pivotal issue. In North Harbor, compare the notice assumption with Garcia’s and Smith’s accounts. A large conditional swing is a reason to examine receipt and reading evidence, not a measured probability that either account wins.

  • Read the Red Brief against the same sources

    The Red Brief develops the other side’s dispositive argument element by element from supplied record excerpts. Open the cited basis, compare versions and mark the argument for review. Evidence ingestion can schedule a debounced refresh; a changed claim spine may appear a cycle later. The result is adversarial work product, not a ruling.

  • Carry the priority into discovery and trial

    Use the disputed fact and the opponent’s argument to decide what request, deposition topic or scoped device question could test the assumption. Then connect the reviewed fact-to-element chain to the witness, exhibit and foundation needed at trial. The proof ledger does not directly rewrite the published valuation probability; scenario results do not change the live headline range by themselves.

How the range is built

  • Keep the middle estimate whole

    BRON authors the matter's claims, costs and coverage in five separate runs. It selects the middle eligible run by expected net outcome, keeping its model and simulation together rather than blending their percentiles. The returned-run count stays visible.

  • A range from p10 to p90

    Available

    BRON starts with the 10th to the 90th percentile of simulated recovery amounts, conditional on a recovery. That band is distinct from the chance of any recovery and from net trial economics after costs.

  • Disagreement widens the range

    The uncertainty adjustment can widen the simulated band but never narrow it. Without enough scorable resolved outcomes, it falls back to the returned estimates' spread, or the sampler band when that spread is unavailable. That fallback does not establish a measured coverage rate.

  • Outstanding discovery widens it too

    Served requests still awaiting answers widen the uncertainty band under a disclosed request-count assumption. Drafted, unserved requests do not move the figures. No served requests means the discovery posture is unknown, not finished.

  • Damp re-authoring drift without calling it evidence

    The dollar magnitude is AI-authored. A default-on magnitude anchor damps changes toward the prior published magnitude when monetary inputs are unchanged and a usable prior exists. Material monetary changes or counsel's explicit re-run pass through fresh figures. This does not validate the assumptions or turn the estimate into a measurement.

What bounds the figures

  • A cap needs every active claim's limit

    A policy limit caps the range only when every active claim has a known limit. Where a cap applied, BRON shows how often it bound the outcome and how much it took off.

  • Keep the range beside demands and offers

    Each demand and offer is recorded with the range that stood that day, so you can compare an offer with the estimate it met.

Testing your position

  • The issue the matter may turn on

    Available

    BRON proposes one issue the matter may turn on and stores it, so every screen shows the same answer. Whether it is the right issue is counsel's call.

  • If the other side wins a pivotal issue

    Available

    For each issue that decides at least one claim, BRON re-runs the simulation as if the other side won it, and shows how the range moves.

  • The case against your position

    Available

    The Red Brief drafts a dispositive motion the other side could file, element by element, from your own record. Each new version shows what changed since the last.

On the record

  • Each valuation is recorded before the outcome

    Available

    Before the outcome is known, each valuation is recorded in a log that is never edited, so it can be scored once matters resolve.

What it reads from

  • Claims, elements and the facts recorded for them
  • Billing, coverage and damages records
  • Served discovery, and what has come back
  • Demands and offers on the matter

Who can do what

  • Counsel decides the recommendation and retains negotiation posture as private work product. Client negotiation authority and acceptance are separate recorded acts; a reviewed closing statement and individual payment approvals follow their own process.
  • Counsel compares conditional swings, the Red Brief’s source-pinned opposing argument and version changes before choosing discovery priorities, a trial proof plan or a recommendation to the client.
  • BRON drafts the matter's claims, costs and coverage five separate times. It keeps the middle estimate whole and shows how many of the five came back.
  • The middle eligible model is selected by expected net outcome, with its simulation kept whole. Dollar magnitudes, claim probabilities and dependencies are model-authored assumptions for counsel to examine.
  • The default-on magnitude anchor damps drift toward a usable prior published magnitude when monetary inputs are unchanged. Material monetary changes or an explicit re-run use fresh figures; the anchor does not validate the model.
  • Counsel decides what to recommend and how to answer an offer. BRON proposes and compares.
  • Each demand and offer is recorded with the range that stood that day.
  • Each valuation goes into a log that is not edited, before the outcome is known.

What it produces

  • Settlement discussions
  • The Red Brief
  • The client conversation
  • The valuation log

Limits

  • This is an estimate, not a measurement. No range has been scored against a resolved outcome.
  • BRON does not claim that outcomes fall inside the range at any stated rate.
  • An AI model writes the dollar figures, so re-authoring with no new record can still move them. The default-on magnitude anchor damps drift where a usable prior exists; it does not eliminate authored uncertainty or establish outcome performance.

What BRON will not do

  • Give one dollar figure as the value of the matter.

    A simulation supports a range, not a point. Valuation text that names one dollar figure without a second, different one is refused.

  • Narrow the range because the data is thin.

    Thin data widens the range. The adjustment for disagreement can widen the simulated range but never narrow it.

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

    An offer reflects the other side's position, not the matter. BRON compares it with the finished range, and it never changes the outcomes simulated.

  • Assume a policy limit it does not have.

    One unknown limit means no cap. A cap built on a guessed limit could understate what is recoverable.

When something goes wrong

  • An absent or untrustworthy saved snapshot cannot provide a dependable current negotiation basis. Deriving figures from that snapshot does not validate its model-authored assumptions or resolve the evidence.
  • Red Brief refresh is scheduled and debounced, not guaranteed immediately on upload. A model-conditioned counterfactual is not a validated win probability, and no valuation range has been scored against a resolved outcome.
  • Valuation text that names one dollar figure without a second, different one is refused.
  • The adjustment for disagreement between the five estimates can widen the simulated range but never narrow it.
  • Conformal adjustment needs enough scorable resolved outcomes. Without them, the band falls back to ensemble spread or the sampler band; a fallback is not measured calibration. The conditional recovery range is distinct from the chance of recovery and net trial economics.
  • An offer is compared with the finished range. It never changes the outcomes simulated.
  • Where a policy limit capped the range, BRON shows how often it bound the outcome and how much it took off.
  • The dollar figures are written by an AI model, so estimating again with nothing new in the record can still move them.
  • No range has been scored against a resolved outcome. It is an estimate, not a measurement.
  • The North Harbor views are static illustrations of the recorded basis and an adverse assumption. They run no simulation, resolve neither witness account and supply no scenario dollars or fitted distribution.