We build systems that make an organization's judgment auditable and scored.
Companies keep a system of record for money, customers and operations. They do not keep one for strategic judgment. BRIER is that system of record: the conclusions an organisation commits to, locked, resolved against what happens, and scored.
Every vendor sells the model. BRIER sells the record.
An organisation that pays for AI-assisted judgment has no way to know whether it is getting better judgment or only faster prose. Running a model well is expensive, and consistency and trustworthiness cannot be judged without a record of what was concluded, when, on what evidence, and how it turned out.
The expensive decisions, buy this company, hold this position, enter this market, answer this competitor, are still made on slides, memory and instinct. An answer engine produces an answer and forgets the question. BRIER is built on the opposite premise: the answer is the least important output. The committed claim, the reasoning trail, the watch-list and the score afterward are the product.
- Set the board
The objective, the actors, the forces, the constraints and the live uncertainties, laid out as one map of everything that bears on whether the decision succeeds, on top of the organisation's own memory of what happened before and why.
- Run the futures
The situation is played forward many times against an opponent that pursues its own goals. The output is not one answer but a spread of outcomes with probabilities, the two or three points that actually decide the contest, and what would have to be true to flip them.
- Hear the panel argue
Several viewpoints look at the same situation, the strategist, the competitor, the economist, the skeptic, the historian, the operator, the red team, and an adjudicator shows the consensus, the dissent, the historical base rate and the reconciled probability side by side. Disagreement is shown, not hidden.
- Commit the forecast
The one button that matters. A committed claim is written to the ledger and locked by fingerprint: the claim in one sentence, the probability and its confidence, the evidence for and against, the conditions that would change the call, what to watch, where the answer will come from, and the date it will be judged.
- Get woken up
When new information moves a committed forecast, the owner is told what changed, why, what moved it, and what the options are. Executives read it the way they read the markets in the morning.
- Keep score
Outcomes are recorded separately from the locked claims. The Brier rule over what has resolved, the reliability diagram drawn from decile bins, unsmoothed. Nothing is called an improvement unless the held-out scorecard says so.
- The AI suggests; a named human approves. The runtime cannot reach an approve surface.
- No invented numbers. Every probability traces to a reference class and a record or is labelled a judgment weight.
- Locked forecasts are never edited. Outcomes are recorded separately.
- Nothing is called an improvement unless the held-out scorecard says so.
- Customer material is segmented. Shared learning crosses only the human-operated gate.
- The model is swappable behind one adapter. Closed dependency set; new modules standard-library only.
- Documents claim only what the code proves. Zero em dashes.
The Scorecard on this site is kept with the instrument BRIER sells. Every forward claim in the firm's research is locked in the brief's own words with its publication date, resolved only against a dated public source, and scored with the Brier rule over what has resolved. The firm's own calibration engine has scored 314 of its engagement calls by confidence band and domain since June 2026, and on 9 July 2026 it locked a seven-item forward benchmark that resolves from 18 September. Nothing in a locked claim is edited afterwards, and a claim that proves unresolvable is voided in the open rather than quietly dropped. The site you are reading is the demonstration.
Fewer than ten claims have resolved (n = 0). The diagram draws the points it has and is not yet a calibration record.
This is the Scorecard kept on this site, drawn from the same ledger, with the same instrument.
The instrument is industry-agnostic. An edition is the set of question types, resolution sources, certification criteria and voice that an instance registers for one kind of decision; the record underneath does not change.
The founders answer for it.
BRIER Systems is a division of Aegean Intelligence Group LLC, not a separate company. Every contract is with the firm, and the two founders stand behind each instance. Reach them on the Advisory page or write to either directly.