RESEARCH

The problems we are solving to build a decision engine.

What Voltaire is building, and the paradigms it is built on. Written as working notes, not marketing.

PARADIGMS

The decision object

Situation, move, confidence, evidence, test, outcome. Every decision is made from a record with its prediction attached.

Deterministic reasoning

The model reads and drafts; code decides, gates and scores, and every conclusion traces to the rule and the records behind it.

The human gate

Every consequential move is surfaced for judgment and reviewed by a person. The engine ranks; human experts decide.

The compounding graph

Measured outcomes write back into the graph. Every recorded decision makes the next ranking better.

RESEARCH AREAS

DECIDING

01
Which outcomes a company should own, proposed from its own history and market, with odds that are defined, recorded before the action and scored after.
02
What a confidence figure may claim at each evidence tier, and what the screen refuses to say.

LEARNING

03
What the engine would have caught last quarter, as of each date, with the misses listed.
04
Replaying recorded runs before any change ships, scored per action class.
05
Turning a correction in Slack into a test the next version has to pass.

ACTING

05
Autonomy per action class, earned from approvals, edits and measured outcomes, under ceilings people set in code.
06
Reading untrusted text without letting it steer a write.
07
A ledger that records the prediction before the effect, and a reconciler that freezes the engine on any effect it did not authorize.

RUNNING

07
An engine that lives for months: state persisted, runs rotated, one clock for production and replay.
08
Connectors kept fresh with scoped credentials, webhooks reconciled by reads.
09
One accountable director per outcome, delegating to specialist agents behind a budget gate.

WRITING

Founding note21 July 2026

Why Voltaire exists

I could not answer a simple question about my own company: which growth path was the right one. Nobody could sell me the answer. That gap is the company.

Research22 October 2026

Autonomy is earned one action at a time

An agent that holds the keys to billing and the inbox should not get more freedom because people clicked Approve. Here is the rule we use instead: four numbers, three of them set by people, and the evidence only moves one.

Research22 October 2026

Three tests before an engine earns a decision

A backtest, a replay and a forward ledger answer three different questions. Most systems that act on a company answer none of them first. Here is what each test can prove, what it cannot, and the order we run them in.

Research22 October 2026

What a confidence figure has to mean

A card that says a move is worth $1.4M at 92% confidence is making three claims at once. Here is what each one has to be backed by before we let it on the screen, and what the screen refuses to say.

Research24 July 2026

The knowledge nobody wrote down

Every company has a system that seven people know how to use. That knowledge is the reason most automation stalls, and it is the thing an agent has to be given before it can be trusted with anything.

Analysis17 July 2026

The pay cut problem

Your law firm is not slow to go AI-native because it lacks talent. It is slow because going AI-native requires the partners to vote for a smaller cheque.

Research14 July 2026

Memory an agent can reason with

Most AI memory is a pile of text fetched by resemblance. If software is going to spend your budget, memory has to be something the system can compute with, not something it has to re-read.

Research9 July 2026

How a small brain watches a large business

A reasoning engine is fast because it holds everything in memory, which means it cannot hold much. The trick is to stop shipping it data and start shipping it answers.