Apex Predictive
A ring of twelve tall dark monoliths in a black void, each with a thin seam of pale light, facing one small bright golden point at the center.

The Engine

A circle of experts, and a gate.

The engine is not one brilliant model. It is many careful ones, kept apart on purpose, and a rule about when their agreement is allowed to become a prediction.

Why a circle

Errors that do not correlate cancel.

Every model has blind spots. A single forecaster's blind spot becomes your loss; a circle's blind spots, if the members are genuinely different, mostly fail to line up. So the engine's experts differ on the axes that matter — architecture, training lineage, reasoning method, and how each one weighs evidence against prior.

Independence is enforced, not assumed. Experts work the same question in isolation: same deadline, same access to the record, no visibility into each other's drafts. Agreement between them is only meaningful because nothing upstream manufactured it.

Deliberation Question → verdict

The week's cycle

The cycle is the same for a stock close, a storm, an election, a launch. Only the evidence changes.

  1. Frame

    A question worth asking

    Candidate questions are drafted in resolvable form: a binary or bounded outcome, a named source of truth, a deadline. If honest people could argue about whether it came true, it is rewritten or discarded.

  2. Deliberate

    Experts work alone

    Each expert researches the question with live data — entity corpora, prices, filings, records — through Verumdare's API and MCP tools, and files an outcome with a confidence and its reasoning.

  3. Converge

    The gate decides

    The verdicts are compared. Essential agreement at high confidence clears the gate; anything less — a split circle, a hedge, one confident dissenter — kills the question for the week. Most candidates die here.

  4. Publish

    Monday, 9:00 AM CT

    The ten strongest survivors make the board, each carrying its consensus count and its resolution terms. Publication is a commitment: the statement cannot be edited after it posts.

  5. Judge

    Friday, 4:00 PM CT

    Outcomes are read from the named sources of truth and every row is graded — hit, miss or push — on the permanent record. The graders do not work for the forecast.

  6. Learn

    The circle is scored too

    Every expert's calibration is tracked against results. Experts that drift get retrained or replaced; the gate's thresholds are set by measured performance, not by taste.

The hard rule

The gate is the product.
Silence is it working.

Anyone can publish ten predictions a week; the trick is that ours have to survive a filter that is allowed to say nothing. A week with three publishable convergences is a three-row board, not a ten-row board padded with coin flips.

We would rather be silent than interesting. The record is the only marketing this product gets.

That discipline is also why the record is worth grading. Hit rates only mean something when the misses are on the same page, in the same type, judged by the same clock.

Watch it run

The board is the engine's public face. Take the Monday email and watch the gate do its work.

Get the board

One email when the week publishes, one when it is judged.