Skip to content
Evidence / CalibrationOpen access · Freely available
The coverage record

Uncertainty should
be measured, too.

If a range claims 80% coverage, does it actually contain outcomes about 80% of the time? We publish the record so you can inspect the method's behavior—not just its intended behavior.
Current service audit. This panel is loaded from the live service. Check its observation window, map date, sample size, and sequential verdict. A warning or unavailable record should not be read as a pass.
Loading…
Rolling service record · 5-day band
This is a dated, rolling record of past predicted-versus-realized coverage, not a guarantee and not evidence of trading edge. Inspect the full receipt.
Interpreting the evidence

Two records.
Two different questions.

01 / Rolling service record

Is it holding up now?

The rolling record follows published bands as their outcomes settle. It can reveal deterioration or condition-specific failures that an older validation result would miss.

Use the dated panel above. A sequential breach is a reason to investigate or withhold the uncertainty claim.

02 / Frozen evaluation

Did this method work
on this test?

A held-out evaluation examines a defined method on a fixed sample and protocol. It is evidence about that experiment, not a substitute for current service status.

Read the frozen evaluations
Dated research finding · September 4, 2026

Why a condition-specific check matters.

Study 103

Gap-day calibration
Held-out sample: 385 anchors
July 2024–August 2026

On the study's held-out gap-day sample, the original five-session band covered 70.1% of outcomes. A multiplier of 1.48, fitted on an earlier sample, increased coverage to 82.6%.

This is a specific, dated finding—not today's overall coverage rate. The correction addressed under-coverage on non-earnings gaps; it does not establish that every market condition is adequately covered.

Read the protocol and result
Before using a range

Coverage is only one part of the question.

Read the width.

A wider interval can cover more outcomes while providing less information. Evaluate coverage and width together.

Check the condition.

Aggregate performance can hide weak slices. Earnings proximity, gap days, and the requested horizon deserve their own checks.

Keep the claim narrow.

A band at one horizon is not an entire-path guarantee. Calibration does not establish direction, trading profitability, or execution feasibility.

Raw live audit → · Methodology · Research ledger