Withdrawn · 2026-10-14
The page you followed has been retired.
Earlier versions of this site presented Junctiq as an engine that scored markets and published forecasts, probabilities and a track record of its own calls. Those pages, the numbers on them and the programmatic endpoints behind them are retired. Their addresses now lead here; the endpoints answer 410 Gone.
Why
Junctiq does not publish forecasts, probabilities, scores or ratings of companies. A forecast asks the reader to trust Junctiq’s estimate of the future. A record lets the reader check the past: what a company published, when it was fetched, and what Junctiq can and cannot show about it. Junctiq’s governing document retires the forecasting measures by name and forbids bringing them back.
This track record has been withdrawn. It was not a measurement of anything.
Until 19 September 2026 this page published a calibration and accuracy record for a Junctiq forecasting engine. It is withdrawn in full. The figures are kept below rather than removed, because a company that sells custody of the record does not get to quietly edit its own.
- 01None of it was engine-verified. Production holds 325 committed predictions: 319 carry provenance_class = "seeded", 6 carry "hybrid", and none carries "engine_verified". The block headed "Engine Verified (Live)" was computed over seeded rows.Part X.10 — no synthetic contamination
- 02The classifier read a column the writer does not fill. It decided engine-versus-seeded from provenance_flags.origin, which is NULL on 311 of the 325 rows and equals "seeded" on none of them — and it treated a missing value as engine. That is also why the Seeded block read n = 0: every seeded row had already been counted as engine.Part X.1 — unknown must never silently become known
- 03A missing probability was replaced with 0.5. Of the rows counted, 161 carried no forecast probability at all. The code substituted 0.5 for each. A Brier score is a function of those probabilities, so the published 0.258 was largely a measurement of that substitution.Part X.1
- 04The outcomes were self-certified. "Correct" was read from lifecycle_state on the prediction record itself, rather than from an external adjudication in outcome_resolutions against a predeclared source and rule.Part X.5 — external resolution, never self-certified
- 05The measure itself is retired. Junctiq does not author probabilities. Brier scores, calibration curves and calibration gates are named in the governing blueprint as retired and not to be reintroduced, whatever their inputs.Part XIII
Junctiq has no externally resolved track record. Not a low one — none. No claim on this system has been adjudicated against a predeclared external source and rule, so there is nothing to score and no score is shown. The commitments the engine recorded remain stored as records of what was committed and when, and are no longer displayed on this site; their resolution states are self-reported and are not evidence that an outcome occurred.
An actor track record built from claims extracted from captured artifacts, each with its resolution contract declared before the outcome was known, and each resolved externally. That machinery is not built, and it is gated behind a kill test that has not yet returned. When it exists, what it produces will be counts of resolved claims with their artifacts attached — not a probability, and not a Brier score. How it works →
Subscriptions
The earlier Explorer, Strategy and Platform subscription tiers described the retired product and are no longer offered on this site. If you hold an existing subscription or have a billing question, write to us.
What Junctiq is now
Junctiq captures the public pages where companies state their commitments, keeps the exact bytes it received, hashes them and records when they were fetched. What it can and cannot yet show is set out on How it works, and the captured pages themselves are listed in the Vault.