Event Horizon IQ

The method

EHIQ measures what institutions actually say — and shows its work. Every instrument is named and versioned. Every number a venue produces can be recomputed from the frozen queries on /audit. And every line we have ever closed is on this page at full size, because a method that can’t name what it killed never killed anything.

The detector

The core instrument reads language against a subject’s own baseline — never cross-subject, never pooled genres (the genre-pooling lesson is written into the doctrine). Eight dimensions plus the quiet triad, published in the SSRN-lineage detector. Versioned: LAS v4, lexicon v2 (cutover 2026-09-10 — v1 numbers are labeled, never pooled with v2).

Surviving reads, ranked by evidence: candor measurement, and the tb-group separation (p=0.0028). Two-sided screens are shift detectors, not rankers: a PASS is within-entity vs. its own baseline, in both directions, with genre varying.

The discipline

  • Blind proposals. A proposer never sees the market price. A number formed with the price visible is a parrot, not a view — the blind path is law (PR #199).
  • Planted failures. Every gate is tested by plants observed to fire — a control that has never been seen refusing is a control that might not exist.
  • Frozen denominators. A rate is computed by a frozen, published query, footnoted, or it doesn’t ship. A threshold changed after the data arrives is worthless.
  • Negative results at full size. A closure prints at the same size as a launch. See below — this is the part nobody else does, and it is the whole point.

The instrument graveyard

Every line we closed, with its verdict. Printed at full size, per the covenant.

CLOSED — 100% detection failure
Crypto whitepaper fraud scorer

All 2026-09-07 verdicts were downgrades; detection was 100% dead against a blinded set. The fraud-detector line is closed. Surviving: candor + tb-group separation (p=0.0028) — kept because they measured, not because we hoped.

CLOSED — signal-separation wall
The excision predictor (redaction engine)

The recovery engine validated (3,304-pair key). The predictor closed 2026-09-08: the signal does not separate from the noise at the wall we could measure. The line-merge redesign was withheld at full size on 2026-09-10 (18/300; 185 fragment-absent — the same wall).

DEPRECATED — both pilots FAIL (structural)
TIQ-24 instrument, v7.0

Both pilots failed structurally; deprecated 2026-09-07. Survivors are internal only. The SSRN-lineage detector is unaffected — every venue runs it only.

RETIRED — no live replacement
v22 (ETH model line)

Retired 2026-09-07. There is no live ETH model — the allocation anchor is a hardcoded 25%, and calls stamped by v22 stay on the board labeled as legacy stamps with no basis on file. A dead instrument’s output is named as such, never laundered into the new one.

The gates these instruments sit behind are tested by planted failures observed to fire; the plants are in the repo. Recompute any figure on this site at /audit. The next closure prints here at the same size as this page.

The method — every instrument named, every closed line at full size