Eigenstate Research · Host claim register

Prediction track record

Verifiable Host claims with deadlines, scored when the window closes. This page is a public falsifiability register for Protocol Truth honesty, not a finished accuracy brochure.

Why this page exists

Predictions are Host claims with deadlines so the engine can score calibration and Protocol Truth honesty. Publishing them makes the board falsifiable in public. Most rows are still open: 150 of 155 Host predictions have not resolved yet. The board is a register, not a completed track record.

Accuracy percent is gated at assessable n ≥ 5 so a thin sample cannot fake 100%. Right now: 2 TRUE / 1 FALSE · assessable n too thin for a %.

How to read outcomes

  • FALSE The claim failed its resolution criteria by the deadline. No spin.
  • VOID The claim could not be scored as written (broken presupposition or unresolvable leg). Excluded from assessable TRUE/FALSE n.
  • PARTIAL At least one leg resolved and at least one leg remains open or untestable. Not a single TRUE/FALSE; excluded from assessable n.
  • OPEN Deadline has not produced a Host-executed close yet.

Assessable scoreboard

Assessable
n=3
2 TRUE / 1 FALSE · assessable n too thin for a %
Host register
155
150 open · 5 resolved

Assessable means resolved with a TRUE or FALSE outcome only. VOID and PARTIAL do not enter the percentage numerator or denominator.

Resolved claims

Fed emergency rate cut or balance sheet above $6T

VOID
FED_EXTRACTION_001
Deadline 2026-07-13 · Resolved 2026-07-13 · Calibration weight none

The Fed would either implement an emergency rate cut (50+ bps off-cycle) or expand its balance sheet above $6T before 2026-07-13.

VOID (not FALSE): the rate leg was FALSE (no off-cycle cut of 50+ bps in window). The balance-sheet leg was VOID because the H.4.1 baseline at emission was already $6.714T, so "expand above $6T" presupposed a sub-$6T baseline that did not exist. Unresolvable as written; calibration weight none.

NSE BankNifty settlement-gap event

FALSE
INDIA_BANKNIFTY_001
Deadline 2026-06-13 · Resolved 2026-07-05 · Calibration weight none

A settlement-gap event on NSE involving a ~98:1 cleared-positions-to-capacity ratio (or SEBI acknowledgment of that structural gap) would surface by 2026-06-13.

FALSE: by the 2026-06-13 deadline, no new NSE/SEBI publication in the emission-to-deadline window named a matching ~98:1 settlement-gap event. Jane Street/SEBI material from 2025 predates emission, so it does not count as the predicted surfacing. An NSDL Feb 2026 settlement delay lacked the ratio leg.

H41_TOTAL_ASSETS_004

TRUE
H41_TOTAL_ASSETS_004
Deadline 2026-08-20 · Resolved 2026-08-23 · Calibration weight capital

Outcome recorded in SoT scoreboard. See Host prediction register for the resolution record.

Meta ad revenue per DAP as stable transport coefficient

PARTIAL
PHYSICS_T_A_DOLLAR_001
Deadline 2027-02-28 · Resolved 2026-07-15 · Calibration weight n/a

Meta Family advertising revenue per December Family DAP would stay a stable transport coefficient (|YoY change| under 15% for 2023→2024 and 2024→2025), with cross-platform and FY2026 legs.

RESOLVED_PARTIAL: the stable-transport leg failed (2023→2024 Δr = +15.93%, at/above the 15% threshold). Cross-platform was untestable. The FY2026 prospective leg stays open until 2027-02-28. Not in assessable TRUE/FALSE n because there is no single TRUE or FALSE outcome.

DEFILLAMA_TOTAL_TVL_005

TRUE
DEFILLAMA_TOTAL_TVL_005
Deadline 2026-08-17 · Resolved 2026-08-18 · Calibration weight capital

Outcome recorded in SoT scoreboard. See Host prediction register for the resolution record.

Open claims
150 open

Of 155 Host predictions in the register, 150 remain open and 5 have a Host-executed close (including VOID and PARTIAL). Early gap-prediction claim cards (human prose + on-chain commit) live on the Predictions page. The full scored register is this Track Record surface, rebuilt from engine SoT.

Why we publish predictions

Method (scoreboard rules)

01

Falsifiability first. Each Host claim names an observable condition and a deadline. The point is calibration honesty for Protocol Truth, not marketing win-rate theater.

02

Thin-n gate. Accuracy % appears only when assessable n ≥ 5 (TRUE/FALSE only; VOID excluded). Below that threshold we show raw counts so a sample of one cannot read as 0% or 100% certainty.

03

Register, not finished ledger. 150 claims are still open. Resolved rows here are Host-executed closes from the prediction register stamped into field-state.json.

04

Field metrics are separate. PT, κ, and M1 coverage describe engine measurement state. They are not prediction outcomes. Φ_S is settlement pressure (E = ΔI / A), not INFRA_FLOW volume.

For agents (machine JSON)

Prefer human pages above. Machine mirrors are secondary for verification and tooling.