# 008 antibind-mechanism — RESULT (harvested 2026-08-01, tier T3-exploratory)

**Verdict: NO meaningful early anti-binding dip at 10%-budget — the marginal below-chance
readings are noise-scale; the known anti-binding must form LATER, alongside the surface
code. The prereg's parity-arithmetic insight stands as the leading explanation, untested
at full budget.**

## Numbers (3 seeds × 17 dense checkpoints over 5e7 tokens)
- Deepest dips: 0.494 [0.487,0.500] (s0 @600), 0.493 [0.486,0.499] (s1 @800), 0.495 (s2
  @400/800) — magnitude ≤0.007 below chance, 1-2 of 17 checkpoints per seed at n_boot 200;
  consistent with noise + checkpoint selection, not a robust phenomenon.
- Surface cross barely develops at this budget: 0.51-0.54 final (vs ~0.70 at full 5e8);
  within-ceiling 0.50-0.51 — the probe-power gate flagged in the prereg is effectively
  binding: nothing much has formed by 5e7 tokens.
- No coincident surface-rise at the dips (surface ≈0.513-0.517 there, flat from init).
  Mechanism decomposition (conditional on a replicated dip) correctly did not trigger.

## Key conceptual yield (from the prereg, pre-data)
The stimulus grid has 5 families → **12 flipped : 8 same-parity ordered pairs**, so a pure
surface-position reader mechanically scores BELOW 0.5 on the pooled primary. "Anti-binding
from init" is therefore predicted to be the arithmetic shadow of the surface code — a
hypothesis this budget couldn't yet test (surface code unformed). Full-budget dense
checkpointing over 5e8 tokens is the follow-up.

## Brier: (a) dip in ≥2/3 seeds (0.70) → T-by-letter (all seeds have ≥1 below-chance
ckpt) but trivial in magnitude — scored T with the caveat; (b) dip↔surface-rise (0.65)
→ F; (c) anti-alignment (0.40) → unscored (conditional gate correctly not triggered);
(d) recovery by 5e7 (0.50) → T. Mean (scored 3) **0.192**.

## Caveats: 10% budget by design (early-dynamics scope); n_boot 200; the historical
anti-binding observation came from full-budget organisms — nothing here contradicts it.