← Complete research archive
Plans & synthesisTheory47 lines

Reasoning Attack Plan — 2026-07-17 (post-Codex alignment)

Result May claim Must not claim --- --- --- SCEB typed 25.4% Controller localization; host-exec control Internal model reasoning / Shohin arithmetic NL SCEB 15.7% Op-selection without schedule in prompt Full executor internalization Halt-first 23.8% Decode/stop policy cashes late…

docs/research/concepts/REASONING_ATTACK_PLAN.mdOpen original Markdown ↗

Reasoning Attack Plan — 2026-07-17 (post-Codex alignment)

Claim hygiene (locked)

ResultMay claimMust not claim
SCEB typed 25.4%Controller localization; host-exec controlInternal model reasoning / Shohin arithmetic
NL SCEB 15.7%Op-selection without schedule in promptFull executor internalization
Halt-first 23.8%Decode/stop policy cashes latent answersNew weights or deeper compute
SSC 115/256External schedule owns cursorAutonomous reasoner

Codex Sol is correct: host arithmetic is a systems result, not a reasoning breakthrough. Use SCEB as a strong control and localization clue only.

Live scoreboard (honest)

SystemScoreClass
SSC scheduled115/256External control
SCEB typed closed-loop65/256Control (host math)
Halt-first decode61/256Decode policy
NL op-selection closed-loop8/51Controller signal (no schedule in prompt)
Typed v1 joint LM42/256Internal joint emission (weak)
Direct / whole16 / 9 per 256Baselines

Active lanes (do not collide)

OwnerExperimentStatus
CodexCausal carry motor (691928)RUNNING ~5h
This laneResult-digit motor (692100)~19.2M motor (total ≈144M <150M); r2 wide MLP
This laneNL op headsFrozen as controller control only

Theorem reminder (Codex)

Every deterministic single-pass one-bit consumer collapses to a two-motor bundle. Do not invent cosmetic one-bit “primitives.” Prefer grammar-gated output motors on frozen residuals (carry / digit sites).

Goal for this lane

Move execution serialization into Shohin: post-DRS residual already moves digit log-odds (~+31). Build a tiny grammar-gated result-digit motor (parallel to Codex’s carry motor), frozen backbone, no host apply_op. Success = autonomous multi-step exactness without external arithmetic.