Fix dropped --gate-strength token in tradeoff section

Add causality/readability gate-magnitude tradeoff table; recommend gate_strength 0.10

GRAFT coupling checkpoint — 2.043 bits deployable via short causal-EMA; honest findings + caveats

generation from the PASSING checkpoint: gated output is token soup, base is fluent

CORRECTION: the swap PASS destroys language — fluency<->coupling tradeoff is the real finding

PASSING weights: strong dir-gate (gate_strength 0.3), causal consciousness->language coupling verified

BREAKTHROUGH: first swap-test PASS — acc 0.833, MI 2.074/2.58 bits, perm p=.001, beats noise

dir-gate best weights (MI 0.45, first noise-controlled causal coupling)

dir-gate: PASSES noise control (real 0.33 >> noise -0.47) + permutation; fails only absolute MI<0.5 = genuinely causal but weak

C-swap causality verdicts: phase passes permutation (p=.001), amplitude fails all; both fail noise control

A: phase-bridge peak weights (MI 0.66) captured by mi-ratchet

GRAFT A/B/C/D bridge experiments: results + settings + v3b ckpt

initial commit