L7 — embodiment and real-time control: control is continuous settling, end to end analog
Control is continuous settling in a clock-free, end-to-end analog sensorimotor loop: sense the delayed target, clean up, predict ahead to compensate loop delay, act. The forward model holds a moving target where a reactive controller lags and open-loop drifts; analog I/O beats a 1-bit ADC/DAC loop under noise (0.001 vs 0.159) and dimensional advantage scales (N×gap 4.6 → 42.1).
This layer carries the analog I/O thesis as its organizing test. Prediction’s embodiment role is compensating loop latency (E1); analog dominates the dimensional axis and the tax closes at 2 bits (E2); prediction-sufficiency suffices for control, resolving the inherited L6 [O] (E3); and a non-empty stable bandwidth/latency band exists (E4). The honest two-sides caveat (E5, [O]×2): single-channel bit depth is Shannon-capped and exact arithmetic needs a brief digital hand-off — the analog win is dimensional, not per-channel precision.
L7 closes the loop with a world. Cognition runs off the body, and the same substrate that stores and predicts must now control in real time — “input becomes a pattern at once” plus “real-time appropriate response.” The loop is built end to end on the frozen L0 + L6, with a non-circular read-out (tracking error against the world’s true target) (session v0.8, digest 17aa27bf…).
The closed loop, in four steps
Each tick: sense a graded-analog observation of the delayed target → clean up by L0 settling (the D4 noise immunity) → predict +1 with the L6 forward model to compensate the one-tick sensorimotor delay → act with a continuous, rate-limited phase step. Prediction’s role here is concrete: compensating loop latency.
Four milestones, all verified
| milestone | result | grade |
|---|---|---|
| E1 control by settling | forward holds target, err 0.000→0.004; reactive lags ≈0.96, open-loop drifts ≈0.77 | [V] |
| E2 analog beats ADC/DAC + scales | analog 0.001 vs 1-bit 0.159; tax closes at 2 bits; N×gap 4.6→42.1 | [V] |
| E3 prediction-sufficiency suffices | predicted-state control tracks identically (cosine 0.803); resolves the L6 [O] | [V] |
| E4 stable band exists | non-empty bandwidth/latency band; self-corrects from displacement | [V] |
The analog I/O thesis, tested not asserted
E2 is the directive’s headline: on the dimensional axis analog dominates — each channel carries a continuous value, all channels are integrated by one relaxation in parallel, and the aggregate advantage grows with state dimension. The O(1)-vs-O(N) latency form (the optical-Fourier point) is recorded as the inherited R4 principle [O], since physical parallelism stays deferred.