Start with a system that keeps moving. Its phase describes where it is in a cycle, while an internal state m carries a trace of past activity. The research begins by letting that trace affect the next movement.
The mechanism behind the question
The public T7 work unit proposes a phase increment ω + κm. Input changes m; firing also changes m. This closes a history–motion–response loop. Later studies add other forms of memory, so m is not a name for every learned quantity.
Make an observation
Compare initial m = −0.7 and m = 0.2 with the same phase and pulses. Then set κ to zero.
Change memory. Watch when the unit fires.
Green: phase. Dots: firing. Dashed lines: input pulses at 200 and 1,400. No firing within this window is not proof of permanent silence.
Update rule, scope and numerical values
The discrete T7 work-unit update order is ported from the public OTB15 model. ω=2π/258, ε=0.000065, β=0.6, refractory=512. Background drive=ε×(−1.2). The input schedule, initial-state controls and direct phase kick are teaching interventions. No companion, network, eligibility or learning is simulated.
φ* = φ + ω + κm fire = (φ* ≥ 2π) and (cooldown = 0) m′ = (1−ε)m + drive − β·fire φ′ = 0 if fire, else φ* mod 2π
Firing steps: 320, 1270, 1937
| Step | m | φ | Refractory |
|---|---|---|---|
| 0 | -0.4000 | 0.0000 | 0 |
| 320 | -0.4211 | 0.0000 | 512 |
| 400 | -0.4251 | 1.1021 | 432 |
| 800 | -0.4450 | 0.2093 | 32 |
| 1200 | -0.4644 | 5.4035 | 0 |
| 1600 | -0.4783 | 2.1821 | 182 |
| 2000 | -1.0943 | 6.0942 | 449 |
| 2400 | -1.0970 | 4.8788 | 49 |
What to notice
The timing can change even when the scheduled inputs are identical. With κ = 0, changing m no longer directly changes phase speed.
Check your understanding
Follow this idea into the research
Textbook revision 1 · Evidence snapshot: 05 Nov 2026 · Changes & corrections ↗