SSM-AI – Appendix A — Bands, Stamps & Checklist (A4–A7)

Ship with guardrails: clear bands, stamped context, quick turn-on.

A4) Band policy (recommended defaults)
• promote_if: RSI_env >= 0.60
• pause_if: RSI_env in (-0.60, +0.60)
• block_if: RSI_env <= -0.60
Apply hysteresis if desired: h_up = 0.02, h_dn = 0.02 (promotion/demotion gates).

# Band thresholds (with optional hysteresis)
def to_band(RSI_env, h_up=0.02, h_dn=0.02):
    if RSI_env >= 0.90 + h_up: return "A++"
    if RSI_env >= 0.60 + h_up: return "A+"
    if RSI_env <= -0.90 - h_dn: return "A--"
    if RSI_env <= -0.60 - h_dn: return "A-"
    return "A0"  # (-0.60, +0.60)

A5) Stamp fields (append to your one-liner)
Append these key-values to your stamp tail:

"|g=" + fmt(g_t)
+"|RSI_env=" + fmt(RSI_env)
+"|gate_mode=" + mode
+"|lanes=" + fmt(F_t,D_t,L_t,E_t,V_t,Q_t)"

A6) Acceptance checklist (pass/fail)
• Purity. phi((m,a)) = m before/after gating.
• Boundedness. |RSI_env| < 1 for all vectors.
• Monotonicity. If any lane increases (worsens), g_t must not increase.
• Determinism. Same manifest + inputs ⇒ same g_t, RSI_env (within dtype eps).
• Bands. to_band(RSI_env) matches thresholds (honor hysteresis if declared).

# Quick acceptance asserts
assert phi((m,a)) == m              # purity
assert abs(RSI_env) < 1             # boundedness
assert not increases_when_lane_worsens()  # monotonicity
assert deterministic_under_same_manifest() # determinism
assert to_band(RSI_env) == expected_band  # bands

A7) One-minute turn-on

  1. Paste one preset into your manifest.
  2. Run V1–V3 locally; match outputs within dtype tolerance (6-dec print OK).
  3. Enable gating on one surface (tools or decoding) and log RSI, g_t, RSI_env, band beside m.
  4. Roll up bands daily using the standard U/W fuse on the lane; phi((m,a)) = m holds throughout.
# Daily roll-up (lane-only; classical m unchanged)
U_day := SUM w*atanh(a_day)
W_day := SUM w
RSI_day := tanh( U_day / max(W_day, eps_w) )
band_day := to_band(RSI_day)  # or to_band(RSI_env_day) if gating applied

One-line takeaway. With crisp bands, stamped context, and a tiny acceptance list, you can turn on the calm gate in minutes — lanes stay bounded and auditable, while classical values remain untouched via phi((m,a)) = m.

Navigation
Previous: SSM-AI – Appendix A — Gate Presets & Acceptance (A1–A3)
Next: SSM-AI – Appendix B — Symbolic Search Lens (SSM-Search)

Directory of Pages
SSM-AI — Table of Contents