SSM-AI – Ready-Made Gate Presets (4.4)

Pick a starting profile, then stamp and ship.

4.4 Ready-made gate presets (pick one to start)

Preset A — Conservative (production hardening).
When to use. Early rollouts, safety-critical flows, or unstable vendors.
Effect. Stronger braking under contradictions/errors; smoother, less jitter.

# Manifest block (paste-ready)
"gate": {
  "weights": {"F":1.5, "D":1.0, "L":1.0, "E":1.5, "V":0.5, "Q":0.5},
  "rho": 0.30,
  "g_min": 0.10,
  "eps_g": 1e-12,
  "safety_notch": {"enabled": true, "s_thr": 0.80},
  "mode": "mul",                    # RSI_env := g_t * RSI
  "band_policy": {
    "promote_if": "RSI_env >= A+",
    "pause_if":   "RSI_env in A0",
    "block_if":   "RSI_env <= A-"
  }
}

Quick intuition. Contradiction and error lanes (F, E) weigh more; rho=0.30 damps spikes; g_min=0.10 prevents total stall; notch clamps when any lane is severe.

Preset B — Agile (experimentation).
When to use. Sandboxes, fast A/Bs, iteration loops.
Effect. Lighter braking, preserves high RSI regions with u-space scaling.

"gate": {
  "weights": {"F":1.0, "D":0.5, "L":0.5, "E":1.0, "V":0.5, "Q":0.0},
  "rho": 0.15,
  "g_min": 0.00,
  "eps_g": 1e-12,
  "safety_notch": {"enabled": false, "s_thr": 0.80},
  "mode": "u_scale",                # RSI_env := tanh(g_t * atanh(RSI))
  "band_policy": {
    "promote_if": "RSI_env >= A0",
    "pause_if":   "RSI_env in A0 with drift_up",
    "block_if":   "RSI_env <= A-"
  }
}

Quick intuition. Lower weights on drift/latency; rho=0.15 is snappier; mode:"u_scale" damps less near high confidence.

Preset C — Throughput-aware (high QPS platforms).
When to use. Token/latency budgets are tight; small slowdowns ripple.
Effect. Penalizes latency/queue pressure more; keeps motion with a floor.

"gate": {
  "weights": {"F":1.0, "D":1.0, "L":1.5, "E":1.0, "V":0.5, "Q":1.0},
  "rho": 0.20,
  "g_min": 0.20,
  "eps_g": 1e-12,
  "safety_notch": {"enabled": true, "s_thr": 0.85},
  "mode": "mul",
  "band_policy": {
    "promote_if": "RSI_env >= A+",
    "pause_if":   "RSI_env in A0",
    "block_if":   "RSI_env <= A-"
  }
}

Quick intuition. Heavier weight on L (latency) and Q (queue); g_min=0.20 avoids thrash under load.

Sanity mini (calculator-fast).

# Example lanes: F=0.20, D=0.10, L=0.30, E=0.15, V=0.20, Q=0.10
# RSI=0.70, rho=1 for instant view (no smoothing)

# Preset A (Conservative)
W=6.0; mix=(1.5*0.20 + 1.0*0.10 + 1.0*0.30 + 1.5*0.15 + 0.5*0.20 + 0.5*0.10)/6 = 0.172
g=clamp(1-0.172,0,1)=0.828 → RSI_env = 0.828*0.70 = 0.5796 → A0/A+ boundary

# Preset B (Agile, u_scale)
W=3.5; mix=(1*0.20 + 0.5*0.10 + 0.5*0.30 + 1*0.15 + 0.5*0.20)/3.5 = 0.171
g=0.829 → RSI_env = tanh(0.829*atanh(0.70)) ≈ 0.665 → solid A+

# Preset C (Throughput-aware)
W=6.0; mix=(1*0.20 + 1*0.10 + 1.5*0.30 + 1*0.15 + 0.5*0.20 + 1*0.10)/6 = 0.187
g=0.813; floor g_min=0.20 → RSI_env = 0.813*0.70 = 0.569 → A0

How to choose (rule-of-thumb).
• If incidents are costly → Preset A.
• If exploring fast and logging everything → Preset B.
• If infra is the bottleneck → Preset C.
Freeze one preset per environment, stamp the manifest, and keep phi((m,a)) = m inviolable.

---

Navigation
Previous: SSM-AI – Gate Knobs & Purity (4.3)
Next: SSM-AI – Two-Minute Gate Calibration (4.5)

---

Directory of Pages
SSM-AI — Table of Contents