Jam Musical Policy

Rule: preserve this boundary. Keep creative choices model-owned and runtime guarantees code-owned.

Tempo Resolution Order (Required)

Always resolve tempo in this exact precedence:

1. Explicit BPM
2. Half/double time
3. Model-relative tempo intent

Written explicitly: explicit BPM > half/double time > model-relative tempo intent.

Additional rules:

• If explicit BPM and half/double-time language both appear, explicit BPM wins.
• Relative tempo language (faster, slower, push, lay back) applies only when no explicit BPM or half/double-time match exists.
• If confidence is low, hold current tempo/context (no_change is valid).

Phrase-to-Intent Guidance

Interpret phrase strength before deciding magnitude. Continuity and pocket come first.

Tempo

Energy

Arrangement

Anti-Overshoot Guidance

1. Match change magnitude to wording strength; subtle language should produce subtle deltas.
2. Prefer one primary move per turn; secondary changes should be supportive.
3. Ramp large relative changes over multiple rounds unless the directive is explicitly immediate.
4. Avoid simultaneous extreme tempo and energy jumps unless explicitly requested.
5. When ambiguous or low confidence, choose continuity-first behavior (no_change or minimal delta), not a hard reset.

Creativity-First Behavior

1. Start from intent, then invent role-specific realization.
2. Keep musical responses fresh across rounds; do not reuse one fixed mechanic.
3. Use Strudel capabilities broadly as a substrate, not as a narrow recipe list.
4. Maintain role boundaries (drums, bass, melody, FX) while maximizing expressive freedom inside each role.

Warning: Avoid Style/Template Lock-In

• Examples are anchors, not templates.
• Do not map phrases to one invariant pattern formula.
• Do not collapse recurring directives into a single house style.
• Vary strategy across groove, timbre, density, and phrasing while staying coherent.

Progressive Disclosure and Token Budget

Use a three-pass policy and stop at the smallest sufficient depth:

1. Fast pass (default): detect anchors, apply precedence, choose primary move.
2. Standard pass: add only role-specific details for dimensions that changed.
3. Deep pass: brief conflict/ambiguity rationale only when needed.

Token discipline:

• Keep outputs concise by default.
• Expand reasoning only for conflicts, uncertainty, or explicit request.
• Spend tokens on actionable intent translation, not long prose.

Quick Self-Check

• Precedence honored: explicit BPM > half/double time > model-relative tempo intent
• Model-owned creativity preserved; code-owned determinism not redefined
• Tempo, energy, and arrangement intent all resolved with phrase strength sensitivity
• Anti-overshoot and continuity safeguards applied
• No style/template lock-in introduced