TUNING_MAP.md

Tuning Map - Feedback Loop Constants

Cartography of the ~30 most critical numeric constants governing the four feedback loops and the trust system. Every constant below has at least one interaction partner - changing one in isolation will shift the system's emergent behavior.

1. Density Correction - coherenceMonitor

Compares actual vs intended note output. Feeds a correction bias back into the density product so the system listens to its own song.

Constant	Value	Role	Interaction Partners
SMOOTHING	0.55	EMA factor for coherenceBias - higher = slower adaptation	BIAS_FLOOR/CEILING, phaseGain
BIAS_FLOOR	0.60	Min density correction multiplier	SMOOTHING, conductorConfig density range
BIAS_CEILING	1.38	Max density correction multiplier	SMOOTHING, profileAdaptation restrained hint
WINDOW_SIZE	16	Rolling observation window (beats)	decayFactor, entropy variance calc
ENTROPY_DECAY	0.92	Exponential decay for entropy signal	rawEntropy offset/scale, entropyRegulator strength
phaseGain bell	0.25 + 0.3.sin(pi.phraseProgress)	Correction strength peaks mid-phrase (0.55), drops at edges (0.25)	SMOOTHING, attribution spread boost
decayFactor	0.5	Phrase-boundary decay for window entries - prevents stale data from dominating	WINDOW_SIZE
rawEntropy offset	0.04	Variance baseline subtracted before scaling	ENTROPY_DECAY, scale factor 2
Attribution spread boost	1.0 + clamp(spread - 0.25, 0, 0.5)	When density biases disagree (spread > 0.25), correction gets up to 50% stronger	phaseGain, SMOOTHING

Sensitivity: SMOOTHING is the single most impactful constant - current value 0.55 provides responsive correction with moderate oscillation risk. Raising above 0.85 makes correction sluggish; lowering below 0.4 causes visible density oscillation. BIAS_CEILING must stay below the negotiation playScale upper clamp (1.8) to avoid runaway density gain.

2. Entropy Steering - entropyRegulator

Measures combined pitch/velocity/rhythmic entropy. Steers cross-layer systems toward a target curve driven by section position.

Constant	Value	Role	Interaction Partners
SMOOTHING	0.3	EMA factor for smoothed entropy	Pitch/velocity/rhythm weights
WINDOW_NOTES	10	Max note history per layer (halved for faster turnover)	Pitch entropy uniqueness calc
Pitch weight	0.4	Contribution of pitch diversity to combined entropy	Velocity weight (0.3), rhythm weight (0.3)
Velocity weight	0.3	Contribution of velocity variance to combined entropy	Pitch weight, rhythm weight
Rhythm weight	0.3	Contribution of rhythmic irregularity	Pitch weight, velocity weight, rhythm divisor (2)
Arc floor	0.2	Minimum target entropy from section-shape arc	Arc amplitude (0.7) - range is [0.2, 0.9]
Arc amplitude	0.7	Bell curve amplitude: 0.2 + 0.7.sin(pi.progress)	Arc floor, section length
Arc-intent blend	0.45 / 0.55	45% section arc, 55% sectionIntentCurves target	sectionIntentCurves.entropyTarget
PID gain	2.0	Proportional response: scale = 1 + error.strength.2	regulationStrength, scale clamp
Scale clamp	[0.3, 2.0]	Min/max regulation scale - prevents extinction or explosion	PID gain, negotiation entropy modulation

Sensitivity: The 0.3/0.7 arc-intent blend is a critical mixing ratio. Raising arc weight above 0.5 makes section shape dominate, reducing intent responsiveness. The PID gain of 2 is aggressive - lowering to 1.5 yields smoother but slower correction.

3. Sustained-Condition Hints - profileAdaptation

Watches for sustained low-density / high-tension / flat-flicker streaks. Produces advisory hints consumed by conductorConfig.

Constant	Value	Role	Interaction Partners
DENSITY_LOW_THRESHOLD	0.55	Density below this increments low-density streak	STREAK_TRIGGER, coherenceMonitor bias
TENSION_HIGH_THRESHOLD	1.4	Tension above this increments high-tension streak	STREAK_TRIGGER, negotiation conflict threshold
FLICKER_FLAT_THRESHOLD	1.05	Flicker within 0.05 of 1.0 counts as flat	STREAK_TRIGGER
STREAK_TRIGGER	6	Beats before a hint activates	Hint ramp divisor, conductorConfig consumption
Trend mods (density)	0.75 rising / 1.25 falling	Streak increment varies with signal telemetry trend	signalTelemetry.getTrend()
Trend mods (tension)	1.25 rising / 0.75 falling	Tension trend amplifies/dampens streak	signalTelemetry.getTrend()
Hint ramp divisor	8	Beats past trigger before hint reaches 1.0: (streak - trigger) / 8	STREAK_TRIGGER, conductorConfig phase profiles

Sensitivity: STREAK_TRIGGER = 6 at default tempo (72 BPM) means ~5 seconds of sustained condition before hints activate. The ramp divisor of 8 means full hint intensity at streak = 14 (~12 sec). Lowering STREAK_TRIGGER below 4 risks false positives from momentary lulls.

4. Negotiation - negotiationEngine

Integrates trust scores, entropy regulation, and intent targets to produce final playProb / stutterProb values.

Constant	Value	Role	Interaction Partners
Play scale formula	(0.75 + density.0.45) * (0.9 + trust.0.08)	Computes play probability scale from intent + trust	playScale clamp, adaptiveTrustScores
playScale clamp	[0.4, 1.8]	Prevents play probability extinction or saturation	coherenceMonitor BIAS_CEILING (1.38)
stutter scale formula	(0.6 + interaction.0.75) * (0.85 + trust.0.1)	Computes stutter scale from interaction target + trust	stutterScale clamp
stutterScale clamp	[0.25, 2.2]	Wider range than play - stutter is more exploratory	Play scale clamp
Entropy play modulator	0.7 + entropy.0.3, clamp [0.5, 1.5]	Entropy regulation adjusts play probability	entropyRegulator scale output
Entropy stutter modulator	0.75 + entropy.0.25, clamp [0.5, 1.5]	Entropy regulation adjusts stutter probability	entropyRegulator scale output
Conflict threshold	0.8	Trust conflict above this triggers dampening	adaptiveTrustScores conflict detection
Play conflict dampen	0.92	8% play reduction during high conflict	Conflict threshold
stutter conflict dampen	0.9	10% stutter reduction during high conflict	Conflict threshold
Cadence gate: phase	>= 0.45	Min phase confidence for cadence allowance	phaseAwareCadenceWindow confidence
Cadence gate: trust	>= 0.7	Min cadence trust weight for cadence allowance	adaptiveTrustScores weight

Sensitivity: The play scale clamp [0.4, 1.8] is the single most important range in the system. If BIAS_CEILING (1.38) * playScale max (1.8) were to compound, density could exceed 2.48*. The cadence gate thresholds (0.45/0.7) determine how often cadences fire - lowering them increases cadence frequency dramatically.

5. Trust Governance - adaptiveTrustScores

EMA-based trust weights per cross-layer module. Payoff table defined in MAIN_LOOP_CONTROLS.trustPayoffs. All trust system names are canonical constants defined in trustSystems (src/utils/trustSystems.js) -- 9 scored systems in trustSystems.names, 13 heat-map systems in trustSystems.heatMapSystems. Never hardcode trust name strings.

Constant	Value	Role	Interaction Partners
EMA decay	0.9	Weight on previous score: score = score.0.9 + payoff.0.1	Weight formula, payoff clamp
EMA new-data	0.1	Weight on new payoff observation	EMA decay
Score clamp	[-1, 1]	Trust score range	Weight formula
Weight formula	1 + score * 0.75	Converts score to multiplicative weight	Weight clamp
Weight clamp	[0.4, 1.8]	Same range as negotiation play scale - by design	negotiationEngine play scale clamp
Default decay rate	0.01	Per-call decay toward neutral when no observations	Score clamp, EMA factors

Sensitivity: The weight multiplier 0.75 means a score of 1.0 yields weight 1.75, and score -0.8 yields weight 0.4 (floor). The EMA rate 0.9/0.1 means ~10 observations to converge halfway. Lowering to 0.8/0.2 doubles learning speed but risks oscillation.

6. Breathing & Probability Adjust - processBeat

Final probability adjustments applied in the probability-adjust pipeline stage.

Constant	Value	Role	Interaction Partners
Complement fill urgency	0.3	playProb *= (1 + fillUrgency * 0.3) - max 30% boost	restSynchronizer complementary rest
Breathing decrease: play	0.96	4% play reduction on heat-map breathing decrease	interactionHeatMap heat level
Breathing decrease: stutter	0.94	6% stutter reduction on decrease	interactionHeatMap heat level
Breathing increase: play	1.03	3% play boost on breathing increase	interactionHeatMap heat level
Breathing increase: stutter	1.04	4% stutter boost on increase	interactionHeatMap heat level

7. Pipeline Coupling Management - pipelineCouplingManager

Self-tuning decorrelation engine for all 15 compositional dimension pairs. Nudges density/tension/flicker biases to reduce inter-signal correlation.

Constant	Value	Role	Interaction Partners
DEFAULT_TARGET	0.25	Default coupling target for any pair	Per-pair overrides in PAIR_TARGETS
GAIN_INIT	0.16	Starting gain for new pairs	GAIN_MIN, GAIN_MAX, effectiveness gate
GAIN_MIN	0.08	Minimum gain floor	GAIN_INIT
GAIN_MAX	0.60	Maximum gain ceiling	GAIN_INIT, density-flicker scale, product guards
GAIN_ESCALATE_RATE	0.02	Per-beat gain increase when stuck	GAIN_EMERGENCY_RATE, effectiveness gate, floor dampening
GAIN_EMERGENCY_RATE	0.06	3x escalation when |r| > 2x target	GAIN_ESCALATE_RATE
GAIN_RELAX_RATE	0.02	Per-beat gain decrease when resolved	Entropy pair 2x multiplier
_TARGET_ADAPT_EMA	0.02	~50-beat horizon for rolling |r|	Entropy pair 2.5x multiplier
_TARGET_RELAX_RATE	0.0015	Per-beat target relaxation (intractable)	_TARGET_TIGHTEN_RATE
_TARGET_TIGHTEN_RATE	0.0015	Per-beat target tightening (resolved)	Density product sigmoid scalar
_AXIS_BUDGET	0.24	Per-axis gain budget cap (#9 hypermeta)	Product guards, axis coupling ceiling
Effectiveness gate threshold	0.50	Pairs below this get rate *= max(0.25, eff/0.50)	GAIN_INIT, gain ceiling cap
Effectiveness gain ceiling	eff < 0.40	Cap = GAIN_INIT + (MAX-INIT) * max(0.40, eff)	GAIN_MAX, GAIN_INIT
Flicker product guard	hysteresis 0.90/0.96	'guarding' state caps flicker-pair gains at 0.45	_FLICKER_PAIR_GAIN_CAP, flicker product
Density product guard	hysteresis 0.75/0.82	'guarding' state caps density-pair gains at 0.45	_DENSITY_PAIR_GAIN_CAP, density product
Severe bypass	|r| > 2x target	Bypasses coherence gate for severely overcoupled pairs	Coherence gate values
Transition spike dampener	R33: velocity-based	Boosts effective gain when coupling velocity spikes	_VELOCITY_EMA_ALPHA, _VELOCITY_TRIGGER_RATIO
_VELOCITY_EMA_ALPHA	0.08	~12-beat horizon for coupling velocity EMA	_VELOCITY_TRIGGER_RATIO
_VELOCITY_TRIGGER_RATIO	2.0	Boost when instantaneous velocity > 2x EMA	_VELOCITY_EMA_ALPHA
_VELOCITY_BOOST_BEATS	3	Cooldown beats after velocity spike trigger	_VELOCITY_GAIN_BOOST
_VELOCITY_GAIN_BOOST	2.0	Effective gain multiplier during velocity boost	_VELOCITY_BOOST_BEATS
Per-pair effectiveness temporal	min/max/activeBeats	R33 E5: Tracks effectiveness range across active beats	effectivenessEma
Heat-penalty cooldown threshold	0.30	R34 E5: Above this hp, rate scaled by max(0.35, 1-hp)	heatPenalty, escalation rate
Heat-penalty cooldown floor	0.35	R34 E5: Minimum rate multiplier at hp=1.0	Heat-penalty cooldown threshold
_AXIS_SMOOTH_ALPHA	0.15	R34 E2: ~7-beat EMA horizon for axis coupling totals	getAxisCouplingTotals(), axis energy share
Universal exceedance gain cap	0.30 when |r| > 0.85 + active decay	R37 E4: gain *= 0.95/beat when |r| > 0.85 and gain > 0.30. R36 density-flicker had 76 exceedance beats despite 0.30 cap because cap only limited growth, not existing gain	Cap + decay closes the oscillation loop

Sensitivity: GAIN_ESCALATE_RATE (0.02) and _globalGainMultiplier (controlled by homeostasis) together determine effective escalation speed. The effectiveness gate (R32 E1) reduces rate for low-eff pairs to max 25% of normal. Floor dampening (from homeostasis) is multiplicative and can suppress 75%+ of all escalation. Product guards cap gains at 0.45 when pipeline products are severely compressed. R33 velocity-based spike dampener detects the spike beat itself (not the beat after), providing preemptive 2x gain boost for 4 beats total (trigger + 3 cooldown). R34 E2: Axis coupling totals are now returned as a running EMA (alpha=0.15) rather than per-beat snapshots, preventing phase axis collapse when phase pair correlations are intermittently null. Internal gain scaling still uses raw per-beat values. R34 E5: Heat-penalty escalation cooldown adds a second throttle layer: when hp > 0.30, the gain escalation rate is scaled by max(0.35, 1.0 - hp), creating graduated suppression that prevents density-flicker oscillation cycles from persisting.

8. Axis Energy Equilibrator - axisEnergyEquilibrator (#13)

Two-layer self-correcting controller. Layer 1 detects pair-level hotspots. Layer 2 balances axis-level energy shares. Interacts with #1 (targets), #9 (gain budget), #12 (homeostasis).

Constant	Value	Role	Interaction Partners
_HOTSPOT_RATIO	2.0	Pair hot when raw > 2.0x baseline	_HOTSPOT_ABS_MIN, Layer 1 tighten
_HOTSPOT_ABS_MIN	0.25	Ignore hotspot unless raw crosses this floor	_HOTSPOT_RATIO
_COLDSPOT_RATIO	0.3	Pair cold when raw < 0.3x baseline	_COLDSPOT_ABS_MAX
_COLDSPOT_ABS_MAX	0.10	Only relax if raw below this cap	_COLDSPOT_RATIO
_PAIR_TIGHTEN_RATE	0.004	Per-beat baseline tightening for hot pairs	tightenScale (graduated gate), giniMult
_PAIR_RELAX_RATE	0.002	Per-beat baseline relaxation for cold pairs	N/A
_PAIR_COOLDOWN	3	Beats before a pair can be readjusted (Layer 1)	_AXIS_COOLDOWN
_AXIS_OVERSHOOT	0.22	Axis share above this triggers tightening	_FAIR_SHARE (0.167)
_AXIS_UNDERSHOOT	0.12	Axis share below this triggers relaxation	_FAIR_SHARE
_AXIS_TIGHTEN_RATE	0.002	Per-beat tightening rate for overshoot axes	tightenScale, giniMult, pairScale
_AXIS_RELAX_RATE	0.0009	Per-beat relaxation rate for undershoot axes	pairScale (E2 nudgeable scaling)
_AXIS_COOLDOWN	4	Beats before axis pairs can be readjusted	_PAIR_COOLDOWN
_SHARE_EMA_ALPHA	0.08	~12-beat horizon for smoothed axis shares	Raw shares from coupling manager
_GINI_ESCALATION	0.40	Gini above this -> 1.5x rate multiplier	axisGini from coupling manager
_EFFECTIVE_NUDGEABLE	d/t/f=5, e/tr/ph=3	Nudgeable pair count per axis	_RELAX_RATE_REF
_RELAX_RATE_REF	5	Reference count for rate scaling	Relaxation scaling = 5/count
Graduated coherent gate	evolving=0.6, coherent=0.0, exploring=1.5	tightenScale multiplier for Layer 1+2	regimeClassifier.getRegime()
CROSS_INHIBIT_WINDOW	6	Beats after one handler adjusts a pair before the OTHER handler can reverse direction. Prevents pair-vs-axis oscillation when PAIR_COOLDOWN=3 expired before AXIS_COOLDOWN=4. Phase floor/extreme-collapse bypasses.	_PAIR_COOLDOWN, _AXIS_COOLDOWN

Sensitivity: The graduated coherent gate is the most critical interaction. CROSS_INHIBIT_WINDOW (6 beats) is larger than either cooldown period (3/4) to guarantee directional protection across the full recovery window. Phase floor/extreme-collapse bypasses the inhibit unconditionally since phase recovery takes absolute priority over oscillation prevention. R35 E4: Evolving tightenScale raised from 0.4 to 0.6 to compensate for exploring absence (R34: 0% exploring, tighten budget collapsed 35->14). R34 E4: Exploring 1.5x amplification. It prevents tightening from destabilizing coherent regime (0.0 during coherent) while allowing partial correction during evolving (0.4). The _AXIS_OVERSHOOT/_UNDERSHOOT thresholds (0.22/0.12) with fair share at 0.167 create a +/-33% deadband. _RELAX_RATE_REF scaling gives trust/entropy/phase 1.67x faster undershoot relaxation to compensate for fewer nudgeable pairs. R33 E2: Symmetric tighten-rate scaling applies the same _EFFECTIVE_NUDGEABLE / _RELAX_RATE_REF ratio to the overshoot tightening path. Entropy/trust/phase axes now tighten 1.67x faster (matching relaxation), preventing asymmetric response where overshoot persists because tightening is slower than relaxation.

9. Coupling Homeostasis - couplingHomeostasis (#12)

Global coupling energy governance. Tracks total energy, detects redistribution, controls gain multiplier and floor dampening.

Constant	Value	Role	Interaction Partners
_GAIN_FLOOR	0.20	Minimum global gain multiplier	_globalGainMultiplier in coupling manager
_GINI_THRESHOLD	0.40	Gini above this triggers redistribution detection	Pair-level Gini coefficient
_REDIST_RELATIVE_THRESHOLD	0.008	Redistribution detection: |deltaEma| < 0.008 AND turbulence > 0.008	pairTurbulenceEma
Floor dampen window	0.35	totalEnergy / floor mapped to [min, 1.0] over [1.0, 1+window]	_totalEnergyFloor, floor dampen min
Floor dampen min	0.20	Minimum floor dampening value (base)	Floor dampen window, chronic decay
_CHRONIC_DAMPEN_THRESHOLD	20	Consecutive sub-0.50 beats before decay	_CHRONIC_FLOOR_RELAX_RATE
_CHRONIC_FLOOR_RELAX_RATE	0.005	Per-beat floor downward nudge during decay	_CHRONIC_DAMPEN_THRESHOLD
_CHRONIC_FLOOR_RELAX_CAP	0.60	Max effective floor dampen during chronic decay	Floor dampen min
_totalEnergyFloor	asymmetric tracking	Rises fast when energy drops (alpha=0.04), falls slowly when energy rises (alpha=0.005)	totalEnergyEma, floorDampen
Proportional control	Kp in [0.5, 2.0]	Multiplier adjustment: target / max(totalEnergy, 0.1)	energyBudget, totalEnergyEma

Sensitivity: Floor dampening is the dominant throttle mechanism. When totalEnergyEma is close to _totalEnergyFloor (ratio < 1.35), floorDampen drops below 0.50, suppressing >50% of all gain escalation. The asymmetric floor tracking (fast up 0.04, slow down 0.005) means the floor quickly captures energy minima but very slowly releases them -- this can create chronic dampening lock. R33 E3 chronic decay breaks this lock: after 20 consecutive sub-0.50 beats, the floor is nudged downward (0.5%/beat) and the effective minimum rises toward 0.60 (0.01/beat). The floor never drops below 60% of EMA, maintaining minimum 40% suppression. Any energy surge that raises rawDampen above 0.50 resets the counter.

10. Phase Energy Floor - phaseFloorController (#14)

Self-calibrating phase share recovery controller. Derives collapse thresholds, boost multipliers, and streak activation counts adaptively from rolling volatility and coherent regime duration. Now bidirectional: boosts when phase is low, attenuates when phase recently overshot.

Constant / State	Value	Role	Interaction Partners
_SHARE_EMA_ALPHA	0.06	~17-beat EMA for phase share	persistentLowSharePressure, getLowShareThreshold()
_RECOVERY_EMA_ALPHA	0.08	Recovery attribution window EMA	recoveryFactor in computeBoosts
RECOVERY_ATTRIBUTION_WINDOW	12	Beats after boost before unattributed failure	phaseFloorControllerBeatsSinceBoost
getCollapseThreshold()	[0.01, 0.04]	Share below = phase collapse (base 0.015 + volatility*0.5)	getExtremeCollapseStreak()
getLowShareThreshold()	[0.02, 0.16]	Share below = low share streak accumulates	getFloorActivationStreak(), fair share 0.167
getFloorActivationStreak()	[4, 16]	Beats below low-share threshold before floor activates	getEscalatedBoostStreak()
computeBoosts() base boost range	[6.0, 22.0]	phaseFloorBoost graduated by streakDepth, deficit, recovery	emergencyBoost in axis adjuster
phaseCollapseBoost	[3.5, 11.0]	Boost when share < collapseThreshold	isPhaseCollapse detection
phaseRetractionMult	[0.7, 1.0]	Anti-overshoot: dampens boost when phase recently > 115% of fair share. Bypassed during genuine collapse.	phaseFloorControllerHighShareStreak, axis adjuster floor path
phaseFloorControllerHighShareStreak	counter	Increments each beat share > fairShare1.15; resets below fairShare1.05	phaseRetractionMult = 1.0 - streak*0.025
System phase scaling	oscillating=0.6, stabilized=0.85, other=1.0	Dampens boosts during oscillation	hyperMetaManager.getSystemPhase()

Sensitivity: phaseRetractionMult prevents overshoot feedback loops where aggressive boosts push phase above fair share, triggering further boost applications on the next starvation episode. The 0.025 per-beat decay means 12 beats above threshold produces ~70% retraction. The retraction hard-bypasses on genuine collapse (share < collapseThreshold) so the controller can't be stuck in retract mode during actual emergencies.

11. Dimensionality Expander - dimensionalityExpander

Prevents phase-space collapse by injecting orthogonal perturbations when effectiveDimensionality falls below threshold. Now gated by per-axis cooldowns to prevent continuous nudging before the axis has responded.

Constant	Value	Role	Interaction Partners
DIM_THRESHOLD	2.2	Below this, begin injecting perturbations	effectiveDimensionality from systemDynamicsProfiler
DIM_CRITICAL	1.5	At this, perturbation is at full strength	Urgency ramp between 1.5-2.2
MAX_PERTURBATION	0.15	Maximum perturbation per axis	Coupling-based perturbation direction vectors
DEAD_AXIS_THRESHOLD	0.05	Variance ratio below this = axis considered dead	DEAD_AXIS_PERTURBATION, nudge gap
DEAD_AXIS_PERTURBATION	0.12	Nudge magnitude scaling (severity * 0.12)	DEAD_AXIS_MIN_GAP, SMOOTHING
DEAD_AXIS_MIN_GAP	8	Minimum beats between dead-axis nudge applications. Gap resets when axis exits dead zone. Prevents continuous nudging before axis has had time to respond.	Per-axis axisNudgeGaps[i] counter
DOMINANT_AXIS_THRESHOLD	0.35	Variance ratio above this = axis dominant	DOMINANT_AXIS_DAMPENING, DOMINANT_MIN_NUDGE
DOMINANT_AXIS_DAMPENING	0.12	Suppression scaling for dominant axis excess	DOMINANT_MIN_NUDGE floor
DOMINANT_MIN_NUDGE	0.02	Floor so threshold boundary produces meaningful force	DOMINANT_AXIS_DAMPENING
SMOOTHING	0.25	EMA factor on bias outputs	Prevents discontinuous jumps in bias values
COUPLING_THRESHOLD	0.30	Only break correlations stronger than this	Coupling matrix pairs: dt, tf, df

Sensitivity: DEAD_AXIS_MIN_GAP (8 beats) is deliberately larger than DIM_THRESHOLD response latency (~3-5 beats at SMOOTHING=0.25). Without the gap, continuous nudging drives the EMA toward a fixed non-neutral setpoint even when the axis is recovering. The gap = 0 reset when axis exits dead zone means recovery is recognized immediately and further nudging stops.

12. Self-Organized Criticality - criticalityEngine

Monitors distance from the "edge of chaos." Triggers avalanche corrections when accumulated energy exceeds threshold, maintaining ~20% avalanche rate. Now health-aware: worse system health lowers the effective trigger threshold, producing more corrective avalanches when the system is under stress.

Constant	Value	Role	Interaction Partners
WINDOW	16	Beats to accumulate energy before avalanche check	THRESHOLD_MIN/MAX, energyBuffer
TARGET_RATE	0.20	Desired fraction of beats with avalanche	ADAPT_RATE, threshold self-calibration
THRESHOLD_MIN	0.08	Minimum energy threshold (calibrated down from 0.15 to activate at normal energy levels)	ADAPT_RATE, avg window energy
THRESHOLD_MAX	1.20	Maximum energy threshold	ADAPT_RATE
ADAPT_RATE	0.02	Per-beat threshold adjustment toward TARGET_RATE	Rate-based adaptation: +0.02 if rate > 0.20, -0.02 if rate < 0.10
SNAP_STRENGTH	0.96	How hard avalanche snaps bias toward neutral (4% correction)	critSnapScale from hyperMetaManager, RECOVERY_BEATS
RECOVERY_BEATS	3	Beats of recovery after each avalanche	Bias ramp-down during recovery
criticalityHealthScale	clamp(healthEma/0.7, 0.5, 1.4)	Effective threshold multiplier. healthEma=0.35 -> scale=0.5 (more avalanches). healthEma=1.0 -> scale=1.4 (fewer avalanches).	hyperMetaManager.getSnapshot().healthEma
Section-bypass gate	back half (progress > 0.55)	tensionBias() returns 1.0 in Q3/Q4 -- prevents compounding tension suppression	timeStream.compoundProgress('section')
Health grade scale	healthy=1.0, strained=0.5, stressed=0	Per-signal bias gating by signalHealthAnalyzer.getHealth() grade	Multiplied into avalanche bias for each signal

Sensitivity: criticalityHealthScale decouples the per-signal bias gating (which uses categorical grade) from the trigger threshold (which uses continuous healthEma). At nominal health (0.7), the formula gives scale=1.0 -- no change. Under stress the threshold drops proportionally, making the engine more responsive precisely when corrective avalanches are most needed. SNAP_STRENGTH=0.96 was tuned in R35 (from 0.92) after tension arc regression showed gentler snap was needed while the engine was newly activated.

13. Rhythmic Feedback - feedbackOscillator

Creates rhythmic round-trip impulses between L1/L2 with complementary interval biasing. Highest hotspot rate (31%) of any trust system.

Constant	Value	Role	Interaction Partners
DAMPING	0.55	Energy decay per round-trip -- prevents runaway feedback	CIM setCoordinationScale, trust EMA weight
COMPLEMENT_MAP	interval->complement	Biases response pitch toward complementary interval of stimulus	harmonicContext key/mode
MAX_PENDING	4	Maximum concurrent feedback rounds -- caps multi-layer accumulation	L1/L2 activate cycle timing

Sensitivity: DAMPING at 0.55 produces ~2 effective round-trips before energy is negligible. Lower values (< 0.4) risk self-reinforcing oscillation between layers. Higher values (> 0.7) mean the feedback loop dies after one round, losing the "dialogue" character. Score range [0.23-0.64] with chronic trust inflation (weight 1.2+ despite avg score 0.34) suggests the EMA floor may be propping up a system that rarely delivers. Profile-dependent: performs well in high-tension sections (S1-S2), badly in atmospheric/low-density sections (S0, S5) where the round-trip energy has less material to work with.

14. Stutter Contagion - stutterContagion

Cross-layer module that propagates stutter rhythmic events from one layer to the other with adaptive decay. The #2 hotspot system with extreme peak pressure (0.944) and a dramatic regime-dependent performance swing.

Constant	Value	Role	Interaction Partners
Adaptive decay	EMA-based	Adjusts stutter propagation intensity over time	CIM setCoordinationScale, regime
Stutter variants	19 types	Selection weighted by 12 signal dimensions	stutterVariants.selectForBeat()

Sensitivity: Performance is strongly regime-dependent: score surges from 0.30->0.60 at S2 transition (evolving->coherent emergence), then declines through exploring-dominant sections (S3 score=0.58, S4=0.40). Coherent sections provide stable rhythmic patterns that stutter can predict and amplify; exploring sections are too unpredictable. In intense rivalry with phaseLock (50/50 dominance split, 116 overtake events per run). S4 hotspot rate 92% -- the explosive profile overwhelms stutter's prediction capacity.

Cross-Constant Invariants

These relationships must hold to prevent runaway behavior:

1. Density ceiling chain: coherenceMonitor.BIAS_CEILING (1.38) * negotiationEngine.playScale max (1.8) = 2.48. This is the theoretical maximum density amplification. The floor chain: BIAS_FLOOR (0.60) * playScale min (0.4) = 0.24. Exceeding ~2.5 on the ceiling causes audible note-cramming.

2. Trust-weight symmetry: adaptiveTrustScores.weight clamp [0.4, 1.8] matches negotiationEngine.playScale clamp [0.4, 1.8]. This ensures trust cannot push play probability outside the negotiation's own range.

3. Entropy regulation headroom: entropyRegulator.scale clamp [0.3, 2.0] * negotiation entropy modulator [0.5, 1.5] gives effective range [0.15, 3.0]. The negotiation's own clamps prevent this from manifesting fully.

4. Streak - hint timing: At 72 BPM, STREAK_TRIGGER = 6 ~= 5 seconds. Full hint (ramp / 8) at 14 beats ~= 12 seconds. Section lengths are typically 16-64 beats, so hints activate within one section.

5. Coherence monitor responsiveness: coherenceMonitor.SMOOTHING (0.55) means correction responds within ~2 beats (half-life). At BIAS_FLOOR 0.60, correction can reduce density by up to 40%. The check-tuning-invariants.js script validates all cross-constant invariants automatically on every run.

6. Coupling tail regime scaling: check-manifest-health.js applies the same regime scale factors to coupling tail p90 and exceedance thresholds as to the coupling matrix gate. Exploring regime relaxes thresholds by 10% (x1.10), coherent tightens by 5% (x0.95). This prevents transient sectional coupling spikes from failing the health gate during exploratory behavior.

7. Equilibrator-homeostasis interaction: The axis energy equilibrator (#13) tightens pair baselines (reducing targets), which increases gain escalation in the coupling manager, which increases total coupling energy, which triggers homeostasis (#12) to throttle _globalGainMultiplier. The throttled multiplier slows gain escalation, reducing tightening effectiveness. This creates a negative feedback loop: aggressive equilibrator tightening is self-limiting via homeostasis dampening. Floor dampening compounds this: when floorDampen < 0.50, equilibrator tightening produces <50% of intended decorrelation.

8. Effectiveness-gain ceiling chain: When effectivenessEma < 0.40, the gain ceiling cap limits max gain to GAIN_INIT + (GAIN_MAX - GAIN_INIT) * max(0.40, eff) with floor at GAIN_INIT * 1.5 = 0.24. This means low-effectiveness pairs cannot exceed gain 0.24-0.34 (depending on eff), far below GAIN_MAX (0.60). The freed budget is available to responsive pairs whose effectiveness > 0.50.

9. Graduated gate-axis balance tradeoff: Coherent gate (0.0 during coherent) freezes all equilibrator tightening. Long coherent phases (>100 beats) can cause axis imbalance (axisGini increase) because no rebalancing occurs. The evolving gate (0.4x) provides partial correction. System health requires coherent % < 50% to maintain axis balance; the regime self-balancer targets 15-35% coherent to ensure sufficient non-coherent beats for equilibrator operation.

10. Regime hysteresis and entry convergence (R37 E1/E2/E3): Consecutive-streak hysteresis (REGIME_HOLD) replaced by majority-window (R37 E1): _REGIME_WINDOW=5, _REGIME_MAJORITY=3. R36 proved REGIME_HOLD=3 still insufficient -- 87 raw coherent beats (10.6%) produced 0 resolved coherent because P(3 consecutive|p=0.106) ~ 0.12%. Majority-window: P(>=3 of 5|p=0.106) ~ 4.7%. effectiveDim coherent gate tightened 4.0->3.5 (R37 E2) -- dim almost always > 4.0, swallowing potential exploring beats. Exploring coupling gate widened 0.40->0.50 (R37 E3) -- coupling averages 0.19-0.44 made the 0.40 gate too restrictive. coherentThresholdScale starts at 0.65, floor 0.55, nudge 0.006/beat.

11. Phase retraction-floor interaction: phaseRetractionMult (0.7-1.0) dampens phaseFloorBoost when phase was recently above 115% of fair share. It does NOT affect phaseCollapseBoost or the emergency-starved 3.0 path. Minimum: at 12 beats above 115%, retractionMult = 0.70, so phaseFloorBoost is reduced to 70% of its computed value. The retraction is bypassed for genuine collapse (share < collapseThreshold), preventing deadlock. Net effect: the floor controller now resists the overshoot-boost-overshoot cycle without needing manual cap adjustments.

12. Cross-adjuster inhibit scope: CROSS_INHIBIT_WINDOW=6 only prevents DIRECTION REVERSAL between pair and axis handlers. It does NOT prevent the same handler from re-adjusting a pair in the same direction -- that uses the standard pairCooldowns (3 or 4 beats). The 6-beat window means a pair tightened by the pair-adjuster cannot be relaxed by the axis-adjuster for 6 beats, regardless of which cooldown expired first. The inhibit is asymmetric for phase: floor/extreme-collapse bypasses are unconditional because phase starvation is higher priority than oscillation prevention.