std.adaptive_dynamics

Markdown Open in ChatGPT Open in Claude

Generated by sema doc from crates/sema-runtime/assets/stdlib/sema/adaptive_dynamics.sema. Import with from std.adaptive_dynamics import …. For a narrative introduction see std.adaptive_dynamics.

`adaptive_dynamics`

std.adaptive_dynamics - bounded adaptive-model lifecycle primitives.

The module separates state updates, parameter refits, structural switching, and safe unknown outcomes. Proposed models cannot become active without held-out validation, invariant evidence, dwell time, hysteresis, and an audit record. Numerical fitting and domain equations remain application responsibilities.

`enum ModelLifecycle`

Variants

proposed
validated
active
retired
rejected

`enum AdaptationMode`

Variants

retain
state_update
parameter_refit
structural_switch
ensemble
safe_unknown

`enum RegimeStatus`

Variants

stable
suspected
shifted
unknown

`enum ValidationStatus`

Variants

passed
failed
inconclusive

`enum TransitionKind`

Variants

observation
regime_suspected
regime_shifted
parameter_refit
candidate_proposed
candidate_validated
candidate_rejected
model_activated
horizon_contracted
horizon_restored
action_abstained
action_executed

`struct ModelDescriptor`

Fields

field	type	descriptor
`id`	`str`
`family`	`str`
`version`	`int`
`lifecycle`	`ModelLifecycle`
`parameter_names`	`list[str]`
`parameters`	`list[f64]`
`structure_signature`	`str`
`assumptions`	`list[Assumption]`
`validity`	`ValidityRegion`
`fit_error`	`f64`
`validation_error`	`f64`
`created_at`	`f64`
`activated_step`	`int`

`struct SelectionPolicy`

Fields

field	type	descriptor
`min_validation_improvement`	`f64`
`max_validation_error`	`f64`
`max_invariant_violations`	`int`
`min_dwell_steps`	`int`
`hysteresis_margin`	`f64`
`required_horizon_steps`	`int`

`struct CandidateValidation`

Fields

field	type	descriptor
`model`	`ModelDescriptor`
`status`	`ValidationStatus`
`training_error`	`f64`
`validation_error`	`f64`
`relative_improvement`	`f64`
`invariant_violations`	`int`
`evidence`	`list[Evidence]`
`evaluated_at`	`f64`

`struct ResidualDetector`

Fields

field	type	descriptor
`window_size`	`int`
`min_samples`	`int`
`score_threshold`	`f64`
`required_consecutive`	`int`
`baseline_mean`	`f64`
`baseline_scale`	`f64`
`history`	`list[f64]`
`consecutive_high`	`int`
`score`	`f64`
`status`	`RegimeStatus`
`updated_at`	`f64`

`struct ModelTransition`

Fields

field	type	descriptor
`sequence`	`int`
`kind`	`TransitionKind`
`from_model_id`	`str`
`to_model_id`	`str`
`regime`	`RegimeStatus`
`adaptation`	`AdaptationMode`
`reason`	`str`
`evidence`	`list[Evidence]`
`occurred_at`	`f64`

`enum AdaptationDecision`

Variants

selected(AdaptationMode, ModelDescriptor, str)
abstained(AdaptationMode, str)

`enum ActivationDecision`

Variants

activated(ModelDescriptor, ModelDescriptor, ModelTransition)
blocked(str, ModelTransition)

`def model_with_lifecycle`

def model_with_lifecycle(model: ModelDescriptor, lifecycle: ModelLifecycle, fit_error: f64, validation_error: f64, activated_step: int)

Parameters

name	type
`model`	`ModelDescriptor`
`lifecycle`	`ModelLifecycle`
`fit_error`	`f64`
`validation_error`	`f64`
`activated_step`	`int`

`def make_detector`

def make_detector(window_size: int, min_samples: int, score_threshold: f64, required_consecutive: int, baseline_mean: f64, baseline_scale: f64) -> ResidualDetector !{}

Parameters

name	type
`window_size`	`int`
`min_samples`	`int`
`score_threshold`	`f64`
`required_consecutive`	`int`
`baseline_mean`	`f64`
`baseline_scale`	`f64`

Returns ResidualDetector

Effects !{}

`def append_bounded`

def append_bounded(values: list[f64], value: f64, limit: int) -> list[f64] !{}

Parameters

name	type
`values`	`list[f64]`
`value`	`f64`
`limit`	`int`

Returns list[f64]

Effects !{}

`def update_detector`

def update_detector(detector: ResidualDetector, residual: f64, observed_at: f64) -> ResidualDetector !{}

Parameters

name	type
`detector`	`ResidualDetector`
`residual`	`f64`
`observed_at`	`f64`

Returns ResidualDetector

Effects !{}

`def relative_improvement`

def relative_improvement(reference_error: f64, candidate_error: f64)

Parameters

name	type
`reference_error`	`f64`
`candidate_error`	`f64`

`def validate_candidate`

def validate_candidate(active: ModelDescriptor, candidate: ModelDescriptor, training_error: f64, validation_error: f64, invariant_violations: int, evidence: list[Evidence], evaluated_at: f64, policy: SelectionPolicy) -> CandidateValidation !{}

Parameters

name	type
`active`	`ModelDescriptor`
`candidate`	`ModelDescriptor`
`training_error`	`f64`
`validation_error`	`f64`
`invariant_violations`	`int`
`evidence`	`list[Evidence]`
`evaluated_at`	`f64`
`policy`	`SelectionPolicy`

Returns CandidateValidation

Effects !{}

`def choose_adaptation`

def choose_adaptation(regime: RegimeStatus, parameter_fit: CandidateValidation, structural_fit: CandidateValidation)

Parameters

name	type
`regime`	`RegimeStatus`
`parameter_fit`	`CandidateValidation`
`structural_fit`	`CandidateValidation`

`def transition`

def transition(sequence: int, kind: TransitionKind, from_model_id: str, to_model_id: str, regime: RegimeStatus, adaptation: AdaptationMode, reason: str, evidence: list[Evidence], occurred_at: f64)

Parameters

name	type
`sequence`	`int`
`kind`	`TransitionKind`
`from_model_id`	`str`
`to_model_id`	`str`
`regime`	`RegimeStatus`
`adaptation`	`AdaptationMode`
`reason`	`str`
`evidence`	`list[Evidence]`
`occurred_at`	`f64`

`def blocked_activation`

def blocked_activation(active: ModelDescriptor, candidate: ModelDescriptor, step: int, sequence: int, reason: str, evidence: list[Evidence])

Parameters

name	type
`active`	`ModelDescriptor`
`candidate`	`ModelDescriptor`
`step`	`int`
`sequence`	`int`
`reason`	`str`
`evidence`	`list[Evidence]`

`def activate_validated`

def activate_validated(active: ModelDescriptor, candidate: ModelDescriptor, step: int, sequence: int, evidence: list[Evidence], policy: SelectionPolicy) -> ActivationDecision !{}

Parameters

name	type
`active`	`ModelDescriptor`
`candidate`	`ModelDescriptor`
`step`	`int`
`sequence`	`int`
`evidence`	`list[Evidence]`
`policy`	`SelectionPolicy`

Returns ActivationDecision

Effects !{}