Adaptive FIM next-experiment proposals

BL-80 turns leakage and exact-state-retention counts into conservative proposals for a future static lambda_fim batch. It is an offline experimental-design surface, not a realtime controller and not evidence that FIM feedback protects hardware coherence.

Decision path

1. Supply disjoint leakage and exact-state-retention event counts from one shot block with provenance.
2. Approve the complete paired control/candidate shot plan through the BL-47 no-submit budget.
3. Compute a two-sided Wilson score interval for the configured witness.
4. Decrease lambda_fim only when the harmful-direction confidence bound clears the target and deadband. Ambiguous, missing-count, and underpowered evidence holds the current value.
5. Export an observer record and an unapplied scalar proposal to BL-33.

The product deliberately has no increase action. The committed repeated IBM follow-up found that lambda_fim = 4 increased leakage and reduced retention for the tested circuit family, so BL-80 does not reward larger feedback without a new, separately preregistered policy.

from scpn_quantum_control.analysis.adaptive_fim_feedback import (
    AdaptiveFIMConfig,
    FIMWitness,
    plan_adaptive_fim_schedule,
)

witness = FIMWitness.from_counts(
    leakage_events=60,
    retention_events=400,
    shots=512,
    depth=2,
    source="simulator",
    artifact_id="example-count-block",
)
plan = plan_adaptive_fim_schedule(
    4.0,
    (witness,),
    policy_id="ci_dry_run_only",
    shots_per_arm=128,
    config=AdaptiveFIMConfig(target_leakage=0.05, step_gain=4.0),
)

assert plan.allowed
assert plan.steps[0].decision == "decrease"
assert plan.budget.would_submit is False

shots_per_arm describes the future paired control/candidate batch. BL-47 gates that complete plan before any interval is evaluated or observer is created. An over-budget or hardware request returns a refused plan with no steps and no observer records.

Interfaces

Interface	Role
FIMWitness.from_counts(...)	Count-bound, provenance-carrying leakage and retention witness
wilson_score_interval(...)	Closed-form binomial interval used by the decision gate
propose_count_aware_lambda(...)	One conservative decrease-or-hold proposal
adaptive_count_aware_schedule(...)	Deterministically thread proposals through committed witnesses
plan_adaptive_fim_schedule(...)	BL-47 budget composition and BL-33 observer creation
codesign.adaptive_fim_proposal_port(...)	Convert a step into an unapplied one-parameter ControllerProposal
propose_next_lambda(...)	Legacy point-estimate compatibility helper; not a product/evidence route

Evidence

Run:

PYTHONPATH=src python scripts/run_adaptive_fim_evidence.py

The committed evidence contains three synthetic calibration controls and an offline replay of circuits 0, 3, and 7 from the already committed repeated FIM raw-count artefact. The source SHA256 and exact job identifier are bound into the payload. The synthetic actions are decrease -> hold -> hold; the selected historical adverse witnesses produce three deterministic decreases.

This establishes input validation, uncertainty gating, deterministic replay, budget refusal, and hardware refusal. It does not establish that executing a proposed batch would improve leakage or retention. Closed-loop efficacy remains untested.

• Evidence JSON
• Evidence summary
• Historical protocol boundary

Scientific basis and limits

Adaptive experimental design chooses later experiments using information from earlier outcomes. Ferrie, Granade, and Cory developed that structure for bounded Hamiltonian estimation (AIP Conference Proceedings 1443, 165); Hincks and colleagues evaluated online design against fixed sweeps at matched data volume (arXiv:1806.02427). For binomial uncertainty, Brown, Cai, and DasGupta document the poor coverage of the Wald interval and recommend Wilson or Jeffreys alternatives (Statistical Science 16, 101).

Those sources motivate sequential proposal semantics and interval-aware count handling. They do not validate this proportional update rule. BL-80 claims no optimal policy, Bayesian posterior, realtime feedback, provider submission, hardware execution, FIM protection, control stability, or quantum advantage.