Drift-Aware Calibration Scheduling

Commands

1. Simulation evaluation (no AWS spend)

python -m calib_sched.run_sim --config configs/default.yaml --out results/sim

2. Generate synthetic logged data with propensities

python -m calib_sched.run_sim \
  --config configs/default.yaml \
  --out results/sim \
  --log-out data/sim/logs.parquet

3. Off-policy evaluation (IPS + DR + LCB)

python -m calib_sched.ope_eval \
  --config configs/default.yaml \
  --log data/sim/logs.parquet \
  --policy conservative_bandit \
  --out results/ope/ope_report.json

4. Hardware sentinel submit (manual + confirmation gate)

Dry-run preview (safe default):

python -m calib_sched.run_hardware \
  --config configs/default.yaml \
  --device rigetti_ankaa3 \
  --shots-scale 1.0

Explicit preflight validation only (no metadata write, no submission):

python -m calib_sched.run_hardware \
  --config configs/default.yaml \
  --device rigetti_ankaa3 \
  --shots-scale 1.0 \
  --preflight-only

Actual submit (requires both confirmation flags):

python -m calib_sched.run_hardware \
  --config configs/default.yaml \
  --device rigetti_ankaa3 \
  --confirm-spend \
  --yes-i-understand \
  --shots-scale 1.0

5. Parse completed hardware tasks + append time-series CSV

python -m calib_sched.parse_hardware \
  --config configs/default.yaml \
  --metadata data/hardware/raw_tasks/<metadata_file>.json

Sentinel Plan in configs/default.yaml

Default hardware sentinel suite per timepoint (5 circuits):

1. readout_zero on [0,1]
2. readout_one on [0,1]
3. coherent_rx repeats=4 on [0]
4. coherent_rx repeats=8 on [0]
5. crosstalk parity probe on pair [0,1]

Default shots per circuit: 1500 (scaled by --shots-scale).

Default device: sv1 (safe simulator default in config). Use --device rigetti_ankaa3 explicitly when you intentionally want QPU submissions.

Tests

python -m unittest discover -s tests -p 'test_*.py'