Skip to content

Configuration Reference

Shared sections

project

Metadata used for run identity and deterministic seeds.

Key Meaning
name Human-readable project name
version Workflow/config schema identity, not necessarily package version
random_seed Base seed for deterministic selection or simulation

paths

Roots for raw products, processed data, catalog snapshots, and reports. Paths are interpreted relative to the repository working directory unless absolute.

ingest

Key Meaning
mission Mission queried; supplied configs use Kepler
cadence Cadence class; supplied configs use long cadence
author MAST product author filter
max_products Maximum products per target; null means all matches
cache Reuse local raw products where present

preprocess

Quality bitmask, asymmetric sigma thresholds, maximum interpolation gap, Savitzky-Golay window/polynomial, break tolerance, iterations, and masking sigma. visualization_targets selects baseline examples only.

bls

Key Supplied value
minimum_period_days 0.5
maximum_period_days 50.0
durations_hours 1, 2, 4, 8, 12
top_k 5 where proposals are constructed
frequency_oversampling 5.0
duration_oversampling 10
minimum_peak_separation_fraction 0.01
objective snr
method fast

Durations must be strictly shorter than the minimum period. At the supplied 0.5-day bound, the configured 12-hour duration is excluded; consult BLS diagnostics for durations_hours_used and durations_hours_excluded.

Baseline-only sections

catalog and dataset

Define the NASA Exoplanet Archive table/output and the mission-product manifest. dataset.verify_fits controls structural validation of downloaded products.

targets

The fixed baseline list uses KIC identifiers, display names, optional archive aliases, and documented selection reasons.

machine_learning

Defines grouped folds, folded-view bins, negative target construction, and RF / CNN hyperparameters. The baseline RF uses 400 trees; the scale-up RF uses 600.

scaleup

  • workers: acquisition concurrency;
  • target_file: dynamically generated positive target list;
  • selection: period, S/N, magnitude, quarter, class-balance, and query limits;
  • artifacts: accepted output paths for data, models, experiments, and report.

The supplied configuration uses serial MAST acquisition because concurrent requests previously produced partial responses.

Key Supplied value / role
sample_size 250 deterministic targets
minimum_available_quarters 8
magnitude bounds 10.0–15.0
model_path Frozen RF v2 binary
model_selection Selection-policy JSON
shortlist_size 20
required_label Explicit unconfirmed-candidate label
sanity thresholds Odd/even, secondary ratio, available centroid significance
artifacts Pool, selection, candidates, shortlist, figures, report, run record

Independent validation

inputs

Paths to the frozen shortlist, eligible pool, and processed target light curves.

fap

The supplied config uses 1,000 permutations per target, eight workers, and independent segment circular shifts. minimum_roll_fraction is now honored and must be between 0 and 0.5. Unsupported shuffle methods are rejected. Content-addressed null caches include input data, settings, seed, and runtime fingerprints; legacy caches without matching metadata are not reused. See the workbench guide for migration and isolation.

vetting

Defines formal thresholds for odd/even significance, secondary eclipse, grazing shape, companion radius, Gaia risk, TESS period/S/N, and ExoFOP match.

artifacts

Defines frozen input, FAP, vetting, Gaia, crossmatch, final ranking, report, and run-record destinations.

Warning

Configuration values are coupled to the reported evidence. A changed value creates a new experiment and must use new artifact paths and interpretation.