prop/priv/backtest_reports/baseline_naive_gradient.md
Graham McIntire ab04cb9168 Phase 0: backtest harness
Add Microwaveprop.Backtest: a feature-evaluation framework that runs
a (lat, lon, valid_time) -> float function over the historical QSO
corpus and a matched random-time baseline, reporting distribution
statistics, distance-binned lift, and band-stratified lift.

Adds four baseline feature wrappers around the current scorer inputs
(naive_gradient, td_depression, time_of_day, pressure), a mix backtest
CLI, and the first set of baseline reports under priv/backtest_reports
so downstream phases have a frozen reference point to compare against.
2026-04-09 16:10:54 -05:00

1.3 KiB

Backtest: Microwaveprop.Backtest.Features.naive_gradient

QSO sample: 5000
Baseline sample: 5000

Matched distribution

Set N Mean Stddev p50 p90 Min Max
QSO times 5000 -113.971 42.207 -105.154 -69.043 -331.397 -35.521
Random baseline 151 -118.002 43.405 -107.528 -72.860 -268.079 -41.047

Lift by distance (km)

Set N Mean Stddev p50 p90 Min Max
0-100 1249 -114.766 43.617 -107.614 -66.251 -321.312 -35.521
100-250 2295 -114.555 42.595 -103.392 -71.537 -331.397 -44.027
250-500 1288 -112.728 41.507 -105.238 -65.613 -321.312 -39.224
500-1000 167 -109.723 29.535 -115.466 -69.348 -206.669 -45.098
1000+ 1 -90.610 0.000 -90.610 -90.610 -90.610 -90.610

Lift by band

Set N Mean Stddev p50 p90 Min Max
10000 MHz 4444 -114.037 42.734 -105.238 -68.535 -321.312 -37.154
24000 MHz 465 -116.766 37.728 -107.614 -82.149 -331.397 -64.361
47000 MHz 77 -98.886 35.255 -93.461 -68.535 -227.800 -35.521
75000 MHz 14 -83.273 13.340 -93.461 -68.535 -99.652 -68.535