Build and Push / Build and Push Docker Image (push) Failing after 4m44s

Details

algo: recalibrate per-band weights from prod corpus (2026-08-01)

First full recalibration since 2026-05-24. Corpus is 1,726,757 PSKR
samples (50/144/432 MHz) plus 74,072 contacts joined to hrrr_profiles
(microwave), yielding overrides for 12 bands.

The 6m fit moved most, and coherently: humidity 0.125 -> 0.204 and pwat
0.120 -> 0.185, paid for by season 0.217 -> 0.147. 6m contributes 1.63M
of the 1.73M PSKR samples, so ten weeks of new spot density pulled the
band off a seasonal prior and onto direct moisture signal. Next largest
is 144 MHz refractivity 0.111 -> 0.053.

75 GHz (n=93) and 122 GHz (n=63) appear in the evidence table but get
no overrides -- they fall under MIN_N_FOR_FIT, which is correct; 75 GHz
reports rho_pr of -0.774 on 93 samples, which is noise, not physics.

Caveat: the refractivity multipliers are numerically unstable. They are
computed as |band_rho_grad| / |ref_rho_grad| against the 10 GHz
reference, whose rho_grad is +0.030 -- an order of magnitude smaller
than the other reference correlations. Two runs minutes apart moved
47 GHz refractivity 0.0688 -> 0.0522 (-24% relative) purely on new rows
landing between queries. The stable factors barely moved. These weights
are live once deployed, so this commit is isolated from the script fixes
to keep it independently revertable.

2026-08-01 16:35:11 -05:00

4.6 KiB

Raw Blame History

Full Recalibration Report — 2026-08-01

Auto-generated by scripts/recalibrate.py. The companion JSON file at priv/algo/band_weights.json is the machine-readable output that Elixir loads at runtime. This report is the human-readable audit trail.

Connection: postgres://prop:***@10.0.19.30:6432/prop Generated: 2026-08-01T21:30:41+00:00

Data sources

PSKR (VHF/UHF, label = spot_count): bands [50, 144, 432, 1296, 2304, 3400, 10000], 1,726,757 samples
Contacts ↔ HRRR (microwave, label = distance_km): 12 bands, 74,072 matched contacts

Per-band evidence

Band	Source	n	label avg	ρ_tc	ρ_dpc	ρ_pr	ρ_pwat	ρ_grad
50 MHz	pskr	1,633,976	2.24	+0.113	+0.102	+0.044	+0.114	-0.033
144 MHz	pskr	87,724	3.07	+0.057	+0.101	+0.140	+0.117	+0.004
222 MHz	contacts	7,517	229.17	-0.082	+0.078	-0.012	+0.069	+0.041
432 MHz	pskr	4,152	1.62	+0.094	-0.205	+0.123	-0.188	+0.207
902 MHz	contacts	1,753	172.23	-0.070	+0.094	+0.077	+0.047	-0.041
1296 MHz	contacts	2,951	160.98	-0.035	+0.100	+0.023	+0.052	-0.072
2304 MHz	contacts	718	154.72	-0.053	+0.169	+0.082	+0.128	-0.094
3400 MHz	contacts	334	154.14	-0.052	+0.149	-0.025	+0.145	+0.076
5760 MHz	contacts	320	127.92	-0.058	+0.071	+0.151	+0.097	-0.020
10000 MHz	contacts	47,260	208.25	-0.001	-0.024	-0.089	-0.029	+0.030
24000 MHz	contacts	3,330	95.36	-0.215	-0.333	-0.204	-0.366	+0.005
47000 MHz	contacts	661	62.80	-0.121	-0.187	-0.246	-0.220	-0.027
75000 MHz	contacts	93	49.90	-0.688	-0.696	-0.774	-0.636	+0.495
122000 MHz	contacts	63	16.26	+0.038	-0.130	-0.042	-0.128	+0.356

Derived per-band weights

Band	n	source	humidity	time_of_day	td_depression	refractivity	sky	season	wind	rain	pwat	pressure
50 MHz	1,633,976	pskr	0.2039	0.0184	0.1632	0.0823	0.0679	0.1466	0.0679	0.0116	0.1853	0.0528
144 MHz	87,724	pskr	0.2020	0.0213	0.1617	0.0527	0.0673	0.1271	0.0673	0.0115	0.1836	0.1055
222 MHz	7,517	contacts	0.1947	0.0273	0.1815	0.1000	0.0756	0.1325	0.0756	0.0129	0.1565	0.0434
432 MHz	4,152	pskr	0.1869	0.0225	0.1496	0.1460	0.0623	0.1008	0.0623	0.0106	0.1699	0.0892
902 MHz	1,753	contacts	0.2167	0.0302	0.1787	0.0988	0.0744	0.1103	0.0744	0.0127	0.1256	0.0782
1296 MHz	2,951	contacts	0.2289	0.0345	0.1510	0.1361	0.0763	0.1029	0.0763	0.0130	0.1350	0.0462
2304 MHz	718	contacts	0.1916	0.0289	0.1530	0.1344	0.0639	0.0861	0.0639	0.0344	0.1742	0.0698
3400 MHz	334	contacts	0.1983	0.0298	0.1574	0.1217	0.0661	0.0891	0.0661	0.0503	0.1802	0.0411
5760 MHz	320	contacts	0.1621	0.0299	0.1591	0.0676	0.0662	0.0893	0.0662	0.0797	0.1696	0.1101
10000 MHz	47,260	contacts	0.1262	0.0380	0.1010	0.0986	0.0841	0.1134	0.0841	0.1431	0.1147	0.0967
24000 MHz	3,330	contacts	0.1514	0.0410	0.1211	0.0402	0.0504	0.0748	0.0504	0.2270	0.1376	0.1061
47000 MHz	661	contacts	0.1402	0.0528	0.1122	0.0522	0.0467	0.0756	0.0467	0.2385	0.1275	0.1075

Changes vs prior `band_weights.json`

Largest weight moves (≥0.5 pp), sorted by magnitude:

Band	Factor	Prior	New	Δ
47000 MHz	refractivity	0.0688	0.0522	-0.0166
3400 MHz	refractivity	0.1090	0.1217	+0.0127
2304 MHz	refractivity	0.1439	0.1344	-0.0095
5760 MHz	refractivity	0.0764	0.0676	-0.0088
24000 MHz	refractivity	0.0453	0.0402	-0.0051

Applying these weights

These weights are live scoring inputs, not documentation. BandConfig.weights/1 consults BandWeights.lookup/1 (which reads this JSON) before falling back to the in-source :weights override or the global @weights default. This script still never edits band_config.ex — it only rewrites the JSON.

BandWeights caches the parsed file in :persistent_term on first read, so a running node keeps serving the previous weights until it restarts. To put this run into effect, deploy (the file ships under priv/) or call BandWeights.reset/0 on every node.

4.6 KiB Raw Blame History Unescape Escape