hamsci-physics

Ionospheric science products derived from HF time-standard observations — the science half of the 2026-08-24 hf-timestd split.

hf-timestd keeps the real-time timing core: RTP→UTC sample labelling, the broadcast fusion that produces D_clock, and the chrony feed. Everything that turns those labelled observations into ionospheric products lives here:

Product	Module	Unit
L3 TEC / dTEC fusion	physics_fusion_service	hamsci-physics-fusion.service (held — see below)
Daily reanalysis of arrivals	ionospheric_reanalysis	hamsci-physics-reanalysis.timer
Travelling ionospheric disturbances	tid_detector	(in-process, fusion)
Per-path propagation statistics	propagation_stats	(in-process, reanalysis)
GRAPE / PSWS daily datasets	grape/	grape-daily.timer
IONEX / IRI / space-weather inputs	cddis, scripts/	hamsci-physics-ionex-download.timer, …-iri-update.timer

Where the boundary is

Three rules define the split, and the test suite pins all three:

1. hamsci-physics never imports hf_timestd (tests/test_import_lint.py). Shared engines — TEC solvers, propagation models, the ionosphere stack, the io/schema data layer — come from hamsci-dsp, which both repos depend on.
2. The data root does not move. /var/lib/timestd is a frozen contract: this client reads the timing core's products in place (raw_buffer/, phase2/) and writes its own under phase2/fusion, phase2/science and upload/. Only the config moved, to /etc/hamsci-physics/config.toml.
3. The GRAPE upload pipeline moved verbatim. Its source_id (grape-datasets:/var/lib/timestd/upload) and transport name are hs-uploader watermark keys over a keep-retention spool; changing them would re-ship every dataset PSWS already has (tests/test_deploy_contract.py).

Install

Through sigmond, like any other client:

smd install hamsci-physics
smd config init hamsci-physics     # fills [station] from the station identity
smd start hamsci-physics

deploy.toml is the manifest sigmond reads. hf-timestd must be installed on the same host — it produces what this client consumes.

Development

uv sync --extra dev
uv run pytest tests/            # ~210 tests

hamsci-dsp and hs-uploader are editable siblings ([tool.uv.sources]), so a git pull of either propagates without a reinstall.

Contract surface

hamsci-physics version --json
hamsci-physics inventory --json    # what this client is and writes
hamsci-physics validate --json     # exit 1 on any fail-severity issue
hamsci-physics grape daily         # decimate → spectrogram → package → upload

Status

All three products are live. The L3 fusion service runs continuously — on AC0G/B4 it writes about 14 carrier-phase dTEC records a minute, plus differential dTEC across WWV, WWVH and BPM, anchored to GNSS VTEC.

One known wart, inherited from the HDF5→SQLite migration: the service's upstream-freshness check and channel discovery still read the old file-tree layout. Neither gates the science (the freshness result is advisory — the loop continues either way), but a permanently "stale" health signal is a metric that lies, so both are being moved to DB reads.