predict(): accept one obscode per time (mixed-observatory prediction)

[matthewholman]

Summary

predict() predicted every object to a single observatory for all requested times. The rest of the chain is already per-time — obscodes_to_barycentric reads a per-row stn, _predict sets each Observation's observer position, and predict_sequence light-corrects per detection — so the single-observatory assumption lived in exactly one line (obs_data = [(obscode, t) for t in times_et]).

This lets obscode be either a single string (used for every time, as before) or a sequence of length len(times) giving one observatory per time — a mixed-observatory prediction in one call. Each time is light-corrected against its own station, while each orbit is integrated once across the whole sorted set (with the marching predict_sequence).

Backward compatibility (the important part)

A string obscode is broadcast to [obscode] * len(times), which produces the byte-for-byte identical obs_data and therefore identical output. The test asserts this directly:

predict(data, "X05", times) == predict(data, ["X05"] * len(times), times)   # exactly

Existing callers are unaffected — the predict CLI passes a single station, and interactive / broad-phase callers pass a string. A sequence whose length ≠ len(times) raises a clear ValueError.

Tests

test_predict_mixed_and_single_obscode:

• string output is byte-identical to a same-code-per-time list (backward compat);
• each time of a mixed-obscode call matches a single-obscode prediction at that station (to 1e-11 deg);
• a length mismatch raises ValueError.

Why it matters

Combined with the marching predict_sequence (#428), predicting a catalog to a whole set of (observatory, time) targets — e.g. an ITF attribution sweep — can now integrate each orbit once across all observatories and interpolate at every target, instead of grouping by observatory and re-integrating per group. This PR is independent of #428 (it is correct on its own; #428 just makes the multi-time case fast).

🤖 Generated with Claude Code

[kjnapier]

Nice