mostlyright.core.schemas.observation_merged

Merged observation-grain schema (schema.observation.merged.v1).

The frame-level identity contract for obs(granularity="observation") when the query is UNPINNED — the merged per-report rows produced by _dispatch_strategy + merge_observations, trimmed to the queried UTC window, pass-through with observation_type distinguishing METAR/SPECI (NO bucketing/resampling).

Why a DISTINCT schema. A merged frame’s df.attrs["source"] is the fused-policy tag "merged.live_v1", NOT a single upstream source — but its per-row source column is the TRUTHFUL bare parser tag (awc/iem/ghcnh). The canonical single-source schema.observation.v1 enforces per-row source == df.attrs["source"] (equality), so it correctly REJECTS a merged frame (the CWOP parity firewall stays intact — observation.py is NEVER modified). This schema instead declares:

_registered_sources = {"merged.live_v1"} — the ONLY frame-level identity it accepts. Declared as the PLURAL set (no singular _registered_source) so the validator’s source_drift_allowed audit trap (validator.py:493-500, documented at cwop/_schema.py:) never fires on a clean merged read.
_registered_row_sources = {"awc","iem","ghcnh"} — a NEW ClassVar the validator reads to switch the per-row source check from equality-to-attrs to MEMBERSHIP in this set. Absent (None) on every other schema, so satellite/forecast.station/cwop are untouched.

Columns are declared over the ACTUAL observation-grain row keys that obs(granularity="observation") emits (the merged parser row dicts — see weather/_iem.py:246-277), so a genuine merged frame passes ALL four validator checks (source, column-presence, dtype, enum), not just the source check.

Module Attributes

`MERGED_LIVE_V1`	Frame-level fused-policy identity tag stamped on UNPINNED `obs(granularity="observation")` frames.

Classes

`MergedObservationSchema`()	`schema.observation.merged.v1` — merged per-report observation rows.