[ameersohel45]

Summary

Brings this repo's mappings back in line with the copies helmcharts serves. They drifted when helmcharts#25 landed there first; the three files are byte-identical to that branch again, and pocra already was.

Three changes, two of them the schema pack's:

1. network-specs#12 — a resolved location is a Beckn Location, GeoJSON under geo.
2. network-specs#13 — MandiPrice drops commodity; supportedCommodities is the one coded collection in both modes.
3. sourceName is the document's title, not the ingestion pipeline's filename.

Both upstream PRs are merged and their packs are published, so nothing here is waiting on anything.

Changes

#12 — location becomes a Beckn Location

The weather mapping both emits a location and reads one, and all four sites move:

site	from	to
what it writes	{type, coordinates}	{geo: {type, coordinates}}
required: precondition	location.type	location.geo.type
request half $lat/$lon	location.coordinates	location.geo.coordinates
response half $asked	location.coordinates	location.geo.coordinates

The read sites are the ones worth noting — converting only the write half refuses every select with SCH_INVALID_FORMAT: this capability needs a Point location, which reads like a bad payload and is actually the precondition looking where the geometry no longer is. That was caught in oan-local on the helmcharts side, not in review.

response.location.lat is untouched: that is the upstream's response shape, not Beckn's.

#13 — commodity folds into supportedCommodities

One entry per resource, not the caller's whole list. From that PR's compatibility note: a Direct resource's single prices object applies to every commodity listed. This mapping emits one resource per price record, each with its own prices — so listing more would claim that row's prices for commodities it never priced.

The code stays the caller's and the name the record's, because the upstream takes a commodity code and answers with a name only.

market needed nothing: #13 relaxes it to an anyOf across marketCode/marketName/district/state/location, and this mapping already emits three of the five.

sourceName is the document's title

was   "doc-d7124029cb60-rerun-1788841718"
now   "Karnataka Gruha Jyoti Scheme Guidelines"

No fallback, and the blank is the point. A document with no name_en gets "". Falling back to filename hides the gap behind something that looks like an answer but names nothing a reader can place; omitting the key leaves a consumer unable to tell "this document has no name" from "the provider reported nothing". sourceId survives either way, so the passage stays attributable.

title_en is not a third candidate — it exists on the hit and comes back empty on this corpus.

The tests are the work here

They encoded the old shapes, so each had to be re-argued rather than mechanically flipped.

WeatherObservation — the inbound fixture and the precondition helper both wrap the geometry, so the refusal cases (polygon, line string, multipoint, absent) still exercise the shape the mapping actually reads rather than passing for the wrong reason.

MandiPrice — supportedCommodities moves from the must-NOT-appear list to the required one, and a new assertion pins exactly one entry carrying a code.

KnowledgeAdvisory — the captured providerResponse predates name_en. The live provider returns it today, so the four hits gain the titles their two documents really carry; that makes the fixture current rather than fictional. The assertion now refuses a doc- prefix outright, so a silent regression to filenames fails loudly.

New: TestShippedMappingLeavesAnUnnamedSourceBlank covers the case the change exists to produce — name_en absent, empty, and whitespace-only all yield "" while sourceId survives.

Testing

• go build ./..., go vet, and the full go test ./pkg/... are green
• -race pass over all four capability plugins
• The new blank-source test carries a negative control: restoring the filename fallback fails all three cases with sourceName = "doc-d1", want ""
• The three mappings verified byte-identical to helmcharts release-0.0.1
• The same mapping content was run end to end in oan-local on the helmcharts side — weather and mandi selects both answering on_select against the merged packs
• Not re-run end to end from this repo

Notes

Based on release-0.0.1, not development. pocra's mapping on release-0.0.1 is already identical to helmcharts', and development is behind on it — basing there would have meant either porting someone else's POCRA work as a side effect or silently regressing it.

development is still behind on all four mappings and will need the same sync, including the pocra changes from helmcharts#20 and #21 that never came back here. Worth doing as its own pass rather than folding in here.

[github-actions]

🛡️ Trivy security scan (CRITICAL,HIGH,MEDIUM,LOW)

View full run

Go dependencies

No findings at CRITICAL,HIGH,MEDIUM,LOW.

Container image

Package	Severity	Installed	Fixed in	Advisory
zlib	MEDIUM	1.3.2-r7	1.3.3-r0	CVE-2026-85091

[github-actions]

📊 Test Coverage: ➖ Not applicable — no non-test Go files changed vs origin/release-0.0.1, so there are no lines to measure and no percentage to report.