k8s(prod): promote v0.8.14 by digest

[cboettig]

Promotes prod v0.8.12 → v0.8.14. Replaces #362 (v0.8.13), closed as superseded.

-  image: …:v0.8.12@sha256:332d6df730c1fc66e7b2d317b121828fcb3785339a3a43b49789c3e22e6e0ffd
+  image: …:v0.8.14@sha256:9dda8e42f0bcaec6237e8cb677ce26dc515d16bab17fd1500e108cb27ad8b880

Digest cross-checked two ways — the v0.8.14 tag and the 0f2fc25… git-sha tag resolve to the same digest — and the method validated by resolving v0.8.12, which returns sha256:332d6df…, exactly the digest this commit replaces.

Why this matters

Prod has been on v0.8.12 the whole time, so it is still serving the pre-#357 unweighted overlay guidance — the silent-wrong-answer bug from #356 that a partner reported against the published 2025 Biodiversity Assessment. This is the first release that actually fixes it.

What ships

PR
#357	fractional-overlay examples weight by h3_cell_area, use published hex-weights, bound with the ecoregion land grid (#356)
#360	rebuts "the area terms cancel" — the rationalization that survived #357
#365	coarse-overlay example uses land_area_km2, not the removed nland (#364)
#358	opt-in experimental raster+zarr image — separate tag/endpoint, inert here
#351, #342	guidance-validation runbook fix; dev digest pin + convergence check (#341)

Validation

res-10 path — gated on dev, 3 trials × 2 models: 6/6. Zero occurrences of the pre-fix 28.8% form and zero of the 32.1% wrong-method-inside-tolerance form. Every trial carried the area term and applied the California mask.

res-8 path — verified by hand, because the tier cannot test it. This is the honest caveat: the regression tier has no res-8 question, which is exactly how #364 shipped through a green 6/6 gate. Direct check of the shipped recipe:

shipped recipe, res 8	26.135
res-10 ground truth, same scope	26.135
pre-data-workflows#522 nland count-weighting	25.684
what v0.8.13 shipped	binder error

Coverage gap filed as geo-agent-benchmark#21; do not read the 6/6 as covering the coarse path.

Baseline unchanged: Mojave 7,373,400 · GAP 1+2 26,471,500 · hardwood woodland 13.6% · California 26.08–26.1% · BioRank 21.18–21.2 · channelized 22.5%.

After merge

kubectl apply -f k8s/deployment.yaml
kubectl rollout restart deployment/duckdb-mcp -n biodiversity
kubectl -n biodiversity get pods -l app=duckdb-mcp \
  -o custom-columns='NAME:.metadata.name,IMAGE:.status.containerStatuses[0].imageID'   # all 6 converge

Coordination

Further edits to this guidance block are sequenced under #367, behind the gate-coverage work (geo-agent-benchmark#21, #16, #9). This should be the last patch to that section before a single design pass covering #363/#354/#355/#289/#353.