explorer: halo crescents / disappearing dots over hilly terrain (root cause not yet identified)

[rdhyee]

Symptom

After PR #180 added 1.5px white halos to point primitives in the Interactive Explorer, dots over hilly/mountainous terrain render with partial halos (crescents) at moderate zoom, and disappear entirely at close zoom over higher terrain.

Reproducer: https://isamples.org/explorer.html#v=1&lat=33.2706&lng=-86.2375&alt=311435&heading=360.0 — Birmingham, AL hill country.

Visible in the deployed prod site (current HEAD 9d9291c, post-revert).

What we've ruled out

Two fix attempts so far, both reverted/closed:

• PR explorer: fix halo cutoff over terrain (disableDepthTestDistance) #181 (d681758, reverted via Revert "explorer: fix halo cutoff over terrain (disableDepthTestDistance)" (#181) #183): added disableDepthTestDistance: Number.POSITIVE_INFINITY to all three PointPrimitive.add() sites. Caused worse regressions (Codex review): back-side-of-globe primitive bleed-through and broken hover/click pickability. Cesium docs confirm POSITIVE_INFINITY disables depth-test against the globe ellipsoid, not just terrain.
• PR explorer: fix halo terrain occlusion via scene.globe.depthTestAgainstTerrain (v2) #184 (closed as no-op): set scene.globe.depthTestAgainstTerrain = false at viewer init. Codex caught this — it's already the Cesium 1.127 default. The artifact occurs despite depthTestAgainstTerrain being false.

So whatever's causing the crescents, it's not the standard Cesium terrain depth-test.

Hypotheses worth checking

• Alpha-blended fill becoming invisible: withAlpha(0.8) / 0.85 / 0.9 on the fill (color) means at small pixel sizes the fill blends into bright satellite imagery, leaving only the opaque white outline → reads as an empty ring or crescent.
• Cesium PointPrimitive shader artifact at certain pixelSize / outlineWidth ratios under scaleByDistance (NearFarScalar).
• Atmosphere or translucent ocean layer occluding from one direction at low altitudes.
• GPU / DPR-specific: I cannot reproduce locally at canvas ~919×689 / DPR=1; the user's repro screenshot is at much higher resolution. Possibly a high-DPR-only artifact.

What was attempted

PR	Attempt	Outcome
#181	disableDepthTestDistance: Infinity on all primitives	Reverted (#183) — bigger regressions
#184	globe.depthTestAgainstTerrain = false	Closed — no-op, matches Cesium default

Suggested next steps

1. Reproduce on prod with devtools — toggle the fill alpha to 1.0, toggle outlineWidth to 0, toggle halo entirely. Whichever toggle resolves the crescents identifies the cause.
2. Capture viewport/DPR/GPU info at repro time so future investigators can match conditions.
3. Consider deferring: the halos remain a real a11y win on land regions where the fill is well-contrasted; the artifact is only on certain terrain. May be lower priority than other items in Interactive Explorer rethink: architecture review + UX/feature backlog #163/Interactive Explorer: integrate Hana mockup feedback (Light path on search semantics) #178 follow-ups.

Refs

• PR explorer: native Cesium toolbar + prominent share button + a11y halos #180 (e0043c8) — added halos
• PR explorer: fix halo cutoff over terrain (disableDepthTestDistance) #181 (d681758) — first fix attempt, reverted
• PR Revert "explorer: fix halo cutoff over terrain (disableDepthTestDistance)" (#181) #183 (9d9291c) — revert
• PR explorer: fix halo terrain occlusion via scene.globe.depthTestAgainstTerrain (v2) #184 — closed as no-op
• Interactive Explorer: integrate Hana mockup feedback (Light path on search semantics) #178 — Hana mockup integration umbrella

Filed because we're parking this to return to URL-state observability work; tracking so we can pick it up with full context later.