Clarify fire visualization architecture and define fire_plot recovery path

AGENTS.md

@@ -179,3 +179,18 @@ Docs checks:
 - Preserve backward compatibility unless explicitly instructed otherwise.
 - Add a focused regression test for every bug fix.
 - Use synthetic small cubes for deterministic unit tests.
+
+## 13) Visualization architecture and fire plot recovery plan
+
+The repository currently contains more than one visualization backend. Treat this as a transition state, not as permission to let them drift further apart. The custom HTML/CSS/JS cube viewer exposed through v.plot() is the canonical interactive viewer for general cube rendering. The fire-specific v.fire_plot() workflow currently still exposes a Plotly-based hull figure as its primary interactive display. Future work should move toward shared space-time semantics and clearer renderer boundaries, not ad hoc backend switching.
+
+When touching visualization code, separate four concerns: (1) event or cube data construction, (2) geometry generation, (3) scene or adapter packaging, and (4) rendering backend. Changes that blur those layers tend to create regressions and make migration harder.
+
+Implementation rules for agents:
+- Do not silently replace one viewer system with another without also updating docs, tests, and architectural notes.
+- If you touch axis placement, time-depth direction, cube-attached labels, or camera semantics, read `docs/dev/cube_viewer_invariants.md` first and keep those invariants aligned.
+- Fire-plot work must preserve separation between event geometry generation, climate/event data fusion, and rendering backend.
+- Do not claim fire_plot is fully migrated unless rendering actually runs through the custom cube-viewer path.
+- Do not update screenshots/docs only while leaving contradictory code behavior.
+- Do not change coordinate conventions without updating `docs/dev/cube_viewer_invariants.md` and relevant tests together.
+- When in doubt, prefer an adapter layer over duplicating geometry logic in two renderers.

docs/api/verbs.md

@@ -74,8 +74,7 @@ Plot inside/outside climate distributions for a fire event; side-effecting
 viewer verb.
 
 ### ``fire_plot(...)``
-High-level convenience verb that renders time-hull geometry and climate fields
-as a multi-panel visualization.
+High-level fire workflow that returns event/hull/summary outputs and a Plotly-based interactive hull figure (`fig_hull`).
 
 ### ``fire_panel(...)``
 Compact panel combining time-hull outlines and climate histograms.

docs/dev/fire_plot_architecture.md

@@ -0,0 +1,45 @@
+# Fire plot architecture and migration path
+
+## 1) Purpose
+
+This note documents the current fire-visualization architecture and gives a safe migration path toward shared viewer semantics.
+
+A successful migration will not begin by swapping plotting libraries. It will begin by making the fire-event products renderer-neutral enough that the same event geometry can drive either the existing Plotly hull figure or a future custom cube-viewer scene.
+
+## 2) Current state
+
+- `v.plot()` is the canonical custom HTML/CSS/JS cube viewer for general cube rendering.
+- `v.fire_plot()` currently returns a fire-event analysis bundle and uses a Plotly hull figure (`fig_hull`) as the supported interactive fire-specific display.
+- Fire geometry and climate sampling logic live in backend-agnostic parts of the fire workflow, while the final figure is currently Plotly-specific.
+
+## 3) Stable separations of concern
+
+Keep these layers distinct when modifying fire visualization:
+
+1. Event or cube data construction
+2. Geometry generation
+3. Scene/adapter packaging
+4. Rendering backend
+
+This separation enables one event product to feed multiple renderers and reduces regression risk.
+
+## 4) Migration strategy
+
+1. Make docs and interfaces honest about current backend status.
+2. Factor fire-event outputs into a renderer-neutral scene description or data adapter.
+3. Implement a custom-viewer renderer for that adapter.
+4. Only then consider changing the default fire viewer backend.
+
+## 5) Anti-patterns to avoid
+
+- Swapping rendering libraries before adapter/data boundaries are stable.
+- Duplicating geometry logic independently in multiple renderers.
+- Claiming migration is complete while `v.fire_plot()` still returns a Plotly-only interactive path.
+- Changing coordinate or axis conventions without updating invariants and tests together.
+
+## 6) Recommended next incremental PRs
+
+- Introduce a thin renderer-neutral adapter object for fire scene payloads (without changing current return keys).
+- Add tests that validate adapter payload shape and mapping from hull/summary outputs.
+- Implement an experimental custom-viewer renderer behind a non-default option.
+- Compare semantic parity (axes, time-depth direction, labels, camera defaults) before any default switch.

docs/recipes/fire_event_vase_hull.md

@@ -75,6 +75,13 @@ print("Cube:", getattr(cube, "da", None).shape if hasattr(cube, "da") else type(
     `fire_plot` requests daily gridMET/PRISM data for event windows. Explicitly pass
     `freq="D"` to emphasize daily sampling or a different frequency if needed.
 
+
+## Current viewer status
+
+`fire_plot` currently returns a fire-event analysis bundle plus a Plotly-based interactive hull figure. This is the supported fire-specific viewer path today. It is not yet rendered through the same custom HTML/CSS/JS viewer backend used by `v.plot()`. The long-term goal is to align fire-event visualization with the cube viewer’s shared space-time conventions so that fire-specific displays and generic cube displays live in the same visual grammar.
+
+Developer note: `If you are working on fire visualization internals, treat event construction, hull generation, climate/event fusion, and rendering as separable layers. The safest near-term improvements are the ones that preserve those boundaries.`
+
 ## What you get back
 `v.fire_plot` returns a dictionary with:
 

docs/viz/cube_viewer.md

@@ -4,6 +4,9 @@ The cube viewer turns a 3D climate cube into an interactive HTML scene. Call
 `v.plot()` on an `xarray.DataArray` with dims like `(time, y, x)` to render the
 cube inline in Jupyter or export standalone HTML.
 
+
+> **Transition note:** `v.plot()` is the canonical custom viewer for general cubes. Some specialized visualization workflows may still expose separate rendering backends during the transition period (including the current fire hull path in `v.fire_plot()`). The intended direction is convergence on shared spatial and temporal conventions, not necessarily immediate backend unification.
+
 The viewer is designed to:
 
 - show temporal slices along one axis and spatial slices on the other faces;

mkdocs.yml

@@ -53,6 +53,7 @@ nav:
       - Contributing: dev/contributing.md
       - CI and Testing: dev/ci_testing.md
       - Cube viewer invariants: dev/cube_viewer_invariants.md
+      - Fire plot architecture: dev/fire_plot_architecture.md
       - Figure backend (v.plot): dev/figure_backend.md
       - Viewer backend: dev/viewer_backend.md
       - Legacy reference: dev/legacy_reference.md

src/cubedynamics/verbs/fire.py

@@ -47,9 +47,11 @@ def fire_plot(
     allow_synthetic: bool = False,
     prefer_streaming: bool = True,
 ) -> Dict[str, Any]:
-    """Fire time-hull + climate visualization verb.
+    """Fire time-hull + climate visualization workflow.
 
     Supports cube-first (preferred) and legacy fire-first invocation styles.
+    Returns a fire analysis bundle and a Plotly hull figure (`fig_hull`) as the
+    current interactive backend for fire-specific visualization.
     """
 
     cube_first = da is not None