# Sundog Geometry Workbench Working hook: > It draws the parhelion from the math. Sundog Geometry is the proposed hero application — the page visitors land on when they reach `sundog.cc`. Unlike Balance and Three-Body, it is not a control task with a hidden target and an operating envelope. It is a parametric exercise of the parhelion description: a workbench that renders the optical phenomenon by exposing the geometry to its physical knobs and letting the visitor manipulate them. The workbench is itself the proof — that the description is concrete enough to render, and the rendering is faithful enough to recognise as the photographed phenomenon. The public question is small enough to defend: > Can the parhelion be reduced to a parametric model that renders a > recognisable halo display from a small set of physically meaningful > sliders, while staying interactive in the browser? The workbench should live as `sundog-workbench.html` until the static pose is locked, then promote to `index.html` as the hero. A short snapshot of the locked composition with subtle idle scintillation can serve as the public brand mark; the full interactive workbench remains accessible from the hero via a "tune the math" link. *[nice-to-have #14]* ## Related Documents This roadmap is the workbench's path-to-promotion plan. It is the human-tunable face of a broader **Sundog Generator** whose programmatic face and proof methodology are split into companion docs. Authority is distributed so no document trampolines another's decisions. - [`UI_UX_THEME_FOUNDATION.md`](UI_UX_THEME_FOUNDATION.md) — the site-wide visual frame. Owns: theme tokens (`--sd-*`), shared-stylesheet layer organisation, the cross-cutting **4a/4b/4c** split that places this workbench beside Threebody and Balance, and Migration Steps 1–6 that govern how page-local CSS interacts with the shared sheet. **Step 4a in that document defers the detailed roadmap here.** Any new theme token this workbench needs goes via `sundog-theme.css` and that document, not page-local CSS. - [`SUNDOG_GENERATOR_SPEC.md`](SUNDOG_GENERATOR_SPEC.md) — the programmatic generator architecture (Prompt → Parser → Pose JSON → Geometry Solver → SVG → optional AI). Defines the three modes (render from math / compare to sky / make beautiful) and the external tool schema. **The workbench is Mode 1's user-facing surface; that doc is the canonical statement of what the generator is and isn't.** It also owns the public-framing language we use when describing the project outside this repo. - [`SUNDOG_OVERLAY_PROTOCOL.md`](SUNDOG_OVERLAY_PROTOCOL.md) — the photo-overlay test methodology (Mode 2 procedure). Phase 2 calibration items below cite the overlay protocol as their gating procedure. - [`SUNDOG_V_THREEBODY.md`](SUNDOG_V_THREEBODY.md) and [`SUNDOG_V_BALANCE.md`](SUNDOG_V_BALANCE.md) — sibling application workbench roadmaps. Different shape (control task with operating envelope and pre-registered verdict), same Sundog pattern of *hidden / indirect / transformation / output*. The Pre-Committed Cross-Application Comparison Row near the bottom of this document is the parallel. - `public/poses/*.json` — the named-pose library. `canonical.json` snapshots the locked default; future poses (`low-altitude`, `cza-heavy`, `nine-halo-eye`) ship as Phase 8 lands. - BoxForge tools library: `C:\Users\hughe\Dev\Dungeon Gleaner Main\tools`, agent-accessible. `boxforge.html` and `orb-component.html` are reference primitives for the Phase 6 selective 3D handoff. The animation phase vocabulary (idle / hover / active / handoff·settle) used in Phases 4–6 below comes from that library's three-phase animation system, kept consistent so phase semantics carry across projects. ## Sundog Expression *(pre-staged for the eventual APPLICATIONS.md row)* Geometry does not map onto the indirect-signal pattern the way Balance and Three-Body do — there is no hidden state and no controller. The Expression block is therefore reframed as a description-and-rendering pair, kept in the same row format so cross-application comparison stays legible: - **Hidden description:** the angular geometry of a parhelion display — primary sun position, halo angles (22°, 46°), CZA visibility cutoff, parhelion azimuthal offset as a function of sun altitude, parhelic-circle curvature, ice-crystal dispersion width. - **Indirect signal:** the visible halo display on the sky as a 2D projection. - **Transformation:** parametric optical render — ice-crystal physics → angular geometry → SVG primitives + CSS gradient atmosphere → screen pixels. - **Actionable output:** an interactive workbench that draws the recognisable phenomenon from the slider state, with `Snapshot params` exporting a reproducible JSON pose. This block is here so the cross-application row in `APPLICATIONS.md` and the gallery card's identity triplet do not drift between the roadmap and the eventual writeup. When Phase 7 lands, both surfaces should quote this block verbatim. *[/nice-to-have #14]* ## Current State First executable scaffold landed 2026-05-08. The Sundog tree now includes: - `sundog-workbench.html`: a single-page workbench at the root of the public site (not yet promoted to `index.html` hero). Stage on the left, control rail on the right, mirrors the `threebody.html` idiom. - 16 parameter sliders organised by optical taxonomy: Sun & Pillar, Halos, Arcs, Parhelia (daggers), Composition (the 9-halo-eye fiction), Palette & Atmosphere. - SVG geometry primitives: - Compass-rose sun (4 rays, rotated 45° from diagonal so they read as cardinal directions; bright core scales with `--compass-ray-length` via sqrt-softening). - Sun pillar (vertical gradient stroke through the sun core). - Parhelic arc (quadratic Bezier path, curvature controlled by `--parhelic-curvature`; daggers ride the arc rather than a horizontal line — calibrated 2026-05-08 against source-photo overlay at `c = 0.66`). - 22° halo, 46° halo, two secondary halos for the virtual-sun fiction. - Two CZA arcs (primary + secondary) with prismatic gradient stroke and Gaussian-blurred edge to approximate ice-crystal orientation distribution. - Parhelia daggers (red-inside / blue-outside dispersion gradient, soft-blurred core ellipse + streak line). - CSS atmosphere overlay (radial gradients, `mix-blend-mode: screen`) parametrised by `--dispersion-width`. - Snapshot button that writes the current slider state to console + clipboard as JSON. Known calibration debt (2026-05-08 visual review): - `--parhelic-curvature = 0.66` is locked. - `--cza-bloom` slider was demoted to a baked constant (`stdDeviation = 0.3`, edge-soften only) on 2026-05-08 after A/B testing confirmed Gaussian blur on each arc independently does not produce the eyelid bell curve at any value — the two arcs do not meet within Gaussian-σ of each other at the apex, so their tails never sum to a peaked bell. The bell shape is now produced geometrically by a paint-fill band between the two arcs (`#cza-bell-fill`). A future `--cza-secondary-offset` knob is the right lever for tuning bell width if/when a slider is wanted again; tracked as a Phase 9/10 composition-fiction question, not a Phase 2 calibration item. - The compass-rose 45° rotation is in place but may overlap visually with pillar + parhelic, reducing the X reading. Phase 2 tune candidate. - Sun-altitude slider is wired to a CSS variable but is not yet bound to derived geometry — parhelion azimuthal offset and CZA visibility cutoff are still hard-coded. Phase 3 work. No animation has been wired. Idle scintillation amplitude is exposed as a slider, but only the 22° halo opacity oscillates on it. This section is here so the broadcast surfaces (gallery card, `APPLICATIONS.md`, `claims-and-scope.md`) do not get an entry until the runnable-artifact gates below are met. Until then, the hero workbench sits at **Scaffold** tier — sibling to the Balance Phase 7 smoke read, not yet a promoted public artifact. Promotion to **Calibrated Static Pose** tier requires Phase 0–2 gates below landing. Promotion to **Animated Hero** tier requires Phase 0–5 gates. Promotion to **Live Hero** (replace `index.html`) requires Phase 0–7 gates. ## Parallel Geometry Models (Legacy vs Halo Scaffold vs Halo Governed) We are explicitly avoiding sunk-cost lock-in on a feature-tuned geometry solver. The workbench therefore supports three geometry models that can be run in parallel long enough to compare under overlay. ### Legacy model *(feature-tuned)* This is the current baseline implementation: a small set of visible features are directly controlled and placed, with a few derived relationships. - Parhelic arc: symmetric quadratic Bezier driven by `--parhelic-curvature`. - Parhelia daggers: positioned by sampling that Bezier at fixed `x` anchors. - Sun pillar: a straight vertical line with a gradient stroke. This model is fast to iterate on, but it bakes in the exact failure mode described in the issue: emergent structure gets treated as independent, manually tuned landmarks. ### Halo Scaffold model *(free-ellipse-derived)* The first attempt at the issue's proposal. Treats a free ellipse — not a true halo — as a governing primitive, then derives secondary features from intersections. - Governing scaffold: a **free ellipse** whose bottom point is the sun (so the tangent at the sun is horizontal). Drawn as a sampled lower arc. - Parhelia daggers: derived as the intersection points between the 22° halo circle and that ellipse. Dagger streak direction is aligned to the ellipse tangent at the intersection point. - Compass-rose north/south behavior: rendered as a vesica between two equal-radius circles centered horizontally to either side of the sun. Their radii are tied to `--sun-pillar-length`, not to any halo. This is partway to the issue's intent — daggers do come out of an intersection — but the ellipse axes (`rx`, `ry`) and the pillar circle radii are still hand-fit constants, not halo angular radii or sun-altitude derivations. See *Outstanding gaps* below. ### Halo Governed model *(single-circle scaffold; issue-faithful)* A more committed reading of the issue: insist that the governing primitive be a **circle** (a halo) rather than a free ellipse, and tie the pillar geometry to the same halo system that determines the daggers. - Governing halo: a **circle** whose bottom point is the sun (one radius parameter, currently mapped from `--parhelic-curvature`). The parhelic arc is its visible lower-arc. - Parhelia daggers: closed-form intersection of the governing halo and the 22° halo. Single-step computation — no iterative root-finding. - Dagger streak direction: tangent to the 22° halo at the dagger point (perpendicular to the radial line from the sun). The daggers lie *on* the 22° halo, so this is the physically motivated tangent. - Compass-rose north/south behavior: vesica of two halos **centered at the daggers**, each with radius = (sun-to-dagger distance) + small slack controlled by `--sun-pillar-length`. The pillar geometry is therefore a consequence of the governing halo via the daggers — it has no independent pillar-radius constant. One knob (`--parhelic-curvature`) determines the parhelic arc, both dagger positions, and the pillar shape. Everything else is shared (CZA, secondary halos, compass rays, palette). This model lives in the workbench at the same code path as the others; it is intentionally *not* the default while it is unevaluated against the photo corpus. ### Outstanding gaps from issue #15 The issue raised several specific points. Recording the state of each so the next reviewer doesn't have to rediscover them: - **`~160°` halo/arc family** (issue's headline proposal): not directly modeled. The Halo Governed model's governing halo serves the same structural role (a single primitive that explains arc + daggers) but is parameterised by pixel radius, not by an angular value tied to atmospheric optics. Naming it "160°" in the workbench would be speculative without overlay calibration. - **Ellipses-instead-of-circles for the 22°/46° halos**: not attempted in any of the three models. The 22° and 46° halos remain `` in the SVG. - **Compass-rose center as a halo intersection**: only partially attempted. The Halo Governed pillar uses two real halos (centered at the daggers) but the compass-rose ray group itself is still a hand-placed `` of four ``s plus a core circle. The rays are not yet derived from halo tangencies. - **Overlay fit evidence**: not produced for any of the three models. Phase 2 calibration has not yet been run on Halo Scaffold or Halo Governed. The acceptance-criteria bullet "evidence that it better explains the observed relationships between major features" remains open. - **Halo Scaffold `rx`/`ry` blends**: still placeholder hand-fit constants. The "fewer ad hoc placements" framing is not yet earned for that model. ### How to run in parallel In `sundog-workbench.html`, use the **Geometry Model** selector: - `Legacy (feature-tuned)` — baseline behavior. - `Halo Scaffold (ellipse-derived)` — first-attempt scaffold. - `Halo Governed (single-circle scaffold)` — issue-faithful construction. The selection persists in `localStorage` and is included in `Snapshot params` as `geometryModel`. Reference poses for both alternative tracks live as `public/poses/canonical-halo-scaffold.json` and `public/poses/canonical-halo-governed.json`. Note that until Phase 8 lands the `Load params` button, those poses are documentation artifacts only — they cannot yet be loaded back into the workbench from the UI. ### Halo Atlas v2 fixes (2026-05-11) After overlay inspection three corrections landed: 1. **Parhelic arc direction flipped from frown to smile.** Was drawing the upper arc of a circle whose center sat below the daggers — apex above the sun. The Troels Nielsen photograph's parhelic feature smiles (apex below sun). Now draws the lower arc of a circle whose center sits above the daggers. Applies to both `halo_governed` and `halo_atlas`. 2. **CZA anchored to 46° halo top.** Real atmospheric optics: the CZA is tangent to the 46° halo at its top. Apex now sits at `(sun.x, sun.y − R_46)` by construction; `--cza-curvature` offsets from that anchor. Calibration residual against the Troels Nielsen photo: was (2, 15) px, now **(2, 1) px**. Default 0.85 produces the canonical tangent pose. 3. **Upper tangent arc** is a new atlas primitive. Tangent to the **22°** halo at its top — the "eyelid above the 22° halo" feature distinct from the CZA. Slider `--upper-tangent-intensity` (default 0). Without this primitive, the only "eyelid"-shaped feature in the atlas was the CZA, which sits much higher (tangent to 46° halo) and therefore can't serve double duty. R_46 note: Gemini suggested using the pure 46°/22° angular ratio, `R_46 = R_22 × 46/22 = 460` workbench units (→ 303 photo px). Calibration against the Troels Nielsen photo measures `R_46 = 285 photo px = 432 workbench units`. The atmospheric-optics literature reports the same ~5% gap (refraction in the 46° prism path is more sensitive to crystal orientation than the 22° path). The workbench's `R_46 = 440` gives a 1.8% photo-fit residual; switching to 460 would *worsen* the fit to 6.3%. Holding `R_46 = 440`. Overlay PNG: [`atlas_v2_overlay_troels_nielsen.png`](atlas_v2_overlay_troels_nielsen.png). ### Halo Atlas calibration vs Troels Nielsen DR (2026-05-10) `halo_atlas` extends the Halo Governed v2 reading with three changes the photo overlay said were needed: 1. **Sun-altitude binding** — daggers at `(SUN.x ± R_22 / cos(h), SUN.y)`. At the canonical pose `--sun-altitude = 25°` (the value implied by the photo's parhelion offset), daggers move to offset 240, matching the measured photographed parhelia within 0–3 photo px. 2. **CZA as a true full-ring upper arc** — replaces the broken Bezier-with-buggy-default `applyCza` for atlas mode. The default `--cza-curvature = 0.85` now produces apex y = 81 (canonical), not the collapsed flat line at y = 240. Bell-fill is the band between primary and secondary upper arcs, both as real circles. 3. **Supralateral arc** — new primitive, tangent to the 46° halo at its top, curving up. Optional via `--supralateral-intensity` (default 0). Numerical residuals at the canonical atlas pose (sun-altitude 25°, parhelic-curvature 0.05, cza-curvature 0.85, supralateral-intensity 0.40): | feature | atlas prediction | observed | residual | | --- | --- | --- | --- | | 22° halo radius | 145 (anchor) | 145 | 0 | | 46° halo radius | 290 | 285 | 1.8% | | left dagger x | 240 | 243 | 3 px (was 12 px in v2) | | right dagger x | 560 | 560 | 0 px (was 15 px in v2) | | CZA primary apex | (400, 80) | (402, 65) | (2, 15) px | | parhelic arc | near-horizontal through daggers and sun | photo's parhelic feature is near-horizontal | ✓ | | supralateral arc | upper-canvas arc tangent to 46° halo top | photo has visible bright structure in this region | qualitative ✓ | Overlay PNG: [`atlas_overlay_troels_nielsen.png`](atlas_overlay_troels_nielsen.png). The daggers and parhelic arc residuals close out the two correctable defects v2 left open. CZA y-residual remains ~15 px — fixable by tying `--cza-curvature` default to sun altitude instead of holding it at 0.85. ### Halo Governed v2 calibration vs Troels Nielsen DR (2026-05-09) A first overlay calibration was run on the canonical Troels Nielsen DR reference. Anchor: the 22° halo (sun at pixel `(400, 356)` in the 800×450 photo, observed halo radius `145 px`, so `1 workbench unit = 0.659 photo px`). Overlay PNG: [`v2_overlay_troels_nielsen.png`](v2_overlay_troels_nielsen.png). Numerical residuals: | feature | v2 prediction | observed | residual | | --- | --- | --- | --- | | 22° halo radius | 145 px (anchor) | 145 px | 0 (anchor) | | 46° halo radius | 290 px | 285 px | 1.8% over | | left dagger x | 255 | 243 | 12 px (parhelion **outside** the 22° halo) | | right dagger x | 545 | 560 | 15 px (same) | | CZA apex | (400, 79) | (402, 65) | (2, 14) px | | parhelic arc apex (at default `--parhelic-curvature = 0.66`) | (400, 269) | photo's parhelic circle is essentially horizontal through the parhelia | curvature default too high for this photo | Two calibration findings worth fixing before the workbench can claim a photo-faithful pose: 1. **Daggers are 8% too close to the sun.** The photo has parhelia at distance `≈ R_22 / cos(h)` from the sun, where `h` is the sun's altitude. Solving from the observed offset gives an implied sun altitude of `≈ 25°`. v2 currently fixes daggers at exactly the 22° halo radius (the sun-on-horizon limit). The fix is the Phase 3 binding from `--sun-altitude` to dagger placement: `dagger_x = SUN.x ± R_22 / cos(h)`. 2. **Default parhelic-arc curvature is too high.** At sun altitudes where parhelia are well-formed, the parhelic circle reads as nearly horizontal through the parhelia in the photograph — there is no significant upward arc above the sun in this Troels Nielsen scene. The default `--parhelic-curvature = 0.66` should drop closer to 0 for this pose. What v2 *did* get right: 22° halo (by anchor), 46° halo (1.8%), CZA apex horizontally (2 px), pillar position (centered on sun). Those four are load-bearing; the two findings above are correctable knob defaults plus a Phase 3 sun-altitude binding, not a structural failure of the construction. ### Recommendation No model should become the default until Phase 2 overlay calibration has been run on the canonical photo corpus for at least Legacy vs Halo Governed, with the snapshot of each pose archived for review. Specifically: - The two-vs-three-model decision rests on overlay fit, not on the elegance of the construction in prose. Halo Governed reads cleaner on paper, but prose elegance is not evidence. - Halo Scaffold should not be promoted in its current state — its `rx`/`ry` blends and pillar circle radii are placeholder constants that have to be resolved before any "fewer ad hoc placements" claim can survive review. - If Halo Governed reaches comparable or better overlay fit and the open gaps above are closed (or explicitly accepted), it is the candidate default. Until then it sits beside the other two as a comparison track. - Whichever model ships, the loser(s) should be retained as a regression baseline through at least the Phase 7 hero promotion. ## Why This Workbench The hero of `sundog.cc` carries two loads at once: it is the brand mark, and it is the entry point that has to communicate what the project is in under a second. A static splash can do the brand mark. It cannot do the second load without lying — most static optical hero images either oversell ("we made the sky") or undersell ("this is decoration, scroll past it"). A parametric workbench resolves that tension. Visitors can manipulate the optical knobs and see the rendering respond. The brand mark is visible at rest. The proof is visible on interaction. The two loads collapse into one artifact. The choice is also project-coherent: Balance and Three-Body are both interactive workbenches with public canvases. Routing the hero through the same idiom keeps the site reading as a single voice. A visitor who lands on the geometry workbench, scrolls past, and clicks through to the three-body or balance pages should feel they are using one tool, not three. ## Visual Direction The page should feel like an observation, not a marketing splash. The first screen opens directly into the rendering, with a compact stage copy overlay in the upper-left. Target composition (the "sundog eye"): - foreground: compass-rose sun (pupil), 22° halo (iris), CZA arcs (eyelids), parhelia daggers along the parhelic arc; - background: cool dark sky (`#1a3850` → `#0b1720` → `#050a10` radial), evoking a clear high-altitude observation; - right rail: optical parameter sliders grouped by taxonomy (Sun & Pillar / Halos / Arcs / Parhelia / Composition / Palette); - atmosphere overlay: subtle CSS gradients at warm-amber (sun side) and cool-blue (zenith side), `mix-blend-mode: screen`, parametrised by `--dispersion-width`; - optional 9-halo-eye fiction: when `--secondary-suns-strength > 0`, two virtual suns appear at the upper-left and upper-right with their own halo systems; intended for the active animation phase. Reference photographs (calibration corpus): - Troels Nielsen DR — single-sun parhelion with sundog daggers and CZA eyelid (the canonical pose). - Triple-sun event with stacked CZA arcs forming the iris-layer reading (the 9-halo-eye source). Both references are pinned in the hero design memory (`project_sundog_4a_vision.md`); the workbench's static pose should be recognisable as a stylised composition of these. Implementation preference: - SVG for crisp arcs and rays — no canvas, no WebGL. - CSS gradients + filters for atmospheric softness; `mix-blend-mode: screen` for additive blending on the CZA arc overlap. - Parameters bind via CSS custom properties wherever the SVG attribute accepts them (opacity, stroke-width, transforms). - JS update loop only for derived geometry (parhelic-arc Bezier, dagger positions on the arc, sun-core radius scaling). - BoxForge primitives reserved for the eventual Phase 6 selective 3D handoff (compass-rose ray rotation, possibly), not for the static composition. ## Actionability Audit The workbench must separate parameter tiers from day one. | Slider | Tier | Use | | --- | --- | --- | | `--sun-altitude` | Math-derived (planned) | Drives parhelion azimuthal offset and CZA visibility cutoff in Phase 3 | | `--halo-22-intensity` | Free | Visual knob; opacity of the 22° halo ring | | `--halo-46-intensity` | Free | Visual knob; opacity of the 46° halo ring | | `--cza-intensity` | Free | Visual knob; opacity of both CZA arcs | | `--cza-curvature` | Free | Visual knob; apex of the primary CZA arc | | `--sun-pillar-intensity` | Free | Visual knob; opacity of the vertical pillar | | `--sun-pillar-length` | Free | Visual knob; pillar extent above and below sun | | `--parhelic-circle-intensity` | Free | Visual knob; opacity of the parhelic arc | | `--parhelic-curvature` | Math-derived (planned) | Locked at 0.66 by overlay calibration; Phase 3 will derive from sun altitude | | `--parhelia-intensity` | Free | Visual knob; opacity of the dagger group | | `--parhelia-dagger-length` | Free | Visual knob; horizontal extent of the dagger streaks | | `--dispersion-width` | Free | Visual knob; atmosphere overlay intensity | | `--secondary-suns-strength` | Composition fiction | Controls the 9-halo-eye effect (not standard atmospheric optics) | | `--ring-overlap-bias` | Composition fiction | Golden-ratio knob for the 9-ring intersection | | `--compass-ray-length` | Free | Visual knob; ray length and core radius scale together | | `--rainbow-saturation` | Free | Visual knob; reserved for Phase 2 prismatic gradient tuning | | `--idle-scintillation-amplitude` | Animation | Phase-4 idle drift amplitude; off when 0 | The central typing should be explicit: ```text M = parhelion description (the math) P_t = parameter state at time t (slider values) G(P_t) = derived geometry (positions, curvatures, intensities) R(G) = rendered SVG composition ``` Phase 3 binds `M`-derived params (sun-altitude → CZA visibility, parhelion offset, parhelic curvature) so the workbench cannot render a physically-impossible pose by accident. Free params remain free knobs; the fiction params (composition family) are deliberately outside `M` and visibly labelled in the Composition section of the right rail. The claim is not that the workbench discovers new optics. The claim is that the parhelion description is concrete enough to render parametrically, and the rendering is faithful enough that overlaying it against a reference photograph reads as the same phenomenon. ## Ratified Hook Language Safe hook: > Sundog Geometry renders a parhelion display from a small set of physically > meaningful sliders. The math draws the eye. Short version: > It is the proof, watching you watch it. Long brand-leaning version: > Three suns. Nine halos. One description. Move the sliders and watch the > sky bend to fit the equation. ## Roadmap ### Phase 0 - Parameter Taxonomy and Description Lock Goal: pin down which sliders are free visual knobs, which are math-derived from the parhelion description, and which are composition fiction outside the description. Document the binding equations for the math-derived set so Phase 3 has unambiguous targets. Deliverables: - `docs/SUNDOG_V_GEOMETRY.md` (this document) with the Actionability Audit table populated and the math-binding TODO list explicit. - A short notes file with the parhelion description equations the workbench will eventually exercise (sun-altitude → parhelion offset, CZA visibility window, parhelic-circle curvature derivation). Gate: the Actionability Audit table identifies every slider's tier and the math-binding TODOs are enumerated. **Status: in progress with this document.** ### Phase 1 - Static Composition Scaffold Goal: a runnable workbench that renders the canonical "sundog eye" pose at default slider values, with all parameters interactive. Deliverables: - `sundog-workbench.html` with full slider rail and SVG stage. - All 16 parameters wired to CSS custom properties + JS update loop. - Reset and Snapshot controls. - Site chrome (header, footer) shared from `sundog-theme.css`. Gate: open the page, every slider produces a visible response. **Status: landed 2026-05-08.** ### Phase 2 - Calibration Pass (Static) Goal: lock visual defaults against the reference photograph corpus using the translucent-overlay calibration technique. The procedure is canonical in [`SUNDOG_OVERLAY_PROTOCOL.md`](SUNDOG_OVERLAY_PROTOCOL.md); this phase runs that protocol manually for each open calibration item until each threshold passes against `troels-nielsen-dr.jpg`. Calibration items (live list): - [x] `--parhelic-curvature = 0.66` (calibrated 2026-05-08). - [x] `--cza-bloom` promoted to slider (2026-05-08); default lowered from the over-aggressive `3.2` to `1.4` pending overlay calibration. - [x] `--cza-bloom` demoted to baked constant `stdDeviation = 0.3` (2026-05-08, ass-dyno value). Mechanism mismatch confirmed: Gaussian blur on each arc independently does not sum to a bell curve at the eyelid apex regardless of σ. Replaced with `#cza-bell-fill` paint band between the two arcs. Slider count drops 17 → 16. The eyelid-bell goal is reclassified from a Phase 2 calibration item to a Phase 9/10 composition-fiction question whose right lever is `--cza-secondary-offset` (geometric overlap), not blur. - [ ] Compass-rose rotation angle — currently 45° from original diagonal; test 30°, 22.5°, and back-out-of-rotate to find the pose where the X reads as distinct from pillar + parhelic. - [ ] Default values for `--halo-22-intensity`, `--halo-46-intensity`, `--cza-intensity`, `--parhelia-intensity` against the source photo. - [ ] Default for `--dispersion-width` (atmosphere bloom level). - [ ] Default for `--compass-ray-length` (X visibility vs core dominance). - [ ] Decision: promote 9-halo-eye composition (`--secondary-suns-strength`) default — `0.0` (off) for the canonical pose, or a small non-zero default if the fiction is part of the locked hero. Gate: snapshot the locked param JSON to `docs/SUNDOG_GEOMETRY_POSE.json` or similar; future regressions are reviewable against that file. ### Phase 3 - Math Binding Goal: bind sun-altitude (and any other math-derived sliders) to the parhelion description equations so the workbench cannot render a physically-impossible pose. Deliverables: - A small JS module (`public/js/parhelion-geometry.mjs`) implementing the binding equations: `parhelionOffset(altitude)`, `czaVisible(altitude)`, `parhelicCurvature(altitude)`. - `applyDerivedGeometry()` updates to call the module instead of using hard-coded positions. - A test harness page or unit-test script that pins the equations to known reference altitudes. - Documentation block in this roadmap referencing the parhelion description paper / notes the equations are drawn from. Gate: setting the sun-altitude slider produces visibly correct parhelion-offset shift; setting altitude > 32° hides the CZA arcs automatically. ### Phase 4 - Idle Scintillation Goal: phase-4 of the BoxForge animation discipline — the idle window where the static pose holds. Subtle motion only. Deliverables: - Slow opacity drift on the rainbow gradient stops (saturation pulse, 3–8 second period, low amplitude). - Optional micro-rotation of the compass-rose ray group (≤ 1° peak). - `prefers-reduced-motion` honored — scintillation snaps to off. - All scintillation gated by `--idle-scintillation-amplitude`; at 0 the pose is fully static. Gate: load the page at default amplitude, the pose looks alive but no single element distracts the eye for more than ~2 seconds. ### Phase 5 - Active Reveal Animation (Phases 1-3) Goal: the sequenced reveal that narrates the parhelion construction. This is the BoxForge phase 1-3 of the animation model. Phase 1: empty sky, primary sun fades in (pillar first, then core, then X). Phase 2: 22° halo draws (stroke-dasharray reveal, 1.5s). Phase 3: CZA arcs draw + parhelia daggers fade in. Optional sub-phase: when `--secondary-suns-strength > 0`, virtual peripheral suns and their halos materialize as a Phase 3.5. Deliverables: - Animation orchestrator JS (`public/js/parhelion-phases.mjs`). - Phase trigger UI (replay button, optional auto-trigger on first visit). - `prefers-reduced-motion` collapses to instant snap to phase-4 idle pose. Gate: replay button cycles through phases 1-3 → idle phase-4; reduced-motion visitors see the locked pose only. ### Phase 6 - Selective 3D Handoff (Phase 5+) Goal: the BoxForge phase 5+ of the animation model — only elements that benefit from depth get 3D motion. Halos stay planar (they are sky, not boxes). Candidates for 3D motion: - Compass-rose ray group rotation (slow, hover-triggered or auto). - Possibly the central core treated as a tiny BoxForge orb on hover only. Explicit non-candidates (stay planar): - All halos and arcs (no 3D rotation). - Parhelia daggers (no depth). Deliverables: - BoxForge primitives wired as overlays (orb-component or custom). - Hover and active-state triggers respecting reduced-motion. - Performance budget: target ≥ 50 fps on a mid-tier laptop with the active reveal + idle scintillation + hover motion all running. Gate: hover the sun, the rays rotate; halos do not. Reduced-motion visitors see no rotation. ### Phase 7 - Hero Promotion Goal: replace `index.html` hero with the workbench (or a distilled non-interactive snapshot of it with a "tune the math" link to the full workbench, depending on what reads better in the wild). Deliverables: - New `index.html` hero that either embeds `sundog-workbench.html` content inline OR renders a distilled static pose with a CTA to the workbench. - Decision recorded in this document and in `claims-and-scope.md`. - Old hero canvas (`#parhelion-canvas`) decommissioned. - A short snapshot `.gif` or `.webm` for press / social use, generated from the locked phase-4 idle pose with scintillation. Gate: a first-time visitor lands on the new hero, gets the brand reading + the parametric proof signal, and can reach the deeper workbench (if not already on it) in one click. ### Phase 8 - Reproducible Pose Export Goal: the workbench's `Snapshot params` button currently writes JSON to console + clipboard. Promote that to a deterministic export pipeline matching the schema in [`SUNDOG_GENERATOR_SPEC.md`](SUNDOG_GENERATOR_SPEC.md): identical pose JSON loaded back into the workbench produces a byte-equivalent SVG. Deliverables: - `Load params` button accepting JSON (paste or file picker), reading the canonical schema from the Generator Spec. - Pose pinning script that, given a JSON file, renders the workbench at that pose and exports a PNG/SVG snapshot for archival. - Named-pose library expansion under `public/poses/`: - [x] `canonical.json` — snapshot of the current locked defaults (landed alongside the Generator Spec). - [ ] `low-altitude.json` — sun near the horizon, parhelia close to halo; CZA visible. - [ ] `cza-heavy.json` — sun altitude well below 32°, CZA at maximum intensity, halo and parhelia subdued. - [ ] `nine-halo-eye.json` — full composition fiction with `--secondary-suns-strength = 0.7` and tuned overlap bias. - [ ] `forty-six-halo.json` — 46° halo dominant, 22° subdued. Gate: round-trip a snapshot → load → render → snapshot and confirm byte-equal JSON across two browser sessions. ## Post-Verdict / Conditional Roadmap These phases unlock only after Phase 7 promotion lands and the hero is in the wild for some adoption period. ### Phase 9 - Educational Mode Goal: a toggle that overlays annotations on the workbench — angle measures, crystal-orientation arrows, refraction-angle labels — turning the brand hero into a teaching artifact. ### Phase 10 - Atmospheric Variations Goal: alternative sky palettes (dawn, dusk, twilight, polar) with calibrated photograph references for each. Optional `--sky-tint` becomes a real palette slider rather than a single-axis warm/cool knob. ### Phase 11 - Linked Description Mode Goal: clicking any slider opens a small popover with the equation it represents, the source citation, and a tiny worked example. Couples the workbench tightly to the eventual parhelion-description paper. ## Claim Boundary The Geometry workbench claims: - A parametric rendering of known parhelion physics that is faithful enough to overlay against reference photographs and read as the same phenomenon. - The "sundog eye" composition (pupil = sun, iris = halos, eyelids = CZA arcs, optional 9-halo virtual-sun extension) is an aesthetic identification, deliberately stylised, with no claim that real atmospheric optics produces a 9-halo eye on demand. - Interactive proof that the parhelion description is concrete: every slider corresponds to either a math-derived geometric quantity, a free visual knob clearly labelled as such, or a composition-fiction knob clearly labelled as such. The Geometry workbench does NOT claim: - Novel atmospheric optics findings. The physics is well-known; the contribution is the parametric-rendering format and the composition-as-brand framing. - That the parhelion is "discovered" by the workbench. The phenomenon has been observed and described for centuries; Sundog's contribution is the description style and the workbench-as-proof presentation. - That the workbench renders all halo phenomena. It renders the parhelion + CZA + parhelic-arc subset chosen for the hero composition. Other halo phenomena (Parry arcs, infralateral arcs, sub-suns, light pillars specific to plate-crystal alignments) are out of scope unless added in Phase 10. ## Pre-Committed Cross-Application Comparison Row For the eventual `APPLICATIONS.md` table: | Application | Hidden | Indirect | Transformation | Output | | --- | --- | --- | --- | --- | | Geometry | parhelion description | sky display | parametric optical render | interactive workbench + reproducible pose | | Three-Body | true 18D state | virial / pairwise / inertia signatures | SCAN/SEEK/TRACK | bounded controller for selected regimes | | Balance | upright pole angle | shadow projection on floor | shadow-residual control | bounded cart force command | Geometry occupies a different cell shape than the other two — it is a description-and-render pair rather than a hidden-state-and-controller pair — and the table headers should accommodate that. The "Hidden" column for Geometry reads as the description-being-exercised rather than a withheld truth. ## Suggested First Build Slice The first implementation pass (largely already complete) was deliberately small: 1. SVG stage with the canonical pose hard-coded. 2. Slider rail with all 16 parameters wired to CSS custom properties. 3. Derived-geometry JS for the parhelic-arc Bezier and dagger positioning. 4. Snapshot and reset controls. That slice is enough to decide whether the workbench-as-hero idiom carries the brand load. The locked-pose snapshot from Phase 2 calibration becomes the canonical reference for everything downstream. *[nice-to-have #15]* **Exit criterion for the first slice:** the canonical pose, loaded from a snapshot JSON, reads as the photographed phenomenon when overlaid at 50% opacity against the Troels Nielsen DR reference photograph (vertical pillar matches; 22° halo radius matches; parhelia positions match within ±10 SVG units; CZA apex within the eyelid envelope). Below this floor — the overlay reads as a different phenomenon — re-scope before any Phase 4 animation work. *[/nice-to-have #15]*