# Sundog vs. the Limits of Perception Working hook: > The atlas describes the visible parhelion. The instrument predicts the > parhelion features that exist below the human contrast threshold and > confirms them with hardware. This roadmap promotes the cross-pollination memo at the top of this file's prior life into a structured roadmap. It is the geometry-side analogue of [`SUNDOG_V_MESA.md`](SUNDOG_V_MESA.md) Phase 6: where mesa went from "describing the cliff behaviorally" (Phase 5) to "localizing the mechanism causally" (Phase 6), geometry goes from "overlaying known arcs" (Phase 7 hero) to "predicting where unseen features must lie and detecting them in the real atmosphere." The transition is structural, not aspirational. The atlas already parametrises every visible arc as the upper portion of a complete implied circle derived from sun altitude and crystal orientation. The full geometry of each circle continues below the contrast floor of unaided photography. The instrument's job is to integrate signal along the predicted continuation at the angular positions the atlas already names, exceeding what the eye and naked-sensor camera resolve. ## Why Perception Is the Right Next Roadmap The gravity claim in [`SUNDOG_V_GRAVITY.md`](SUNDOG_V_GRAVITY.md) now carries two independent in-substrate receipts for the field-not-reward framing: mesa's in-vitro entangled 5D `net.7` subspace, and geometry's in-the-wild halo atlas. Both converged on the same shape — small handful of generators, irreducibly entangled, only legible as a whole — by independent methods. The two-substrate convergence is currently a structural argument. Either substrate could be retorted as "you fit a model that happens to look like that." The retort closes when geometry earns a **predictive receipt**: the atlas names a feature that no one had measured, and a measurement confirms it where the math said it would be. That move turns the workbench into an instrument. The atlas already inverts photographs to recover sun altitude with median 1 photo-px residual across seven photos; that is a strong inverse-inference demonstration. The next ratchet is to predict a feature *the photograph does not show* and confirm it with instrumentation that integrates signal along the predicted geometry. Methodologically, perception inherits mesa's discipline wholesale: pre-registered predictions, quantitative smoke gates, named null conditions, and clean negatives as publishable deliverables. The [`MESA_CROSSOVER_NOTE.md`](MESA_CROSSOVER_NOTE.md) already lifted the five transferable findings into Phase 10/11 of geometry; perception extends the same discipline one substrate further. ## Cross-Pollination Provenance This roadmap originated as a cross-pollination memo from the mesa work into the geometry program, written 2026-05-13 alongside [`MESA_CROSSOVER_NOTE.md`](MESA_CROSSOVER_NOTE.md). The memo argued that the atlas's "what it does NOT yet model" list (halo dispersion physics, subhorizon arcs, crystal-orientation mixing, crystal-size effects) is not an apology surface but a **prediction surface** — each item is a falsifiable prediction the 7-formula field makes when extended along its own generators. > **Galvanized from the Sundog Atlas project (2026-06-07) — the prediction > surface is now bounded on one side.** The [Sundog Atlas](SUNDOG_V_ATLAS.md) > ran the directed search for *higher-catastrophe* invisible halos to completion > ([`atlas/ATLAS_PHASE8_STRATA.md`](atlas/ATLAS_PHASE8_STRATA.md): the full 2-DOF > single-crystal sweep — column 60°/90°, Wegener, Lowitz, pyramidal). It came > back **empty: no A₄ swallowtail, no D₄ umbilic anywhere**, confirming Berry > 1994 (the one higher catastrophe found is the Lowitz A₃-lips — an A₃, not a new > arc; two candidate-A₄ false alarms were refuted by code). **So the > "invisible-halo" part of this prediction surface is NOT a new rare *arc* from a > higher caustic** — that direction is closed. The genuine candidates are the > **sub-visible / diffraction-dressing / sub-threshold-continuation** features of > the *known* halos (exactly the dispersion / orientation-mixing / crystal-size / > sub-visible-continuation items this roadmap instruments). The atlas's classified > phase diagram ([`atlas/ATLAS_PHASE7_PHASE_DIAGRAM.md`](atlas/ATLAS_PHASE7_PHASE_DIAGRAM.md)) > is the forward map of *which* known halos are present at *which* elevations — > the structural backdrop the instrument extends along. This roadmap commits to the reframe: the atlas predicts; the instrument confirms or falsifies. Pre-registration discipline is mandatory because without it, "we extended the atlas and it matched" is indistinguishable from confirmation bias. ## Claim Boundary The perception roadmap claims, if Phase 4 lands positive: - In the tested high-cirrus conditions and apparatus tier, the parhelic circle's polarization signature is detectable along the atlas-predicted continuation at angular positions and amplitudes consistent with standard ice-crystal refraction physics, below the contrast threshold of unaided photographic observation. - The atlas therefore functions as a parametric measurement instrument for sub-visible parhelion-display features that exist deterministically as consequences of the field. The perception roadmap does **not** claim: - "The atlas predicts every atmospheric phenomenon." - "Sundog instrumentation can replace HaloSim or full ray-trace simulators." - "Sub-visible halo phenomena were previously unknown to atmospheric optics." (They are not; standard references treat them as known but rarely instrumented in amateur or workbench contexts.) - "A single positive prediction proves the gravity claim." A predictive receipt is one rung on the two-substrate convergence ladder, not the ladder itself. ## Avoid - "Sundog discovered a new atmospheric feature." (Most predicted features are textbook physics; the novelty is the predictive-overlay workflow, not the phenomena.) - "The atlas is a spectrometer." (It is a parametric geometry model that *aims* a spectrometer.) - "The gravity claim is now confirmed in the wild." (A positive Phase 4 ratchets the two-substrate convergence one step; it does not retroactively prove the structural argument.) ## Pre-Registration Template Every candidate prediction in this roadmap, before any apparatus is pointed at the sky, ships with a written spec naming: 1. **Angular position(s)** the atlas predicts, in workbench coordinates and in real-sky coordinates relative to the sun. 2. **Detection regime** — polarization angle, narrowband intensity, long-exposure integration, spectroscopic dispersion, or some named combination. 3. **Predicted signal curve** — the explicit function (polarization angle vs. arc position, intensity vs. wavelength, etc.) the atlas math implies, with the source citation for the physics that connects geometry to signal. 4. **Smoke gate** — the signal threshold below which the apparatus is judged insufficient rather than the prediction falsified, with the noise-floor measurement protocol that defines "insufficient." 5. **Null condition** — what observation would actually count *against* the prediction, distinct from "we couldn't tell." 6. **Positive condition** — what observation would count *for* the prediction, with the discrimination test against confounders (e.g., scattered sky polarization at the observation angle). This template is mandatory. A prediction that cannot fill all six slots does not get an apparatus run. ## Roadmap ### Phase 0 — Scope Lock, Literature Spine, Actionable/Reach Sort Goal: pin the apparatus tier, the candidate prediction slate, the pre-registration template, and the actionable-vs-reach classification before writing any new code or buying any new hardware. Deliverables: - **Apparatus tier locked at Smartphone-Only for Phase 1.** A rotatable linear polarizer film, a tripod, a smartphone with manual exposure control, and the existing pre-rendered atlas overlay matched to camera FOV. Total apparatus cost target: under $50 of polarizer film plus whatever tripod and phone are already on hand. Higher-tier apparatus (narrowband filters, long-exposure astro cameras, fiber-coupled spectrometers) is explicitly deferred to Phase 5 / Phase 6 promotions. - **Candidate prediction slate** — three predictions, ranked by apparatus tier: - **Prediction 1 (Phase 1–4): Sub-visible parhelic circle continuation.** Polarization angle along the great-circle continuation of the parhelic arc beyond the photometric contrast drop-off. Smartphone-tier apparatus. - **Prediction 2 (Phase 5+, conditional): Intersection enhancement.** Integrated brightness at predicted circle-circle crossings should exceed the sum of each circle's local brightness by a factor predictable from prism-path overlap. Requires narrowband filtering and long-exposure capable camera. - **Prediction 3 (Phase 6+, conditional, reach): Color-order inversion at arc tangencies.** Chromatic dispersion gradient cancellation/reinforcement at precise angular positions where two arcs' color orderings meet. Requires a small calibrated spectrometer. - **Actionable/reach sort,** written as the table below: | Item | Tier | Phase | Status | |---|---|---|---| | Methodology transfer from mesa (pre-registration, smoke gates) | Doc-only | 0 | Actionable | | Computing predicted polarization curve along parhelic continuation | Software | 1 | Actionable | | Smartphone + polarizer + tripod + overlay apparatus | Hardware ($30–50) | 1 | Actionable | | First observation run, cirrus afternoon | Field | 3 | Actionable, weather-gated | | Result note in mesa-style discipline | Doc-only | 4 | Actionable | | Narrowband filter + long-exposure camera apparatus | Hardware ($200–400) | 5 | Reach (Phase 4 promotes) | | Intersection-enhancement prediction & smoke gate | Software + Field | 5 | Reach | | Calibrated spectrometer apparatus | Hardware ($500–2000+) | 6 | Reach (Phase 5 promotes) | | Color-order inversion prediction & smoke gate | Software + Field | 6 | Reach | | "Atlas is a measurement instrument" public language | Doc + Public | 7 | Earned-language, gated by Phase 4 positive | | Polarimetric imager / filter wheel apparatus | Hardware ($2000+) | — | Reach, no committed phase | | Subhorizon arc instrumentation (requires observer above horizon) | Field + apparatus | — | Reach, no committed phase | - **Literature spine** — short, structured, not a survey: - Greenler, *Rainbows, Halos, and Glories* (1980): canonical atlas geometry and ice-crystal refraction tables. - Tape, *Atmospheric Halos* (1994): exhaustive halo taxonomy and Parry-orientation chapter. - Cowley, *Atmospheric Optics* (online): canonical photographic references and modern halo nomenclature. - Können, *Polarized Light in Nature* (1985): ice-crystal polarization signatures and the standard derivation for parhelic-circle polarization geometry. - Lynch & Livingston, *Color and Light in Nature* (2nd ed., 2001): chromatic dispersion in halos, color-order conventions, and the standard treatment of arc tangencies. - Standard sky-polarization references (Coulson; Horváth et al.) for distinguishing crystal-refraction polarization from Rayleigh background. - HaloSim documentation (Cowley/Schroeder) as the comparison ceiling for any quantitative prediction. - **Pre-registration template** (the six-slot spec above) ratified and committed to the doc. Exit criterion: the apparatus tier, prediction slate, sort, literature spine, and pre-registration template are pinned and tracked in this file. No Phase 1 work begins until exit criterion is met. ### Phase 1 — Apparatus Build and Predicted-Curve Software Goal: build the smartphone-tier apparatus and the software that computes the predicted polarization curve along the parhelic-circle continuation for any (sun-altitude, observer-azimuth, camera-FOV) tuple, so that Phase 2's pre-registration can name the exact angular positions and predicted signal values. Deliverables: - **Apparatus assembly**: - Rotatable linear polarizer film mount that accepts a standard smartphone lens. Marked with a fiducial scale so polarizer rotation angle is recorded with each exposure. - Tripod with phone clamp and altitude/azimuth read-off (a compass app plus a clinometer suffices; no theodolite required). - Calibration target: a known partial polarizer or a polarization reference card photographed under known conditions, used to verify the polarizer's measured-angle-vs-rotation curve. - **Predicted-curve software** (`scripts/perception-predict-parhelic.mjs` or equivalent under `public/js/` if browser-renderable): - Inputs: sun altitude `h`, observer azimuth (relative to the sun), camera FOV, camera optical axis. - Outputs: the full great-circle path of the parhelic circle in the camera's pixel frame, the predicted polarization angle along that path at sample spacing `Δθ` (default 1°), and the predicted polarization degree using standard ice-crystal refraction coefficients from Können. - Sanity test: at the canonical Troels Nielsen pose (`h ≈ 25°`), the predicted polarization angle at the visible parhelion locations must match the standard literature value to within 1° (this is a sanity check against the implementation, not a novel prediction). - **Atlas overlay extension** (`scripts/atlas-overlay-camera-fov.mjs` or inline in the workbench): - Render the predicted parhelic continuation as a thin curve on the workbench output, distinguishable from the visible parhelic arc (e.g., dashed, half-opacity, separately colored). - Output an FOV-matched PNG that can be loaded as a reference overlay against camera frames. - **Scripted exposure series** — a short shell script or note in `docs/perception/APPARATUS_PROTOCOL.md` describing the exposure sequence: bracket the suspected continuation segment with N polarizer-rotation steps at fixed angular spacing, fixed shutter, fixed ISO, fixed focal length, against the FOV-matched overlay. Exit criterion: the apparatus is buildable from the protocol doc by someone who has not read this roadmap; the predicted-curve software passes its sanity test against the Troels Nielsen pose; the workbench can emit an FOV-matched continuation overlay. ### Phase 2 — Pre-Registration of Prediction 1 Goal: write the full six-slot pre-registration spec for the sub-visible parhelic circle continuation, with the predicted curve computed and the smoke gate quantified, *before* any observation run. Deliverables: - `docs/perception/PHASE2_PREDICTION_1.md` containing: 1. **Angular positions:** the great-circle path in real-sky coordinates and the projected pixel path in camera coordinates for a target observation window (default: sun altitude 15°–30°, mid-afternoon high-cirrus). Continuation segments outside the visible parhelic arc are listed explicitly with start/end angular positions and the photometric contrast value at which the arc's visible portion drops off (estimated from the existing calibration photo set under `docs/calibration/`). 2. **Detection regime:** polarization-angle modulation across rotated polarizer exposures; signal is the modulation amplitude at the predicted angular positions, integrated across the exposure series. 3. **Predicted signal curve:** polarization-angle-vs-arc-position curve from the Phase 1 software, with the predicted modulation amplitude in normalized intensity units. The curve is committed before any observation. 4. **Smoke gate:** noise floor measured at sky-background regions outside the predicted path; smoke gate is the apparatus-judged- insufficient threshold below which a null result is *not* taken as falsifying the prediction. Initial proposal: smoke gate at 3× background modulation amplitude. Calibrated against the reference card before each observation run. 5. **Null condition:** integrated modulation amplitude at the predicted continuation positions does not exceed background by ≥ 3× while the apparatus passes its smoke gate. Distinct from "apparatus failed" or "cirrus was too thin." 6. **Positive condition:** integrated modulation amplitude at the predicted positions exceeds background by ≥ 5× *and* the polarization angle's variation along the path tracks the predicted curve to within an agreed angular tolerance (initial: ±10° along the segment, sharpened in Phase 3 if the first run is positive). - **Discrimination against confounders:** explicit list of confounders (scattered-sky Rayleigh polarization at the observation angle; thin cloud structure mimicking a curved signal; reflective ground; the polarizer's own retardance) and the test that rules each one out. Rayleigh-background is the dominant confounder; the discrimination test is that the predicted parhelic-continuation polarization angle differs from the local Rayleigh polarization angle by a specific amount derivable from sky geometry, which the apparatus must resolve. - **Pre-registration commitment:** the doc is committed to the repo with no observation data attached. Any later edit that softens the null or positive conditions must be filed as a separate amendment block with a written justification. Exit criterion: `PHASE2_PREDICTION_1.md` lands, the predicted curve and smoke gate are quantitative, and no observation has been run yet. ### Phase 3 — First Observation Run Goal: run the smartphone-tier apparatus against high-cirrus weather producing a visible parhelion, capture the scripted exposure series at the polarizer-rotation steps Phase 2 named, and integrate the modulation amplitude along the predicted continuation. Deliverables: - Raw exposure series committed to `docs/perception/observations/ _/` along with the EXIF, polarizer rotation angle, and apparatus-state log per frame. - An analysis notebook or script that reproduces the integration of modulation amplitude along the Phase 2 predicted path, using the Phase 1 software as the geometric reference. - A run-card (`PHASE3_RUN_.md`) recording the apparatus smoke-gate measurement, the cirrus condition assessment, the observation geometry, and whether the run is judged valid or rejected at the smoke-gate level. Exit criterion: at least one observation run passes the smoke gate. Phase 3 may require multiple weather-gated attempts; that is expected and is not a failure. The exit criterion is one valid run, not one successful prediction. ### Phase 4 — Result Note (Mesa-Style Discipline) Goal: write the result note in the same shape as [`mesa/PHASE6_V32_RESULTS.md`](mesa/PHASE6_V32_RESULTS.md) and adjacent notes — clean structure, honest verdict, no rhetorical inflation. Deliverables: - `docs/perception/PHASE4_RESULTS.md` recording: - The Phase 2 pre-registration referenced verbatim. - The Phase 3 valid run(s) summarised by their run-cards. - The measured modulation amplitude at each predicted position, against the smoke gate, null condition, and positive condition Phase 2 named. - The verdict: **positive**, **partial** (signal but does not track the predicted curve closely enough), **null** (no signal above smoke gate), or **inconclusive** (apparatus failed smoke gate across all attempts). - The methodological lessons the run produced, regardless of verdict. - Cross-reference back to the cross-pollination memo at the top of this roadmap and to MESA_CROSSOVER_NOTE.md. - Verdict-conditional next-step routing: - *Positive* → Phase 5 promoted (intersection enhancement). - *Partial* → Phase 2 refinement amendment; tighter prediction bounds, possibly an apparatus calibration upgrade. - *Null with clean smoke gate* → publishable negative; the sub-visible parhelic continuation prediction is falsified at the smartphone-tier apparatus. Phase 5 conditional on whether higher-tier apparatus could still rescue the prediction or whether the falsification generalises. - *Inconclusive* → apparatus-tier revisit; the smartphone tier may be the wrong floor. Exit criterion: the result note lands with a verdict drawn from the four above, and the verdict-conditional routing decision is made and documented. ### Phase 5 — Apparatus Upgrade and Prediction 2 (Intersection Enhancement) Goal: conditional on a Phase 4 positive or partial verdict, promote the intersection-enhancement prediction and the narrowband / long-exposure apparatus. This phase is **gated**, not scheduled. It runs if and only if Phase 4 verdict promotes it. If Phase 4 returns a clean null, Phase 5 may still run with a revised prediction that addresses the Phase 4 null, but the methodological burden shifts: the new prediction must explain why the sub-visible parhelic continuation was null while the new target is expected to signal. Deliverables (when gated open): - Apparatus upgrade: narrowband filter (initial proposal: an H-α or similar narrowband for cleanest signal-vs-Rayleigh discrimination, selected against the predicted dispersion curve), long-exposure capable camera, optionally a small equatorial mount if the observation window exceeds a few minutes. - Predicted-curve software extension to compute predicted intensity enhancement at circle-circle intersections, using prism-path overlap factors from Greenler/Tape. - Pre-registration spec `PHASE5_PREDICTION_2.md` filling all six slots of the template, with the same discipline as Phase 2. - One or more observation runs against the new prediction. - Result note `PHASE5_RESULTS.md` in the same shape as Phase 4. Exit criterion: the intersection-enhancement prediction has a verdict drawn from positive / partial / null / inconclusive, with the result note committed. ### Phase 6 — Reach Tier: Color-Order Inversion at Arc Tangencies Goal: conditional on a Phase 5 positive or partial verdict, promote the color-order inversion prediction and the spectrometer-tier apparatus. This phase is **explicitly reach** territory by the doc's original classification. It enters the roadmap only after both lower-tier predictions have moved. Deliverables (when gated open): - Apparatus: a small calibrated spectrometer (fiber-coupled, sufficient to resolve the dispersion gradient at the predicted tangency position). Apparatus selection note documenting the cost/calibration tradeoffs. - Predicted-curve software extension for chromatic dispersion gradient cancellation/reinforcement at named tangencies (22° halo red-inside / blue-outside meeting CZA's inverted ordering at the 46° halo top tangent, for example). The candidate tangencies are enumerated in `PHASE6_TANGENCY_INVENTORY.md`. - Pre-registration spec `PHASE6_PREDICTION_3.md` filling all six slots of the template. - Observation run(s) and result note `PHASE6_RESULTS.md`. Exit criterion: the color-order inversion prediction has a verdict and a result note, *or* the spectrometer apparatus cost is judged unjustifiable against the cumulative Phase 4 + Phase 5 evidence, in which case the phase is closed unrun with an explicit "not promoted" note. ### Phase 7 — Earned-Language Ratchet and Public Artifact Goal: turn the cumulative result into a defensible public claim, update the gravity ledger and the public-facing sundog.cc surfaces, and ship a public artifact only if there is something earned to show. Deliverables (verdict-conditional): - Update `docs/SUNDOG_V_GRAVITY.md` §The Gravity Claim §Two-substrate convergence: add the predictive-receipt anchor and link to `PHASE4_RESULTS.md` (and Phase 5 / Phase 6 results when applicable). - Update `docs/presentation/claims-and-scope.md` §Unsupported Universal Claims if any Phase 4–6 result reframes what is and is not earned. - Update `docs/promo/PROMO_HIGHLIGHTS.md` if and only if Phase 4 (or later) earned-language survives the discipline review. - Update `docs/SUNDOG_V_GEOMETRY.md` §What the atlas does NOT yet model: move any item that has now been instrumented to a separate §Predictions confirmed by perception section; leave the remaining items as both apology *and* prediction surface, with a forward pointer to this roadmap. - A short interactive page at `perception.html` (sibling to `mesa.html`, `threebody.html`, `balance.html`, `mines.html`) showing the prediction-overlay workflow with the Phase 4 (and later) result as the load-bearing example. Reach: an in-browser playback of the modulation-amplitude integration against an interactive sun-altitude slider, with the Phase 4 observation run as the canonical example. Avoid in Phase 7 language: - "Sundog instrumentation discovered X." - "The gravity claim is now confirmed in the wild." - "Halo physics has been rewritten." Use in Phase 7 language (verdict-conditional): - "Sundog geometry, in the tested apparatus tier and weather regime, predicts a sub-visible parhelion-continuation feature that the apparatus confirms / does not confirm." - "The atlas now ships a predictive overlay alongside its descriptive one." - "The two-substrate convergence on the gravity-frame field-shape gains a third receipt: in-vitro (mesa), in-the-wild observational (atlas overlays), in-the-wild predictive (perception)." Only after a Phase 4 positive. Exit criterion: the public-facing surfaces, the gravity ledger, and this roadmap all use the strongest claim Phase 4 (and any later phases) earned, and no stronger one. If Phase 4 returned a clean null, Phase 7 ships the publishable negative without rhetorical inflation. --- ## Downstream Promotions and Dependencies This roadmap depends on: - **`SUNDOG_V_GEOMETRY.md`** — the halo atlas, the inverse-inference calibration result, and the `public/poses/canonical-halo-atlas.json` pose are load-bearing for Phase 1's predicted-curve software. The atlas formulae are the geometric source-of-truth; this roadmap does not re-derive them. - **`SUNDOG_V_MESA.md`** — Phase 6 methodology (pre-registration, smoke gates, mesa-style result-note discipline) is the template Phase 0–4 of this roadmap inherit. The four stacked methodological lessons from mesa v3.x apply here too: feature availability is not mechanism, variance is not mechanism, partial delivery of the right basis fails, single-component sufficiency fails. The atlas-as- instrument framing is the geometry-side answer to those lessons. This roadmap unlocks: - **Gravity ledger Candidate 12** (new, to be filed): *Predictive Atmospheric Instrumentation.* Promoted into this roadmap. The candidate slot in `SUNDOG_V_GRAVITY.md` is reserved for archival reference once Phase 0 lands. - **Geometry Phase 12 (Atmospheric Variations)** and **Phase 13 (Linked Description Mode)** become richer once the predictive overlay is built: alternative sky palettes can carry their own predicted-continuation overlays; the linked-description popovers can cite Phase 4 results where applicable. This roadmap does **not** absorb: - The Spacecraft / Conservation-Law / Fluid-Wake candidates in the gravity ledger. Those are independent physical-domain experiments that test the gravity claim in domains other than parhelion optics. - The mesa-trap roadmap. Mesa attacks the in-silico falsification mode; perception attacks the in-the-wild predictive-receipt mode. Both ratchet the gravity claim by independent paths. ## Implementation Status **Phase 0:** Not started. This roadmap is the Phase 0 commitment artifact; the deliverables above (apparatus tier lock, prediction slate, sort, literature spine, pre-registration template) are now named here but not yet filed as separate sub-documents. The next move is to file `docs/perception/APPARATUS_PROTOCOL.md` and lock the literature spine in writing, which completes Phase 0. **Phases 1–7:** Not started; gated on Phase 0 completion (Phases 1–4) and on Phase 4 verdict (Phases 5–6) and on cumulative evidence (Phase 7).