# Structural Failure Boundary Map (P0 deliverable) Pre-registration: [`README.md`](README.md) (frozen 2026-05-15 PT) Roadmap: [`SUNDOG_V_GRAVITY.md`](../../SUNDOG_V_GRAVITY.md) ▸ Candidate 13 ▸ Roadmap Frozen: **2026-05-15 (PT)**. Append-only below the **Amendments** rule; the body above it is the P0 artifact and is frozen at write time. This is the P0 deliverable the prereg's *First Falsifier Before Agents* section requires to exist before any agent is admitted. It is the falsifier: a frozen map of where the documented closed-form inverse is ill-posed, what a *traceable* agent must do there, and what a *mere correlate* would do instead — written entirely from existing geometry receipts, **no agent, no render, no training**. Each locus carries the five fields the prereg mandates: **eligible regime · abstain/switch/fail regime · exact source receipt · traceable-agent prediction · mere-correlate prediction.** A row with no crisply-citable receipt is marked **BLOCKED** and halts the program (fix = more geometry specification, not more training). Reconciliation policy (prereg P1): where the project's *coded guard*, the *literature value*, and the *prereg-stated* number differ, the **coded guard is the operative boundary** (it is what any harness will actually enforce); the others are recorded for honesty, not silently averaged. --- ## L1 — Parhelion offset route (the sole promoted inverse handle) - **Inverse.** `offset = R22 / cos(h)` — `phase3.daggerOffset(h)` in `public/js/parhelion-geometry.mjs` (≈ line 1060). - **Eligible regime.** The strict eligible photo set **p2, p7, p13** only — the sole promoted hidden-state route after the Phase 10 audit. - **Abstain / fail regime.** - Low-h, low-leverage: `sec(h) − 1` below ~2% of R22 ⇒ anchor-noise-bounded, not an independent handle. - Parhelion-derived-R22 photos ⇒ **tautological** (R22 came from the parhelion; recovering h from it is circular), not independent evidence. - p26 right side ⇒ **geometrically invalid**, excluded. - **Exact source receipt.** `docs/calibration/HALO_PHENOMENA_ACCOUNTING.md` §A "Sundog / parhelion" (sole promoted inverse handle; strict eligible set p2/p7/p13); `docs/calibration/PHASE10_OPTICAL_AUDIT_HANDOFF.md` (the optical audit that promoted it and bounded eligibility); `public/js/parhelion-geometry.mjs` `phase3.daggerOffset`. - **Traceable-agent prediction.** Succeeds on the strict eligible set; reports *low leverage* or *ineligible* on the low-h / tautological / invalid rows and does **not** count them as independent inverse evidence. - **Mere-correlate prediction.** Emits smooth, confident `h` estimates across low-leverage, tautological, and invalid rows because image style, crop, metadata, or halo prominence co-varies with `h`. ## L2 — CZA visibility cutoff - **Inverse.** Circumzenithal-arc route, gated by `czaVisibleAtAltitude(h)` in `public/js/parhelion-geometry.mjs` (≈ lines 965–968). - **Eligible regime.** Sun altitude **h ≤ 32°** (the coded guard: `czaVisibleAtAltitude` returns true only for `altitudeDeg ≤ 32`). - **Abstain / switch / fail regime.** Above the cutoff the CZA exits the visible hemisphere; a CZA-dependent route must fail, abstain, or switch handles. - **Exact source receipt.** **Coded guard (operative):** `public/js/parhelion-geometry.mjs` `czaVisibleAtAltitude` ⇒ `h ≤ 32°`. **Literature (recorded, not operative):** Tape *Atmospheric Halos* Ch. 6 p63 "about 32°"; the earlier accounting matrix cited the closed-form cutoff h = 32.196°; the prereg states "about 32.2 deg". Operative boundary = **32°** per the coded guard; the ~0.2° spread between coded/literature is itself inside the tolerance band and is recorded here rather than averaged. - **Traceable-agent prediction.** A CZA-dependent route fails / abstains / switches at the cutoff; it does **not** preserve a CZA-apex inverse past disappearance. - **Mere-correlate prediction.** Continues to report altitude through the cutoff because other image features still carry distributional information about `h`. ## L3 — Tangent arc → circumscribed-halo merge - **Inverse.** `tangentArcLocus(h)` upper-tangent handle in `public/js/parhelion-geometry.mjs`. - **Eligible regime.** **h < 29°** (separate upper-tangent handle exists). - **Abstain / switch / fail regime.** **h ≥ 29°**: `tangentArcLocus` returns `null` — the upper/lower tangent arcs have merged into the circumscribed-halo regime; no separate tangent-curvature handle. - **Exact source receipt.** `public/js/parhelion-geometry.mjs` `const TANGENT_ARC_CIRCUMSCRIBED_H = 29` (≈ line 813) and the `tangentArcLocus` `null`-return guard (≈ line 830); cross-validated by `docs/calibration/PASS_C7_OUTPUT.txt` and Tape *Atmospheric Halos* Ch. 6 p62 ("at a sun elevation of 29° the two halos merge … the value 29° is theoretical"). Coded boundary and literature agree at 29° here. - **Traceable-agent prediction.** A tangent-dependent route degrades, abstains, or switches at the merge; it does **not** claim continuous tangent-curvature recovery through the singularity. - **Mere-correlate prediction.** Maintains a continuous tangent-like estimate through the merge — a learned texture/shape correlate, not the canonical tangent handle. ## L4 — Supralateral route (structural-discrimination gate) - **Inverse.** Supralateral angular position → `h` (a *candidate* handle that the audit rejected). - **Eligible regime.** **None.** This row is a permanent *fail* row, not a windowed one: the handle fails the structural-discrimination gate at every altitude. - **Abstain / fail regime.** All `h`. Receipt figure: supralateral angular distance from the sun varies only **~0.5° across h = 0–22°**, below the typical ~5–10 px visual-edge measurement noise even at perfect coverage. - **Exact source receipt.** `docs/calibration/PHASE10_OPTICAL_AUDIT_HANDOFF.md` line 54 ("Supralateral position → h | **fails structural-discrimination gate** | … varies only ~0.5° across h = 0–22°, below the typical 5–10 px visual-edge measurement noise"). The prereg's "~0.3 deg" is the narrower tested-low-altitude-span figure; the operative receipt value is **~0.5° over h = 0–22°** per the handoff. Either way the row is a hard fail. - **Traceable-agent prediction.** Refuses to promote supralateral position as a useful inverse handle under the documented apparatus. - **Mere-correlate prediction.** Treats supralateral brightness, crop position, or co-occurring arcs as a usable altitude channel. ## L5 — Rendered ≠ anchored - **Inverse.** None directly — this is the *evidence-admissibility* boundary that governs every other row. - **Eligible regime.** Only **anchored closed-form** rows (the §A core with a project receipt) may count as inverse evidence. - **Abstain / fail regime.** `rendered-optional`, `named-only`, `not-modeled`, and the hardcoded atlas-placeholder primitives (supralateral / suncave-Parry / Parry-supralateral / infralateral draw code, the parhelic-circle empirical smile, the sun-pillar vesica) may support display / vocabulary / future hypotheses but **never** traceability. - **Exact source receipt.** `docs/calibration/HALO_PHENOMENA_ACCOUNTING.md` §A "Honesty ratchet — rendered ≠ anchored", plus the Status Vocabulary table. - **Traceable-agent prediction.** Counts only anchored closed-form rows as inverse evidence; treats drawn-but-unanchored primitives as non-evidence. - **Mere-correlate prediction.** Uses the presence of any drawn or named primitive as evidence that an inverse is available. --- ## P0 completion gate Every locus L1–L5 is written crisply from a **named, verified source receipt** (file + section/line checked 2026-05-15). No row is BLOCKED. Two coded-vs-stated reconciliations are recorded, not hidden: L2 CZA (coded `h ≤ 32°` operative; literature/prereg ~32.2°/32.196°) and L4 supralateral (receipt ~0.5°/0–22°; prereg ~0.3° narrower span). Neither changes a regime classification. **Verdict: P0 gate PASSES — the boundary map is freezable.** The program is admitted to **P1 (falsifier admission review)**. This is a genuine, non-trivial outcome and the payoff of the Phase 14/15 receipt discipline: the inverse's singularities were already documented precisely enough that the falsifier could be written *before* any agent — exactly the condition the prereg required, and the condition that distinguishes a testable apparatus from a theorem posture. P2 (agent run) remains **blocked** until P1 signs off. No public copy may use theorem / universal-proof language; the prereg Public-Language Constraint is in force. --- ## Amendments Append-only. Each amendment: timestamp (date + zone), author, one-line justification. The body above is frozen at P0 write time. **2026-05-15 (PT) — maintainer.** P1 admission review filed ([`P1_ADMISSION.md`](P1_ADMISSION.md)); verdict **PASS** with carried findings: (1) **L5 re-scoped** — admitted as the *evidence-admissibility rule governing L1–L4 scoring*, not as a standalone behavioral failure-coincidence locus (closes a definitional-escape soft spot; the frozen L5 body above stands as the pre-registered record, this is the reviewed operative interpretation); (2) **B-1** operative L2 boundary fixed at the coded guard `h ≤ 32°` (literature ~32.2°/32.196° recorded, not averaged); (3) **B-2** operative L4 figure fixed at ~0.5°/h=0–22° (`PHASE10_OPTICAL_AUDIT_HANDOFF.md` line 54), L4 remains a permanent-fail row at all altitudes. Justification: independent P1 pass per the roadmap gate; reconciliations ruled in writing; no regime reclassified. The falsifier is now cross-referenced to the existing surface ([`../../debunked.md`](../../debunked.md)) — orthogonal to Pushable Occluder, adjacent to Score Aliasing, and the rigorous instrument for the `Proxy Collapse` avenue; prioritization + verdict-vocabulary bridge fixed in `P1_ADMISSION.md` §C.