# Phase 10 Optical Re-Audit Memo Filed: 2026-05-14 Scope: post-pass state after the Phase 10 attack campaign required passes (`§6` hedges, handoff stop banner, B1, A1a, A1b, A2, A3, C1, B2). C2 remains optional and unrun. ## Verdict The re-audit gate clears the technical-pass wave. No new code, anchor, or route-math blocker was found after the post-pass state was checked against the synthetic three-persona protocol. This is a **clearance to rewrite**, not a clearance to send the old specialist handoff unchanged. The stale handoff and public-framing surfaces must be ratcheted to the post-pass failure taxonomy before external use. ## Persona Re-Audit ### Persona 1 - Tangent / Primitive Taxonomy Pass C1 fixed the load-bearing taxonomy error: p7 is no longer counted as upper-tangent-route evidence because h = 59.4 deg is in the circumscribed-halo regime. The post-C1 sampled tangent set is p2, p13, and p27. The remaining tangent verdict is bounded: the column-peak detector fails on the post-C1 sampled set, but this is a protocol-conditional negative. With C2 unrun *(at original filing — see Post-C2, Post-C4, Post-C5, and Post-C6 Addenda below; all four passes landed later the same day; C2 / C4 / C6 returned not-recovered; C5 manual sample selection recovered the route on p2 with R\_uta\_obs / R22 = 0.824 and RMS = 1.23 px but C6 matched-filter on the same b\* substrate falsified the C5→matched-filter natural extension, putting the route in C5↔C6 substrate tension that recommends specialist re-anchoring as the verify gate)*, the correct handoff language was "unresolved open question" for a wing-based or Lab b* ridge detector, not a class-level tangent failure. Finding: no new tangent-route blocker. The only remaining risk is stale surface language that still says "all four eligible photos" or implies a generic tangent-route failure. ### Persona 2 - CZA / Supralateral Route Semantics Pass A1a/A1b repaired the CZA atlas formula and verified the repair. On p2, the CZA apex y residual collapses from the legacy -19 px class of error to about +1.3 px in the current sign convention. Pass A2 removes p27 from CZA evidence by reclassifying its visible chromatic feature as 46 deg halo top / supralateral merger. The CZA route now fails coverage, not residual reliability: p2 is the only in-window measured CZA anchor; the other anchored photos are either beyond the CZA disappearance altitude or have the literature CZA apex off-frame. The supralateral route remains blocked even if coverage is read permissively. Its h-sensitivity across the relevant low-altitude range is below visual-edge measurement noise, so the route is structurally weak as an inverse handle. Finding: no new CZA or supralateral blocker. The corrected verdict is dataset/aspect-ratio coverage for CZA and physics-discrimination for supralateral. ### Persona 3 - Parhelion Route / Public Claim Pass B1 and B2 fixed the parhelion overclaim without demoting the route. The promoted wording is now restricted to p2, p7, and p13: photos with unambiguous bilateral peaks, valid geometry, non-trivial discrimination, and an independently fittable 22 deg halo. The low-h photos remain informational, not promotional. p20, p25, and p26 have parhelion-derived R22 values; p27 stipulates offset := R22; p26 right side is explicitly invalid because R22 / offset > 1. Finding: parhelion-offset remains the only promoted inverse handle. The old "every eligible photo" phrase is retired and should appear only in retraction or historical-audit contexts. ## Consolidator Verification - Ran `python scripts\test_cza_formula.py`. The test prints h = 22 literature CZA above sun = 45.734 deg, CZA disappearance altitude = 32.196 deg, p2 literature residual = -1.3 px observed-minus-predicted (equivalent to +1.3 px predicted-minus-observed in overlay notes), and p27 literature CZA y = -11.5, confirming the visible p27 feature is not CZA. - Ran `python scripts\overlay_calibrate.py docs\calibration\2.Photometeor-jeff_mod_red.jpg --anchors docs\calibration\p2-anchor.json --sun-altitude 18.6 --supralateral 0.40 --out $env:TEMP\p2_reaudit_overlay.png`. The script reports CZA apex predicted at `(567.0, 114.3)`, residual `(dx, dy) = (+7.0, +1.3) px`, and parhelion residuals near zero. - Checked the post-pass route tables in `docs/calibration/RICH_DISPLAY_OVERLAY_NOTES.md`: A3, C1, and B2 are present and agree on the current taxonomy. - Checked `docs/SUNDOG_V_GEOMETRY.md`: the closeout table already uses the post-pass route verdicts. - Checked p26 schema semantics from the B1/B2 table: the right-side parhelion remains invalid because `323 / 322 > 1`. Verification gate result: all load-bearing numeric and schema checks agree with the post-pass route taxonomy. No persona overstatement survived the consolidator pass. ## Post-Re-Audit Failure Taxonomy | route | gate | failure-mode kind | | --- | --- | --- | | Parhelion offset -> h | promoted, post-hedged | 3-photo strict eligibility: p2, p7, p13 | | CZA apex -> h | fails coverage | dataset/aspect-ratio: only p2 is in-window and measured | | Supralateral -> h | fails structural discrimination | physics-shape: h-spread below visual-edge noise | | Tangent-arc curvature -> h | partially recovered on p2 under manual sample selection; coverage gate fails; C5<->C6 substrate tension flagged *(see Post-C2 + Post-C4 + Post-C5 + Post-C6 Addenda)* | hybrid coverage + tooling with verify-gate: 4 automated detectors fail (C2/C4/C6 plus column-peak); C5 manual recovers on p2 with methodology hedge; p13/p27 substrate-signal absent; recommended specialist re-anchoring of p2 | The pre-audit "three independent failure layers" phrase stays retired. The post-audit phrasing is "three structurally different failure modes": dataset/aspect-ratio, physics-discrimination, and tooling-protocol. ## Required Next Step Proceed to the specialist handoff rewrite and public-framing ratchet: 1. Rewrite `docs/calibration/PHASE10_OPTICAL_AUDIT_HANDOFF.md` from the post-pass taxonomy; keep C2 as an unresolved open question. 2. Ratchet `docs/MESA_CROSSOVER_NOTE.md` and `docs/SUNDOG_V_GRAVITY.md` out of pre-audit hedge mode and replace old residual/failure tables. 3. Keep `index.html` and `chat/claim_map.json` conservative until the public claim copy is explicitly approved against this memo. 4. Check `docs/promo/PROMO_HIGHLIGHTS.md` for stale "all four eligible" tangent wording before any external reuse. ## Post-C2 Addendum *(2026-05-14)* The original memo above was filed when Pass C2 was unrun and the tangent detector was filed as an Unresolved Open Question. C2 has now landed in the same wave (`scripts/tangent_detector.py` + `scripts/test_tangent_detector.py`; captured run output at `docs/calibration/PASS_C2_DETECTOR_OUTPUT.txt`; full receipt in `docs/calibration/RICH_DISPLAY_OVERLAY_NOTES.md` under "### Pass C2 Update"). This addendum records the disposition change for the tangent row of the verdict table above; the rest of the memo stands. **Pass C2 result.** A wing-azimuth-offset Lab b\* ridge detector with 22°-halo-radial-profile subtraction (the literature-standard substrate recommended by `PHASE10_OPTICAL_AUDIT_SYNTHETIC_MEMO.md` §4.8 / Persona 1 §4 item 6) ran on the post-C1 sampled set (p2 / p13 / p27). Pre-registered coherence gate: a wing is ridge-detected if ≥ 8 / 24 samples have residual b\* amplitude ≥ 3.0 AND radial offset ≤ ±10 px from the predicted tangent locus. Result on all three photos: 0-2 / ~24 coherent samples per wing, well below the gate. Verdict: **not-recovered**. **Methodology hedge that did not save the route.** The pilot run *without* the halo-radial subtraction step exhibited boundary-pinned median radial offsets at ±18 px, consistent with the detector finding the 22° halo's own chromatic ridge inside the search window. Adding the halo-radial subtraction step un-pinned the offsets (post-subtraction medians: −7.5 to +10.5 px across the six wings, well inside the ±18 px window) and the negative survived. The failure cannot be attributed to halo-ridge contamination of the detection window. **Taxonomy update.** The "tooling-protocol" classification of the tangent-route failure survives but narrows: the route fails under *two* literature-standard detector families (column-peak intensity at sun meridian; wing-radial Lab b\* with halo-radial subtraction). The residual open question is whether non-literature-standard detector designs (wing-slope geometric curvature, matched-filter against a parameterized arc model, polarization-channel filtering) might recover the route. These are filed as Phase 10 backlog ideas, not Phase 10 deliverables. The "three structurally different failure modes" framing otherwise survives unchanged. **Handoff implication.** `PHASE10_OPTICAL_AUDIT_HANDOFF.md` §2.3 (the specialist question "is a wing-based / Lab b\* detector worth building?") was already answered by the team — the detector has been built and ran negative. The specialist question reframes to: "are the non-literature-standard detector designs (wing-slope curvature, matched-filter, polarization) worth building before the route is called dead?" The handoff was updated in the same wave to carry the post-C2 verdict; this addendum is the load-bearing pointer for any reader following the cross-reference back from the handoff to this memo. **Single-handle verdict unchanged.** Parhelion-offset remains the sole promoted inverse handle. The forward-rich / inverse-narrow asymmetric field-shape pattern survives at the same one-handle resolution; C2 ruled out a *literature-standard* path to two-handle promotion but did not rule out two-handle promotion via a different detector substrate. ## Post-C4 Addendum *(2026-05-14)* The Post-C2 Addendum above narrowed the tangent-route open question to "non-literature-standard detector designs (wing-slope curvature, matched-filter, polarization)." Pass C4 lands the first of those — the wing-slope geometric curvature detector Persona 1 §5 explicitly named — and tests it on the same eligibility set. Implementation: `scripts/tangent_curvature.py` (regression runner `scripts/test_tangent_curvature.py`; captured run output at `docs/calibration/PASS_C4_DETECTOR_OUTPUT.txt`; full receipt in "### Pass C4 Update" of `docs/calibration/RICH_DISPLAY_OVERLAY_NOTES.md`). **Pass C4 result.** A wing-slope luminance-gradient edge detector with circle fit ran on p2 / p13 / p27. Pre-registered gates: gradient magnitude ≥ 1.5 L\*/px AND radial offset ≤ ±12 px from predicted tangent locus AND ≥ 12 surviving candidates AND circle-fit RMS ≤ 8 px AND fitted R\_uta / R22 ∈ [0.7, 1.3]. Result on every photo: 88-100% of wing samples rejected as weak-gradient, leaving 0-4 surviving candidates — too few for any circle fit. **Not-recovered** on all three photos. **Reclassification of Persona 1's §5 alternative.** The Post-C2 addendum classified wing-slope geometric curvature as a "non-literature-standard detector design." Reading Persona 1 §5 more carefully, this is wrong: Persona 1 explicitly named gradient-based edge detection as the literature-standard alternative to peak-based ridge detection (*"real measurement needs either gradient-based edge detection (the spine is at a brightness or chromaticity transition, not a peak) or manual sample selection from visual crops"*). C4 tested the gradient-based edge detection branch. The route now fails under **three** literature-standard detector families. **Taxonomy update.** The "tooling-protocol" classification of the tangent-route failure survives but narrows further. The remaining candidate detector designs / substrates that have NOT been tested: - **Manual sample selection from visual crops** (Persona 1 §5's literature-standard alternative #2). Hand-anchoring rather than automated detection. `tangent_curvature.fit_circle` already accepts arbitrary (x, y) point lists, so if the team hand-anchored arc points on each photo, the existing circle-fit machinery could run. - **Matched-filter detection against a parameterized arc model.** A different signal (whole-shape correlation) than the per-sample edge or ridge picks tested so far. Audit memo doesn't classify this as literature-standard or not. - **Polarization-channel filtering.** Different *substrate* — requires polarizer-equipped follow-up photos. Substrate test, not detector test on the same substrate. - **New calibration photos with stronger tangent display.** Coverage expansion rather than detector design. The wing-region L\*-gradient rejection rate (88-100%) suggests the calibration set may genuinely not carry enough tangent signal for any radial-profile detector at the pre-registered amplitude. **Handoff implication.** `PHASE10_OPTICAL_AUDIT_HANDOFF.md` §2.3 was updated in the same wave to reflect the post-C4 state: the specialist question reframes from "is wing-slope curvature worth building?" (now answered: built and ran negative) to "is manual hand-anchoring, matched-filter detection, polarization filtering, or new calibration photos the right next step?" The §2.3 prose carries the four options. **Single-handle verdict unchanged.** Parhelion-offset remains the sole promoted inverse handle. The "forward-rich / inverse-narrow" framing stays at the same one-handle resolution; C4 ruled out the second literature-standard alternative for two-handle promotion but did not rule out the remaining four candidates. ## Post-C5 Addendum *(2026-05-14)* The Post-C4 Addendum above narrowed the tangent-route open question to "manual sample selection from visual crops" (Persona 1 §5's other literature-standard alternative), matched-filter detection, polarization filtering, and new calibration photos. Pass C5 lands the first of those. **Pass C5 result.** Manual hand-anchored upper-tangent points were added to `upper_tangent_manual_samples` blocks of each post-C1 eligibility-photo anchor JSON. The runner at `scripts/test_tangent_manual.py` applies the C4 circle-fit machinery to the manual points under pre-registered gates: ≥ 5 manual points, circle-fit RMS ≤ 10 px, R\_uta\_obs / R22 ∈ [0.7, 1.3]. Captured run output: `docs/calibration/PASS_C5_DETECTOR_OUTPUT.txt`. Per-photo result: - **p2: route recovered.** 5 hand-anchored points fit a circle with R\_uta\_obs / R22 = **0.824** (within tolerance) and RMS = **1.23 px** (well under 10 px). Center fit: (567.5, 160.3) — essentially on the sun's vertical axis (sun\_x = 567). The upper tangent arc is geometrically real and circular on the cleanest display. - **p13: not recovered.** Only the apex is marginally anchorable (faint chromatic brightening at the halo crown); the wing region is washed into a luminous chromatic gradient with no localizable spine (consistent with Persona 1 §3 p13 entry). Insufficient samples for a circle fit. - **p27: not recovered.** Zero anchorable points. The sun-bloom column dominates the halo-crown region where the upper tangent arc would sit (consistent with Persona 1 §3 p27 entry); no upper-tangent feature is separable from the sun-bloom. The chromatic arc visible at the top of the frame is the supralateral / 46° halo top merger (per Pass A2 reclassification), not upper tangent. **Coverage gate fails** at 1 / 3 photos < pre-registered ≥ 2 threshold. **Reclassification of the failure mode.** The Post-C4 Addendum classified the tangent-route failure as "detection-protocol tooling" under three literature-standard detectors. C5 refines this to a **hybrid coverage + detection-tooling failure**: - On p2, the signal is real and recoverable; the three automated detectors (C1 column-peak, C2 wing-radial Lab b\*, C4 wing-slope luminance-gradient) all miss what manual hand-anchoring picks up. This is *detection-tooling* in a specific sense: the arc's image signature is at the resolution of visual chromatic-dome identification, not at the resolution of per-sample radial-profile peak picks. A matched-filter detector that tests the whole-arc-shape hypothesis (rather than per-sample features) should be able to recover what C2 / C4 miss; C5's positive result on p2 is the strongest evidence for matched-filter being the right next detector. - On p13 / p27, no detector — automated or manual — recovers the signal. This is *substrate-signal absence*, not a detector design problem. Better tooling does not help; new substrate (new photos or polarization filtering) does. The "three structurally different failure modes" framing from the post-A3 + post-C1 taxonomy survives but the tangent classification sharpens: | route | failure mode kind | | --- | --- | | Parhelion offset -> h | promoted on strict 3-photo subset | | CZA apex -> h | dataset / aspect-ratio coverage | | Supralateral -> h | atmospheric-physics discrimination | | Tangent-arc curvature -> h | hybrid coverage (p13/p27 substrate absence) + detection-tooling (p2 signal present but missed by 3 automated detectors) | **Methodology hedge.** The p2 manual anchoring was performed by visual identification at display resolution by the project assistant, with per-point uncertainty estimated at ±15 px. The RMS = 1.23 px fit result is suspiciously tight given that uncertainty and could partially reflect symmetry bias. The points are recorded in `p2-anchor.json` `upper_tangent_manual_samples.points`; a specialist re-anchoring is a one-script operation against the existing JSON. **The C5 verdict should be treated as preliminary pending specialist re-anchoring** before any public-framing surface relies on the p2 recovery as a confirmed positive. **Handoff implication.** The handoff `PHASE10_OPTICAL_AUDIT_HANDOFF.md` §2.3 was updated in the same wave: the specialist question reframes from "is manual hand-anchoring worth doing?" (now answered — done, positive on p2) to "is matched-filter detection the right next literature-standard step?" plus a request for specialist re-anchoring of p2 as a verify gate. **Single-handle verdict unchanged.** Parhelion-offset remains the sole *promoted* inverse handle. The tangent route's verdict shifts from "fails under three literature-standard detectors" to "partially recovered on p2 under manual sample selection; coverage gate fails." The forward-rich / inverse-narrow framing stays at the same one-handle resolution, but the tangent route now has a published positive result on one photo — a constructive update rather than a deeper negative. ## Post-C6 Addendum *(2026-05-14)* The Post-C5 Addendum above named matched-filter detection as the highest-leverage next candidate ("now MORE promising given p2's circular fit"). Pass C6 built the matched-filter detector against a parameterized arc model on the same halo-subtracted Lab b\* substrate C2 used, ran it on the post-C1 sampled set, and **falsified the C5→matched-filter natural-extension hypothesis**. **Pass C6 result.** Detector at `scripts/tangent_matched_filter.py` (regression runner `scripts/test_tangent_matched_filter.py`; captured output at `docs/calibration/PASS_C6_DETECTOR_OUTPUT.txt`; full receipt under "### Pass C6 Update" in `docs/calibration/RICH_DISPLAY_OVERLAY_NOTES.md`). Pre-registered gates: R\_uta\_obs / R22 in [0.7, 1.3]; peak correlation ≥ 0.10; peak prominence ≥ 1.5× baseline. Per-photo result: - **p2: not recovered.** The matched-filter correlation is **negative across the entire R\_uta scan** (from −0.244 at R\_uta = 91 px up to −0.117 at R\_uta = 267 px). There is no positive correlation at any R\_uta in [0.5 × R22, 1.5 × R22], including at R\_uta = 150 consistent with C5's manual fit (R\_uta\_obs = 149.9). The template vs. b\*-residual signal-coupling reverses sign over the wing region. - **p13: not recovered.** Positive correlation peak (+0.191) with 2.05× prominence — but at R\_uta = 106 px = 0.502 × R22, way below the [0.7, 1.3] tolerance. Matched-filter is picking up some chromatic feature at half the predicted curvature, not the upper tangent arc. - **p27: not recovered.** Score essentially zero (−0.001) at all R\_uta. Sun-bloom column dominates the upper region; no coherent chromatic structure correlates with the predicted arc template. **The C5↔C6 contradiction is the load-bearing finding.** C5 manual sample selection recovered p2 with R\_uta = 149.9 (ratio 0.824) and RMS = 1.23 px on the same photo where C6 matched-filter returns correlation = −0.21 at the same R\_uta. Both detectors operate on the same image and (effectively) the same substrate (C6's halo-subtracted b\*; C5's visual identification picks up b\* signal among other things). Yet they reach opposite conclusions. **Two open interpretations:** (a) **Substrate mismatch.** C5's manual identification picked up the arc by *visual gestalt* — combining luminance, chromaticity, spatial coherence, and likely some prior expectation of arc shape. C6 correlates only halo-subtracted Lab b\*. The arc's gestalt signal may not be in halo-subtracted b\* alone — could be in absolute b\* (no subtraction), in L\* magnitude, in chromaticity vector magnitude (`sqrt(a*² + b*²)`), or in some combined channel. C5 positive + C6 negative narrows the substrate space the route's signal lives in; it does NOT rule out automated detection on a different substrate. (b) **Symmetry bias in C5.** The C5 receipt explicitly hedged: *"RMS = 1.23 px is suspiciously tight given ±15 px per-point uncertainty, possibly reflecting symmetry bias."* If the manual hand-anchoring unconsciously favored points consistent with circular shape, C5's clean fit would be partly artifact. C6's null result on the same substrate is consistent with this reading — the matched-filter doesn't share the symmetry bias and reports honestly. Both interpretations are testable. (a) tests by running matched-filter on alternative substrates (absolute b\*, L\* magnitude, chromaticity magnitude). (b) tests by specialist re-anchoring. Both are filed as **Phase 10 backlog**, neither is a Phase 10 deliverable. **Taxonomy update.** The "hybrid coverage + detection-tooling" classification from Post-C5 stands, but the detection-tooling sub-claim now carries a verify-gate flag: > *Four automated detectors (C2 wing-radial Lab b\*; C4 wing-slope > luminance-gradient; C6 matched-filter on halo-subtracted b\*; plus > the original column-peak) all miss p2's signal. Manual sample > selection (C5) recovers on p2 but with a methodology hedge that C6 > now amplifies into a substrate-tension verify-gate.* **Handoff implication.** The handoff `PHASE10_OPTICAL_AUDIT_HANDOFF.md` §2.3 was updated in the same wave: the specialist question now includes a request for independent re-anchoring of p2 as the verify gate, in addition to matched-filter-on-alternative-substrates, polarization filtering, and new calibration photos. **Single-handle verdict unchanged.** Parhelion-offset remains the sole *promoted* inverse handle. The tangent route's "partially recovered on p2" status holds but now carries the C5↔C6 substrate tension and a recommended verify-gate before further propagation into public-framing surfaces. The forward-rich / inverse-narrow framing stays at one-handle resolution.