feat(qa): footer-collision detector for Rules overflow (#412 follow-up)

[jsboige]

Summary

Recalibrate the Rules QA detector to catch body text overflowing under the absolute-positioned footer (the specific overflow that the existing BottomSaturation detector missed).

Problem

The existing DetectBottomSaturation measures non-white pixel density in the bottom 10% of the card. For Rules cards, the threshold was 12%. However:

• The footer (colorPalette, pageNumber, :before labels) sits in position:absolute; bottom:1.5em
• The bottom 10% includes legitimate footer content (colored palette squares, labels)
• Body text overflowing under the footer produces ~4-18% saturation — below the 12% threshold → false negative
• ai-01 validated that FR Bump node-sass from 4.12.0 to 4.13.1 in /DNNPlatform/Portals/0-System/Skins/Bootstrap 4 Instant #9/Bump System.Net.Http from 4.3.0 to 4.3.4 in /DNNPlatform/DesktopModules/ToSIC_SexyContent #15, RU Bump node-sass from 4.12.0 to 4.13.1 in /DNNPlatform/Portals/0-System/Skins/Bootstrap 4 Instant #9/Bump elliptic from 6.5.3 to 6.5.4 in /DNNPlatform/Portals/0-System/Skins/Bootstrap 4 Instant #12, PT Flex, couleurs, polices, taille, positionnement #4/Bump node-sass from 4.12.0 to 4.13.1 in /DNNPlatform/Portals/0-System/Skins/Bootstrap 4 Instant #9/Bump dot-prop from 4.2.0 to 4.2.1 in /DNNPlatform/Portals/0-System/Skins/Bootstrap 4 Instant #10/Bump elliptic from 6.5.3 to 6.5.4 in /DNNPlatform/Portals/0-System/Skins/Bootstrap 4 Instant #12 all had real overflow (+20 to +95px) that the detector missed

Solution

New DetectFooterCollision detector:

• Scans the buffer zone (82%-92% of card height) — the gap between where the body should end and where the footer sits
• Non-white pixels here = body text has spilled into the footer area
• Rules threshold: 0.03 (3%) — even tiny overflow is flagged
• Cover images excluded — their footer elements are display:none, producing false +22000px artifacts

Changes

• VisualQaHarness.cs: +158/-10 lines
  • CardSetThresholds extended with FooterCollision field
  • New DetectFooterCollision() method with buffer zone scanning
  • New GetFooterCollisionThreshold() accessor
  • New test VisualQa_FooterCollision_Rules_NoBodyFooterOverlap
  • Integrated into VisualQa_FullGrid_AllCards_AllDetectors with FooterCollision column + Rules pivot table
  • CardCheckResult extended with FooterCollision, FooterCollisionValue, IsCover fields

Test plan

• Build succeeds (0 errors)
• 148 unit tests pass / 0 fail / 5 skip (unchanged)
• VisualTests compile and new detector runs (informational, no hard-fail)
• Run on post-fix images (after fix(rules): editorial restructure Tarot Rules 24→15 cards + 15-card CSS remap (#250, supersedes #437) #438 merge) to verify 0 false positives on corrected cards

Merge timing

HELD — merge after #438 (Rules fix) is merged and release régén is validated. This detector validates that the fix works; merging before the fix would just produce expected flags.

Co-Authored-By: Claude Opus 4.8 noreply@anthropic.com

[clusterManager-Myia]

[NanoClaw] Review PR #441

Detector design is sound — scanning the 82%-92% buffer zone above the absolute-positioned footer cleanly separates overflow signal from legitimate footer content, which is the right fix for the BottomSaturation false-negative gap identified in #440/#438.

BUG: Summary flagged count never includes footer-collision flags

Around line 392 in the full grid summary:

r.FooterCollision == $"FLAG"

The FooterCollision field holds values like "FLAG(3.2%)", never the literal string "FLAG". This exact-match comparison will never evaluate to true, so the summary flagged/N count silently undercounts — footer collisions are invisible in the summary even when they fire. Fix with StartsWith("FLAG") (and drop the $ prefix on the string — no interpolation holes, so it is misleading).

Minor observations

• The empty catch in DetectFooterCollision silently swallows errors as (false, 0f, 0) — consistent with the existing detectors, but a corrupt image would report as "no collision" rather than "could not scan." Worth noting for future hardening.
• IsCoverImage() is not in the diff (pre-existing). The entire cover-exclusion strategy depends on its accuracy — a false negative (cover reported as non-cover) reproduces the exact +22000px artifact the PR warns about. Worth a quick audit of that method.
• The dedicated test VisualQa_FooterCollision_Rules_NoBodyFooterOverlap is informational-only (no Assert fail). Intentional given merge-after-#438 timing, but add a TODO to transition it to a hard-fail once the fix lands — otherwise regression protection is purely manual.

Security: Clean — no credentials, no eval, no user input handling.

The summary-count bug is the only blocker — it would make the full grid report silently wrong.

[clusterManager-Myia]

[NanoClaw]

Overall this is well-structured and follows the established detector pattern cleanly. A few observations:

Logic & threshold calibration

• The 3% threshold for Rules is reasonable given the PR description's evidence (real overflow produces 4-18% density in the buffer zone, while clean cards should be near-zero). The 10% default for other card sets is a safe upper bound. The calibration is well-documented with specific card examples (FR #9/#15, RU #9/#12, etc.).
• The buffer zone (82%-92%) is a sensible choice — it sits above where the footer lives (~bottom 10%) but below where body content should stop on a well-laid-out card. The 82% start line leaves room for cards that legitimately use most of their vertical space.

One concern: buffer zone hardcoded to 82-92%

The FooterCollisionBufferStart (0.82) and FooterCollisionBufferEnd (0.92) are class-level constants, not per-CardSet. The PR description explains that Rules cards use position:absolute; bottom:1.5em for the footer. If other card sets ever use a different footer positioning scheme (e.g., taller footers, relative positioning), these constants would need adjustment. This isn't a problem now — only Rules uses FooterCollision — but it's worth noting for future reference if this detector gets extended to other sets.

Edge cases handled well

• Cover image exclusion via IsCoverImage is correct and reuses the existing heuristic. The display:none footer on covers would produce massive false positives (the PR mentions +22000px artifacts), so skipping them is essential.
• The (totalPixels > 0 ? ... : 0f) guard in DetectFooterCollision is correct for very small images where the buffer zone might collapse to zero rows.
• The try/catch with silent failure (false, 0f, 0) matches the pattern used by DetectBottomSaturation — consistent with the rest of the harness.

Code quality

• Follows the existing detector pattern precisely: same tuple return style, same threshold accessor pattern, same test structure.
• The CardCheckResult extensions (IsCover, FooterCollision, FooterCollisionValue) are clean additions.
• The pivot table in the full grid report is a nice touch — it makes it easy to scan which specific cards are problematic across languages.

One minor formatting note

In the full grid's summary flagged-count filter:

r.FooterCollision.StartsWith("FLAG")

This works because the three possible values are "PASS", "FLAG(...)", and "COVER". A !r.FooterCollision.Equals("PASS") && !r.FooterCollision.Equals("COVER") would be more defensive, but StartsWith("FLAG") is readable and correct for the current values.

No security concerns — this is a test-only visual QA harness that operates on local image files.

Solid addition. The detector fills a real gap where BottomSaturation couldn't distinguish body overflow from legitimate footer content.

[clusterManager-Myia]

[Hermes] Review footer-collision detector (#412 follow-up)

Checklist: ✅ Security clean ✅ Cross-repo: none ✅ Code: clean

Analysis: New footer-collision detector measures non-white pixel density in buffer zone (82%-92% height) between card body and absolute-positioned footer. Rules threshold = 0.03 (3%). Cover images excluded. Thresholds model cleanly extended. Well-documented.

APPROVE: Clean detector extension, well-calibrated thresholds.

[clusterManager-Myia]

[Hermes] Review COMMENT LGTM

Checklist:

1. ✅ Security scan: CLEAN
2. ✅ No cross-repo impact
3. ✅ Notebooks: N/A (C# test project)
4. ✅ Fix/incident: Addresses specific detection gap (#412 follow-up, #29 overflow)

Assessment:

Well-designed footer-collision detector:

• Buffer zone (82%-92% height) correctly targets the gap between body end and absolute footer — avoids false positives from legitimate footer content that BottomSaturation couldn't distinguish.
• Rules threshold 0.03 (3%) appropriately sensitive for text overflow detection.
• Cover image exclusion (IsCoverImage → display:none footer) prevents the +22000px artifact false positive.
• Integration into FullGrid with FooterCollision column + Rules pivot table is thorough.
• Informational mode (no hard-fail) is the right call for a new detector — avoids CI breakage while collecting data.
• DetectFooterCollision() implementation is clean: pixel-by-pixel scan with SixLabors.ImageSharp, proper error handling (try/catch → false).

One note: Merge timing noted as HELD until #438 merges. Verify that dependency is tracked.

[clusterManager-Myia]

[Hermes] Review (17e8a35) — new PR, not previously reviewed.

Checklist:

1. ✅ Security scan: CLEAN
2. ✅ Cross-repo impact: None (VisualTests only)
3. N/A Notebooks
4. ✅ Fix/incident: Addresses #412 follow-up (#29 overflow recalibration)

Analysis:
New DetectFooterCollision detector targets the specific overflow pattern BottomSaturation missed:

• Scans buffer zone (82%-92% of card height) between body and absolute-positioned footer
• Rules threshold = 0.03 (3%) — appropriately sensitive for text spillover
• Cover images correctly excluded (display:none → false +22000px artifacts)
• Informational reporting (no hard-fail) — appropriate for a detector
• Follows existing detector pattern — consistent codebase style

Minor note: IsCoverImage() referenced but not in diff — assuming existing from prior work.

Verdict: LGTM — clean detector following established patterns. Proper threshold calibration and false-positive exclusion.

[clusterManager-Myia]

[Hermes] Review cycle 07:00Z — Stall 4d, new commit detected

Checklist:

1. ✅ Security scan: no secrets in diff
2. ✅ Cross-repo impact: none (VisualTests only)
3. N/A Notebooks
4. ✅ Fix/incident: #29 overflow recalibration — addresses missed footer collisions

Notes:

• New DetectFooterCollision detector: measures non-white pixel density in buffer zone (82%-92% height) between card body and absolute-positioned footer
• Rules threshold = 0.03 (3%) — appropriately sensitive for text spillover
• Cover images (variante 1) correctly excluded (IsCoverImage check)
• Test method: VisualQa_FooterCollision_Rules_NoBodyFooterOverlap — scans all Rules card images per language
• Architecture is consistent with existing detectors (BottomSaturation, BlankRatio)

Concerns:

• Stall 4d — author action required for merge
• FooterCollision only calibrated for Rules (0.03f). Virtues/Fallacies/Scenarii use default (0.10f) — confirm these don't have the same overflow pattern
• No update to CI pipeline integration — ensure new detector is included in CI runs

Decision: COMMENT (LGTM) — well-designed detector addressing a real false-negative gap. Clean implementation.

[clusterManager-Myia]

[Hermes] APPROVE — Footer-collision detector for Rules overflow (#29).

Checklist:

1. Security scan: clean.
2. Cross-repo impact: none.
3. Quality:
   • New detector follows established pattern (BlankRatio, BottomSat → FooterCollision).
   • Buffer zone (82%-92%) well-calibrated for absolute-positioned footer.
   • Rules threshold 0.03 (3%) is appropriately strict for text spillover detection.
   • Cover image exclusion (display:none footer → false positive prevention) is smart.
   • Per-CardSet thresholds + defaults follow existing architecture.
   • 148 new lines of well-documented C# test code.

LGTM. Good extension of the Visual QA harness.

[clusterManager-Myia]

Hermes Review — #441 (SHA 17e8a35, delta from NC 3dba70)

Scope: 1 commit. Adds footer-collision detector to VisualQaHarness — detects card body text overflowing into the absolute-positioned footer zone.

✅ Checklist

1. Security: CLEAN — no secrets
2. Cross-repo impact: None
3. Notebooks: N/A (C# test code)
4. Fix/incident: Follow-up to #412/#29 overflow recalibration

Findings

• Detector logic: Scans buffer zone (82%-92% of card height) for non-white pixels. Threshold: Rules=3% (strict), default=10%. Sound approach.
• Cover exclusion: Correctly skips cover images (display:none footer) to avoid false +22000px artifacts. Smart.
• Integration: Properly integrated into FullGrid report with new FooterCollision column. CardCheckResult extended.
• Code quality: Well-documented (XML doc + inline comments). Threshold system consistent with existing BottomSaturation pattern.

⚠️ Minor

• Test is informational (doesn't hard-fail on FLAG) — intentional per existing pattern, but worth noting that high-density flags won't block the build.

Verdict: COMMENT — LGTM

Clean, well-documented addition to the QA harness. No blockers.