Four Bugs in One Algorithm Design

TLDR

During the design review of our v1 bipartite column detection algorithm, four compounding mathematical edge-case bugs were identified prior to shipping to production: reversed gutter tolerance directions, viewport-frame instead of content-frame persistence checks, global band count denominator pollution, and two-pointer interval scan failures. Correcting these four geometric constraints produced a bulletproof multi-column layout engine.

Geometric Defect	Faulty Math Formulation	Structural Consequence	Corrected Formulation
Reversed Tolerance	maxX <= X + tol	Overlapping bands counted as clean split	maxX <= X - tol (Shrinks inward)
Viewport Persistence	b.y <= vpHeight 0.20	Rejects short 2-column articles	b.y <= contentTop + contentSpan 0.20
Global Denominator	/ narrowBands.length	Rejects 2-column sections on hybrid pages	/ localCoexistenceBands.length
Two-Pointer Walk	Un-merged maxX/minX lists	Misses overlapping band spans	Interval merging prior to gap search

---

Technical defect analysis & fixes

Defect 1: reversed gutter tolerance direction

• Faulty Logic: Adding tolerance outwards (maxX <= X + tol) allowed bands ending past the split center to count as "committed left," causing overlapping bands to evaluate as a clean partition.
• Correction: Subtract tolerance inward (maxX <= X - tol) to ensure committed populations never overlap spatially.

// REFACTORED: Inward Tolerance Partitioning
const leftOnly  = narrowBands.filter(b => b.maxX <= X - tol);
const rightOnly = narrowBands.filter(b => b.minX >= X + tol);

Defect 2: viewport-relative persistence scaling

Faulty Logic: Evaluating header confinement against viewportHeight 0.20 rejected valid 2-column layouts on pages where text occupied only the top half of the page.

• Correction: Measure persistence relative to the actual vertical content span (contentBottom - contentTop).

// REFACTORED: Content-Relative Persistence
const contentTop = Math.min(...narrowBands.map(b => b.y));
const contentBottom = Math.max(...narrowBands.map(b => b.y));
const contentSpan = contentBottom - contentTop || 1;
const persistenceThreshold = contentTop + contentSpan * 0.20;

Defect 3: global denominator pollution on hybrid layouts

• Faulty Logic: Dividing committed bands by total page bands (narrowBands.length) rejected 2-column sections located on pages that were 70% single-column.
• Correction: Restrict the commitment ratio denominator to the Y-range intersection of left and right populations (the Coexistence Zone).

// REFACTORED: Coexistence Zone Denominator
const coexistTop = Math.max(Math.min(...leftOnly.map(b => b.y)), Math.min(...rightOnly.map(b => b.y)));
const coexistBottom = Math.min(Math.max(...leftOnly.map(b => b.y)), Math.max(...rightOnly.map(b => b.y)));
if (coexistBottom < coexistTop) return false; // Populations are vertically disjoint
const localCoexistenceBands = narrowBands.filter(b => b.y >= coexistTop && b.y <= coexistBottom); const commitmentRatio = (leftOnly.length + rightOnly.length) / localCoexistenceBands.length;
if (commitmentRatio < 0.40) return false;

Defect 4: two-pointer interval overlap failures

• Faulty Logic: Sorting maxX and minX into independent arrays failed to track interval overlap state.
• Correction: Execute interval merging prior to candidate gap evaluation.

// REFACTORED: Interval Merging Before Candidate Search
const sortedBands = [...narrowBands].sort((a, b) => a.minX - b.minX);
const mergedSpans = [];
for (const band of sortedBands) {   if (mergedSpans.length && band.minX <= mergedSpans.at(-1).hi) {     mergedSpans.at(-1).hi = Math.max(mergedSpans.at(-1).hi, band.maxX);   } else {     mergedSpans.push({ lo: band.minX, hi: band.maxX });   } }

Rule of thumb: Evaluate persistence relative to actual content bounds rather than page frames, and calculate layout ratios within local coexistence intersections.