The guard that tested my assumption instead of the code

TLDR

Writing hardcoded numeric assertions in regression guards (e.g. asserting expectLinks: 0 on single-column documents) causes false test failures when algorithms correctly link paragraph continuations across page zone boundaries. Replacing brittle counts with property-based invariants (asserting that zero links touch digit-heavy table fragments) validates algorithmic correctness without locking tests to arbitrary totals.

Guard Assertion Type	Assertion Logic	Resiliency to Valid System Improvements
Count-Based Assertion	expect(links.length).toBe(0)	Fragile (Fails when zone linkers improve)
Property-Based Invariant	links.every(l => !l.isDigitHeavy)	100% reliable across feature updates

---

Technical property invariant test assertion

// Property-Based Invariant Assertion for Paragraph Continuation Linkers
export function verifyContinuationLinkInvariants(detectedLinks) {
  for (const link of detectedLinks) {
    // Invariant 1: Continuation links must never connect numeric table debris
    const hasDigitDebris = isDigitHeavy(link.prevFragment) || isDigitHeavy(link.nextFragment);
    if (hasDigitDebris) {
      throw new Error([InvariantViolation] Continuation link ${link.id} connects numeric table debris);
    }
// Invariant 2: Linked fragments must share font family compatibility     if (link.prevFragment.fontFamily !== link.nextFragment.fontFamily) {       throw new Error([InvariantViolation] Font family mismatch across continuation link ${link.id});     }   }   return true; }

Rule of thumb: Assert structural property invariants rather than exact numeric result counts when testing heuristic extraction algorithms.