CC-DF-2026Q3-C04 | Version 1.0 | 12 October 2026

SOURCE: planning_unit_sensitivity.csv, unchanged from CC-DF-2026Q3-04 (5 October 2026).
geography: Panel or one of seven custom regional groups. Panel overlaps regional rows.
quarter: 2025-Q3, 2026-Q2 or 2026-Q3.
authorities: Matched authority count. This is not a complete national census.
all_approval_unit_mentions: Unit mentions in approved records, without the reserved-matters filter.
excluding_reserved_matters_unit_mentions: Mentions remaining after the upstream case-insensitive proposal-text phrase filter.
removed_unit_mentions: All minus remaining. Does not establish duplicate dwellings.
removed_pct: Removed divided by all, multiplied by 100.

DERIVED: definition_thresholds.csv and thresholds.json (same 16 comparisons).
comparison: Annual (baseline 2025-Q3) or Quarterly (baseline 2026-Q2).
baseline_quarter/current_quarter: Compared periods; current is always 2026-Q3.
baseline_filtered/current_filtered: Retained-group source mention counts.
baseline_removed/current_removed: Removed-group source mention counts.
filtered_change: Current minus baseline in the retained group (delta F).
removed_change: Current minus baseline in the removed group (delta R).
all_change: delta F + delta R.
filtered_change_pct: 100 * delta F / baseline_filtered.
all_change_pct: 100 * (delta F + delta R) / (baseline_filtered + baseline_removed).
within_range_crossing: True only if the zero-change root lies strictly between 0 and 1.
zero_change_weight_pct: 100 * (-delta F / delta R) if inside the interval; otherwise blank.
direction_at_zero/direction_at_one: Sign at the fully filtered / fully restored endpoints.

No blank threshold means zero. It means no interior crossing was found. Source baselines are positive.
The app uses T(w)=F+wR and the SAME w in both periods. Intermediate values are illustrative weighted mentions, not measured dwellings or classifier accuracy. All unrounded calculations use the integer source counts; display rounding does not change results.