Q3 PLANNING SENSITIVITY EXPLORER
CC-DF-2026Q3-C01 | Version 1.0 | 8 October 2026
Author: Matt Lenzie / Construction Capital

REPRODUCTION
Save planning_decisions.csv, reproduce.py and LICENSE.txt together.
Run: python3 reproduce.py
Python 3.9 or newer, standard library only. This writes sensitivity_bounds.csv
and data.json in the same directory. No credentials or private repository needed.
It does not recreate the proprietary upstream extraction process.

SOURCE
ICPSR study 317061 V1, https://doi.org/10.3886/ICPSR317061.V1
Frozen aggregate edition 5 October 2026; source_manifest.json records its hash.
planning_decisions.csv is the unchanged 24-row study input table. This explorer
uses Q3 2025 and Q3 2026 (16 rows); Q2 2026 is retained for source fidelity.
Panel is the total of seven custom regional groups, not an additional region.
No individual lenders, borrowers, addresses or planning applications are exposed.

DERIVED CSV FIELDS
geography: Panel or custom regional group.
authorities: Number of authorities, identical in both compared quarters.
change: Q3 2026 minus Q3 2025 known-outcome approval rate, percentage points.
minimum_change: new.lower minus old.upper, percentage points.
maximum_change: new.upper minus old.lower, percentage points.
direction: Always falls if maximum_change<0; Always rises if minimum_change>0;
otherwise Can change direction (all displayed bounds here are strictly away
from zero or cross zero; no endpoint is exactly zero).
old_*: Q3 2025; new_*: Q3 2026.
year: Calendar year; approved/refused/unknown: recorded counts.
rate: 100*approved/(approved+refused).
lower: 100*approved/(approved+refused+unknown).
upper: 100*(approved+unknown)/(approved+refused+unknown).

SOURCE CSV FIELDS
geography, quarter, authorities: as above and calendar quarter.
received: applications received, a separate flow from decisions.
approved, refused: recorded outcome counts.
withdrawn: withdrawals, excluded from every rate in this explorer.
unknown_outcome: recorded dated decisions without a classified outcome.
known_decisions: approved+refused.
approval_pct: rate; unknown_all_refused_pct: lower; unknown_all_approved_pct: upper.

LIMITS AND RIGHTS
Matched partial panel, not a national planning census. Unknowns are recorded
outcomes, not missing applications or uncaptured decisions. The bounds test
assignments only, not uncertainty probabilities, capture errors or selection
bias. Counts do not establish consent probability for a project or loan access.
Original research text: CC BY 4.0. Original code: MIT. Planning-source rights
reserved to their holders. Read LICENSE.txt before reuse.