Teardown Signal Backtest

Our teardown signal concentrates real redevelopment at 2.95× the base rate

A leakage-safe retrospective backtest: across 1,401,779 Maricopa parcels scored only from time-stable structural attributes as of 2024-06-01, the top-scored decile went on to be demolished or rebuilt at 0.133% — 2.95× the 0.045% base rate. The score inputs predate the outcomes and cannot encode them.

Top-decile lift
2.95×
the top-scored decile was redeveloped at 0.133% vs. a 0.045% base rate
Parcels backtested
1,401,779
Maricopa parcels scored from time-stable attributes as of 2024-06-01
Outcomes graded (live cohort)
0
redevelopment permits matched on the live frozen cohort so far (0 via permits) — early, not yet validated

Redevelopment rate by teardown-score decile

Each bar is the share of parcels in that score decile that were actually redeveloped within the window. We render the real shape, unsmoothed: several middle deciles fall below the base rate — the honest signal is the sharp lift concentrated in the top decile (D10), not a clean monotone ramp.

base rate 0.045%Decile 1: 0.045%D1Decile 2: 0.053%D2Decile 3: 0.012%D3Decile 4: 0.014%D4Decile 5: 0.019%D5Decile 6: 0.022%D6Decile 7: 0.037%D7Decile 8: 0.050%D8Decile 9: 0.066%D9Decile 10: 0.133%D10

What this measures

Outcome: city demolition or new-SFR permit issued after T.

Anchor date (T): 2024-06-01 · Outcome window: (2024-06-01, 2026-06-25 00:00:00].

Signal: structural teardown score (building age + building-to-lot coverage), leakage-safe: time-stable attributes only, no post-T data.

Leakage note: Retrospective validation of the SIGNAL — distinct from the live frozen score-at-first-seen cohort. Score inputs predate and cannot encode the outcome; outcomes are strictly after T.

The live cohort (separate from this backtest)

accumulating — value loop engages when sales land. The backtest above validates the signal retrospectively. Separately, every week we freeze the exact score we showed at first-seen and grade it against real recorded county permits and sales as they land — an immutable, accumulating track record.

9,616
predictions frozen at first-seen, immutably
9
counties accumulating outcomes
0
outcomes graded vs. county reality so far

Honest caveat

This study validates the signal retrospectively; it is not a claim that the live frozen cohort has graded outcomes — it has 0 so far (0 redevelopment permits matched on the live cohort, with the county sale feed lagging first-seen by months). No value self-improvement is claimed until 100 realized sales land.

Permits grade as they're issued (0 redevelopment permits matched on the cohort so far — live, within weeks). Value self-improvement waits on realized SALES (0 so far; the county sale feed lags first-seen by months) and engages at 100 realized sales. The immutable record is real and accumulating; no value recalibration is claimed until then.

Companion report: the Phoenix Teardown Index → (the city-by-city census this signal scores). Machine-readable: /api/flywheel.