Both Sides of the Discount Moved
A seven-source pricing monitor covering 9,077 facilities, an August rate-hike anomaly, and a pre-registered bet on Public Storage's (PSA) Q3 earnings.
9,077facilities tracked across seven collections
129,704advertised unit/offer records daily
4,904 / 4,904listings with simultaneous rate hikes on August 1, measured by its predecessor FindStorage
- Pre-registered Q3 prediction: Sealed before Q3 results exist, predicting deceleration in promotional discount growth and new-customer contract rates below +1.6% YoY, with Q4 rates flat or negative.
- The August 1st event: 4,904 listings moved into "40% off For 4 Months" and every single one received an immediate reference-rate increase (median +$45), while a July control proved rate and promo are independently writable.
- The living ledger: On September 18, the September 14 rate reset was partially walked back with 12% restored across 22,772 units (0.9800 matched-panel index), observed and logged in real time.
The Verification Filter
Turning 57,032 civic observations across 30,587 locations into 4 verifiable commercial opportunities without hallucinatory leads or naive address matching.
57,032observations across 30,587 locations in database
21municipal & county sources mapped
37 → 10 → 4strict pipeline reconciliation: 4 prospect, 4 review, 2 excluded
- The naive scraping myth: Building permits name general contractors rather than tenant businesses, strip-mall addresses collide across dozens of suites, and speculative permits carry no operating tenant.
- Tyler EnerGov reverse engineering: Extracted undocumented multi-tenant API headers (Olathe vs Overland Park vs KCMO) and discovered Overland Park houses commercial activity in Plans while Olathe uses Permits.
- Silent failures caught: Detected portals enforcing silent 10,000-row query truncation caps without error responses, and engineered automated run-audit reconciliation rails.
Room or Radio?
A Wi-Fi CSI system identified devices at 99.7% accuracy. Physical controls broke the device-identity claim—and revealed a physical link state instead.
69,510,186clean beacon frames replayed deterministically
29 / 35distinct ideas / rejection tests carried through 8 paired nights
0.2σ vs 19.6σco-located boards (0.2σ) vs same board across rooms (19.6σ)
- The seductive number: 99.7% chronological holdout accuracy identifying three ESP32 transmitters using subcarrier phase slope, originally attributed to quartz crystal frequency offsets.
- The hypothesis ledger: 29 distinct ideas systematically killed across 35 rejection tests, proving what the feature wasn't before claiming what it was.
- The double difference: Two receivers watching the same transmitter pair in the same instant disagreed by 6.79 standard deviations—proving the measurement is a property of the link, not the radio.