Brooklyn — top redevelopment candidates
1000 parcels ranked by calibrated probability of new-building permit issuance. Click any row or marker for the full feature breakdown and the per-parcel reasoning.
Sign in to view Brooklyn parcel intelligence
The parcel-level workspace surfaces address, BBL, recent sale prices, owner information, FAR utilization, LPC constraints, and per-parcel "why this scored high" reasoning derived from the underlying public data. We gate it behind a free account so we can rate-limit fairly and ensure the data is being used responsibly — these are real properties owned by real people, not anonymized fixtures.
- 1000 ranked candidates with full feature breakdown
- Borough-wide map with rank-coded markers
- CSV export of any sort + filter combination
Free plan includes 5 CityLens runs per month. Parcel intelligence views don't consume run quota.
How the ranking works
Every NYC tax lot (~858k parcels) is scored by a calibrated gradient-boosted classifier trained on PLUTO 2018 features against actual New-Building permits filed 2019-2024. The model emits a per-parcel probability that reflects how much each parcel looks like a known redevelopment.
- Data sources
- PLUTO for lot/zoning, DOB legacy + DOB NOW for prior permit activity (block rollups + recency), LPC gpmc-yuvp + ncre-qhxs for landmark / historic-district constraints, ACRIS for deed history (last sale price + years held). All datasets are official NYC OpenData; the audit log is in the parcel-intel repo.
- Honesty guardrails
- Temporal holdout:features frozen at PLUTO 2018, labels taken from 2019-2024 NB filings — so the model can't see post-event ground truth. Year-built bucketed at 2010 to prevent redevelopment leakage. Stratified hard-negative sampling (5× positives) so the 770k easy negatives don't swamp training. Isotonic post-calibration on a temporal validation slice so the probabilities are monotone and well-ordered, not just relative ranks.
- What a high score means
- P@100 = 0.92: 92 of every 100 top-scored parcels actually received an NB permit in the 2019-2024 holdout window. AUC = 0.98 in aggregate. A high score is nota guarantee — it's "this parcel looks structurally similar to the redevelopments we've seen." Verify against current LPC status, ACRIS owner chain, and DOB filings before acting.
Caveats.Some "parcels" in PLUTO are administrative entities (transit ROW, condo billing units, institutional campuses) and don't represent realistic redev sites. We filter these via lot-area bands and land-use exclusions, but a few slip through. The per-parcel reasoning is rule-based, not a SHAP-style attribution — it surfaces the strongest features the model actually consumed, not a literal coefficient breakdown.