Build Nexus

Measures: approved DOB NOW Build permits issued in the period — work cleared to start, and the headline volume figure of the report. Counts: rows in rbx6-tga4 whose issued_date falls inside the period. Rows with no issued_date are excluded by the filter. One job can draw several permits across trades, so this counts permits, not projects, and will exceed the number of distinct jobs.

Measures: how many DOB NOW Build job filings were submitted in the period — the pipeline's leading indicator, since filings precede permits which precede work. Counts: every row in w9ak-ipjd whose filing_date falls inside the period, with no filter on filing_status, so withdrawn and rejected filings are included; the figure measures intent to build, not approved work.

Measures: how long DOB took to turn a filing into a permit — the construction analogue of days-on-market. Counts: w9ak-ipjd rows whose first_permit_date falls inside the period, measured as first_permit_date minus filing_date. The cohort is defined by when the permit was *issued*, not when it was filed, so the figure is complete on publication rather than drifting as slow filings eventually clear. Rows missing either date, or with a negative interval (data-entry errors), are dropped. Because filings still waiting for a permit are excluded by construction, this measures the experience of jobs that got through, and understates the wait for those that never do.

Measures: net residential units a period's filings would add — the supply a residential market report would eventually see as new inventory, observed years earlier at the filing stage. Counts: proposed_dwelling_units minus existing_dwelling_units on each w9ak-ipjd row filed in the period. Both are text columns, so they are cast in Python. Negative results (demolition and conversion out of residential use) are kept and net against additions, so the figure is a net change and can legitimately be negative for a borough.

Measures: residential units that received a certificate of occupancy in the period — completions, the far end of the pipeline that filings and permits predict. Counts: number_of_dwelling_units on pkdm-hqz6 rows whose c_of_o_issuance_date falls inside the period. That column is stored as text rather than a date, so the whole dataset is pulled and the date parsed and filtered in Python; a server-side range filter on it would compare strings and return wrong rows. Temporary and amended COs are included alongside final ones, so a building can appear more than once across periods as its TCO is renewed.

Measures: DOB violations written in the period — enforcement activity, which moves independently of construction volume and is the half of the market most owners only see after it lands. Counts: rows in 3h2n-5cm9 whose issue_date falls inside the period. issue_date is compact 'YYYYMMDD' text rather than a date column, so the range filter is a lexicographic string comparison; this also excludes the dataset's malformed values (e.g. 'Y9990120'), since 'Y' sorts above any digit. The dataset carries no BBL or neighbourhood, only borough/block/lot, so borough is the finest geography available here without a PLUTO join.

Measures: the dollar cost of enforcement — penalties the Environmental Control Board imposed on violations issued in the period, plus how much of it has actually been paid. Counts: sum of penality_imposed (the dataset's own spelling) over 6bgk-3dad rows whose issue_date falls inside the period, grouped by borough code and severity. Collection rate is amount_paid divided by penality_imposed over the same rows. Penalties are assessed at issue but contested at a hearing months later, so recent periods show a low collection rate purely because the hearings have not happened yet — the rate is only meaningful on periods at least a year old.

Measures: how heavily each borough is policed relative to how much of it there is to police — the figure that lets boroughs be ranked against each other rather than merely by size. Counts: DOB violations issued in the period (3h2n-5cm9, by borough code) divided by the borough's building count from MapPLUTO (sum of numbldgs in 64uk-42ks, grouped by borocode), times 1,000. Borough is the finest grain available: 3h2n-5cm9 carries no BBL or community district, so a finer slice would require a per-lot PLUTO join across 858k rows. PLUTO updates a few times a year, so the denominator is effectively constant within a report period.

Measures: which specific violation codes the city wrote most often in the period — the practical question an owner or GC actually asks, and the table most likely to be quoted. Counts: top violation_type_code values from 3h2n-5cm9 and top infraction_code1 values from 6bgk-3dad, each over rows issued in the period. The two are reported separately rather than merged because they are different code systems — a DOB violation type and an ECB infraction code are not comparable and must not be summed.

Measures: which specific violation codes the city wrote most often in the period — the practical question an owner or GC actually asks, and the table most likely to be quoted. Counts: top violation_type_code values from 3h2n-5cm9 and top infraction_code1 values from 6bgk-3dad, each over rows issued in the period. The two are reported separately rather than merged because they are different code systems — a DOB violation type and an ECB infraction code are not comparable and must not be summed.