The Setup
Los Angeles filed 58,014 civic complaints in the 14 days ending August 3 — 311-style reports spanning street pavement, streetlight outages, sidewalk damage, illegal dumping, tree emergencies, and landscape maintenance, logged against 1,528 distinct H3 cells across the metro. In the same 14-day window, the city issued 4,608 building permits. The complaint volume outruns the permit volume by a factor of 12.6 to 1. Cell-level scoring in Los Angeles is built entirely on the smaller number.
The Chain
Development pipeline scoring, the cell-level field that reads as a proxy for where a metro is actively building, is derived from building_permits — the 4,608-record side of this comparison. The 58,014-record side, civic complaints, does not feed cell scoring at all: the signal_scores fields computed per cell (geoConformal, epistemic, spatialSharpeRatio, sobol, monteCarlo) carry no complaint-derived input anywhere in their composition. The complaints are not noise. Filtering out the catch-all "Information-Only" and "Item Pickups" categories still leaves a maintenance-coded core — 1,997 street pavement reports, 1,765 landscape maintenance, 1,571 streetlight repairs, 1,461 obstructions, 1,431 tree emergencies, 1,253 sidewalk problems — plus 2,453 homeless encampment reports, a category with its own hostility-adjacent read on a cell. That is a denser, more current account of physical condition than a permit count, which only registers where someone has already filed paperwork to build or repair something — not where something is visibly wearing out. A cell can carry the complaint load of Item Pickups and Street Pavement Issues stacked ten deep and still show a permit-derived pipeline score of zero, because the two data streams never touch inside the scoring layer.
The Implication
Permit-based scoring measures intent to build. Complaint volume measures the condition being reported against, in near real time, at twelve times the density. A metro that is complaint-heavy and permit-light — high wear signal, low construction signal — reads identically in current cell scores to a metro with genuinely little going on, because the model has no way to see the 58,014 side of the ledger. That is a meaningfully different cell than "quiet," and the distinction only exists in a data stream the scoring pipeline does not consume.
What to Watch
Whether the complaint-to-permit ratio holds or diverges further as more metros are checked — LA, Chicago (30,720 complaints logged over the same 14 days, against Chicago's own permit count), and San Francisco (1,850) are so far the only three metros with complaint data ingested at all, out of the full tracked metro list, so this is not yet a general finding. Also worth tracking: whether any of the maintenance-coded complaint categories cluster geographically with cells already flagged for other reasons — deferred-maintenance framing elsewhere in this dataset has used permit mix and maintenance-risk framing separately; this is the first pass at complaint volume as its own line.
Limitations
Complaint-data coverage is currently three metros deep against a much longer tracked-metro list, so the 12.6x ratio in Los Angeles cannot yet be read as typical — it is one data point, not a pattern. The most recent complaint record in this window is from August 3, four days stale relative to the analysis date, which is worth flagging given the freshness emphasis elsewhere in this pipeline. "Information-Only" and "Item Pickups," the two largest categories, are not necessarily maintenance signal — they were excluded from the maintenance-coded subtotal above precisely because their content is ambiguous, and a full accounting would need a category-by-category read rather than the top-line total used here.
--- Data as of 2026-08-07. Source: `axiom_events` (event_type='complaint.filed', 'permit.issued'), Axiom Locus civic data pipeline.