VectorGIS

Not a count but a roster, each development with its address, unit mix, who it serves and its AMI cap: The Breezewood (33 units, families), Valley Senior Village (50, seniors 62+) and Veterans Village of San Diego (54, veterans). It also says plainly that this is the regional inventory's record of Escondido and may not be every affordable unit in the city.

This dataset codes the same category two ways. Property Damage Only appears on 191,012 records and PDO on another 20,630. Matching only the spelled-out value returns a number that looks entirely reasonable and quietly drops 20,630 real collisions. It counts both.

Carlsbad is a jurisdiction with no descriptive coverage in the dataset at all; there is no document to retrieve and no text to match. The address channel resolves it directly off the parcel record.

Two questions in one sentence: where they are, and what the designation actually permits; both answered from the same record. The allowance is stored as a literal string in the data and has to be matched exactly, not paraphrased.

A whole-network total computed from the geometry at the moment you ask, not read off a spreadsheet someone exported last quarter. The kind of number that normally means a request to the GIS team.

The answer also says out loud that this is a regional figure covering all of San Diego County, not any single city, because the most likely way to misuse this number is to quote it as one jurisdiction's total.

The dataset spells this class Bike Boulevard. Ask for "bike boulevards" and a system that matches on the phrase you typed returns nothing at all. It matches the stored value, not the question.

A street address in, a full assessment breakdown out. No APN to look up first, no parcel layer to know the name of, and the land and improvement components come back split rather than as a single opaque total.

A spatial join the user never has to name: housing points against transit priority polygons, with the denominator carried along so the number means something. Nothing in the question mentions a layer, a join, or a buffer.

Five classes, with the segment count behind every line, returned as a table in one pass. This is the shape of a capital-planning question: not "how much of X do we have" but "how is X distributed". It is also the shape that usually means exporting the layer and building a pivot table.

729 of the region's 736 tracts have income data and seven do not. Averaging the missing ones in as zeros would drag the floor down and nobody would notice. The answer states the denominator it used.

The reverse of the question above: a parcel number in, the address and assessment out. The same parcel carries two address points (1305 and 1319 Tamarack), and the answer surfaces both rather than silently picking one.

The table has 125 rows for those 40 routes, because a route is stored as several shape geometries. Counting rows gives 125 and looks perfectly plausible. It counts distinct routes.

The wetlands layer mixes land wetlands with open water. The category that sounds closest to what was asked, Estuarine and Marine Deepwater, is 2.4 million acres of open ocean and bay, roughly 365 times larger. Summing the layer and calling it "wetlands" produces a number that is enormously wrong and looks authoritative. The answer breaks the categories apart and says which one is the habitat figure.

Two defensible readings of this question exist in the data: count the collisions flagged as involving a pedestrian fatality, or add up the pedestrian fatalities recorded across all collisions. The second is the one that answers what was actually asked, because a single collision can involve more than one fatality.

A city boundary against a watershed layer, with the answer being a list rather than a number. The kind of overlay question that normally means a request ticket and a two-day wait.

Two numbers exist for this parcel and they do not match. Rather than pick one and present it as the answer, it reports the assessor's recorded acreage as the figure of record and the computed geometry area beside it. Which one is correct depends on what you need it for, and that is the user's call to make, not the software's.

A containment question across two unrelated layers, assessor parcels and FEMA flood zones, with no precomputed join between them. It also volunteers that parcel boundaries come from a tax-assessment fabric rather than a legal survey, which is the caveat that matters if anyone acts on this number.

There is no layer called "schools." The schools live inside a general places layer as one type among many, so answering this means knowing what is in the data rather than matching the word in the question to a layer name. "No" would sound honest here and would be wrong.

The straight version of the question, asked without a false premise attached. It is here so the correction in the next question has something to stand on: the same number, arrived at the same way, whether or not the question tries to push it somewhere else.

The question contains a false premise stated with confidence, which is exactly the prompt most systems will agree with. It corrects the number instead, and it does not quietly fall back to a text lookup to do it.