Data¶
findings.csv¶
Every finding of mass surveillance technology this project has found in Connecticut municipal records, in one file. One row per mention.
7,004 rows · 101 towns · 18 topics · 6,647 from meeting transcripts, 357 from agenda and minutes documents.
Those counts are rewritten from the file itself every time it is published, so they cannot drift away from what you actually download.
Raw URL, stable across regenerations:
This is the complete record. The topic reports are the readable version of the same data, but they cap receipts at 12 per town so the pages stay renderable — so they show roughly a quarter of these rows.
Columns¶
| Column | Meaning |
|---|---|
town |
Connecticut municipality the record belongs to |
source_kind |
transcript (spoken, from meeting captions) or agenda (written, from an agenda or minutes document) |
topic |
Technology category, e.g. alpr, drone, rtcc, facial_recognition |
term |
The specific search term that matched |
meeting_date |
Date of the meeting or document. Empty where it could not be determined |
meeting_body |
Council, commission or committee, where identified |
item_title |
Title of the source recording or document |
url |
Link to the exact second of the meeting video, or to the document. Present on every row |
start_seconds |
Offset within the recording. Empty for agenda rows, which have no timestamp |
context |
The surrounding passage, up to 300 characters |
source_kind is worth reading before you aggregate anything: a transcript row
is somebody saying a word out loud, an agenda row is a word printed in the
town's own paper. They are not interchangeable evidence, and mixing them
silently inflates a "mentions" count.
What a row is, and is not¶
A row is a place in a public record where a search term appeared. It is not a verified claim that the town uses the technology.
Keywords are ambiguous — "drone" can mean a bee, "flock" a bird — and automatic
captions mis-transcribe. Genuine false positives are in this file: one row is a
predictive_policing term matching a Zoning Board of Appeals discussion about a
car repair shop. Every row carries a url for exactly this reason. Follow it
and listen before citing anything.
A town's absence, or a low count, is not evidence it has avoided surveillance technology. It far more often means that town's meeting backlog has not been downloaded yet, its meetings are not captioned, or its video archive is on a platform this pipeline cannot read. Per-town coverage is in the cross-town matrix; read it before drawing conclusions from what is missing.
Redaction¶
Connecticut public comment opens with a ritual: the speaker gives their name, then their home address, for the record.
The names are here as spoken. Addressing a public body is a public, accountable act, and stripping names would gut the record.
The home addresses are not. Where a speaker gives their own residence, the
address is replaced with [address redacted]. Individually those addresses are
public and sit on the town's own video. But this file is a different artifact
from a four-hour video nobody watches: it is searchable, sortable and
downloadable in bulk, and several thousand name-and-home-address pairs in that
shape is the same aggregation this project objects to when a camera vendor
builds it. Some of the speakers are children.
Addresses being discussed rather than lived at — where a camera sits, which
property is before the commission — are untouched, because that is the evidence.
So is I live in West Hartford, which is how a speaker establishes standing.
The town's own recording remains the unredacted record of the same meeting.
Source¶
The transcripts behind every row are published as
release assets
— one .zip per town, each with a MANIFEST.txt stating its own coverage and
redaction count.