SMVaddadi/propensity-data-ct-doi-complaints

ct_doi_complaints — mirror

A convenience mirror of a public dataset, pushed by propensity.data.hf_mirror so that benchmark runs do not depend on a flaky upstream origin.

This is a mirror, not a grant of rights

Copyright and licence remain with the original source named below. Mirroring moves bytes; it does not and cannot change the terms they arrived under. If you intend to redistribute this further, or to use it commercially, read the original terms — not this card, and not the license: field above, which is our best reading and may be wrong.

Provenance

• Source: Insurance Company Complaints, Resolutions, Status, and Recoveries -- Connecticut Insurance Department, via data.ct.gov view t64r-mt64. Attribution: 'Department of Insurance' (the publisher's own string; it travels as a courtesy, not a licence condition). Declared as other + ct-open-data-public-domain, NOT cc0-1.0 (CC0 is a waiver only a rights holder can execute and this repository is not one) and NOT us-public-domain (17 U.S.C. 105 covers FEDERAL works only and a US state may hold copyright). The basis is Connecticut's own written open-data policy, which calls the portal's Public Domain field an act of DEDICATION, plus CGS 4-67o defining open data as data that 'does not have restrictions governing use'. The portal's actual Terms of Use (data.ct.gov/terms) is purely a warranty disclaimer and reserves no rights. WHAT THIS IS: 90,138 rows x 12 published columns -- every consumer complaint against an insurer licensed in Connecticut, with the regulator's adjudication and the dollars recovered. 25,566 distinct complaints over 822 insurers, opened 2015-11-02 to 2026-10-29. Five candidate outcomes: disposition (14 values), conclusion (54), recovery (continuous USD), status (27), days_open. WHY IT IS HERE: the only complaint-grain, prose-licensed, anonymously downloadable insurance asset the sweeps found in North America -- nearly everything else in the lane is an insurer-year aggregate. A monetary severity target on an individual complaint is rare. FIVE MEASURED CAVEATS. (1) A ROW IS A COMPLAINT LINE, NOT A COMPLAINT, and this changes every headline number. file_no has 25,025 distinct values over 90,138 rows and 497 are reused by up to 7 different companies, so it is a per-company filing number, not a key; the complaint is (company, file_no) -- 25,566 groups, median 2 lines, max 108, and 46.1% have exactly one line, which is why the duplication is easy to miss on a sample. recovery, status, closed and opened are CONSTANT across every line of a complaint, so a row-wise recovery.sum() returns $161,582,427.73 against a true $27,696,271.96 -- a 5.83x overstatement. complaint_key is the real key, and is_complaint_primary_line / complaint_line_count are provided. (2) 'Temporal Coverage: 2015-Present' is true and useless: 2015-2020 is 313 of 90,138 rows (0.35%) and the series effectively begins 2022. (3) status is a CENSORING indicator, not an outcome -- 87,789 rows read Closed and the label space collapses with cohort age (2 distinct statuses in the 2024 cohort, 27 in 2026), so predicting it predicts recency. days_open is missing not at random. (4) The two adjudication taxonomies are mostly null -- disposition on 42.9% of rows, conclusion on 27.6% -- and they are DIFFERENT taxonomies, not refinements: disposition is the regulator's finding, conclusion the mechanism. reason has only FOUR values; the resolution is in subreason's 205. (5) The entity key is a NAME with no NAIC code anywhere in the view, so it does not join out without a fuzzy match -- though the strings are hygienic (0 collisions after case/punctuation normalisation, no placeholders, no aggregates). NO PERSONAL DATA: the complainant is represented by no column -- no name, address, policy number, age or narrative. All 822 company values are corporate or fraternal insurers and the longest text value is 71 characters. WHAT IT IS NOT: not a claims file (no amount claimed, no policy, no coverage limit), not a complaint RATE (no exposure denominator, so a large insurer's count is not comparable to a small one's), and no target column -- nothing published says which disposition counts as a win.
• Origin URL: https://data.ct.gov/d/t64r-mt64
• Loader: propensity.data.public_loaders.load_ct_doi_complaints
• Extract: full pull from origin
• Reported source: ct_doi_complaints
• Rows pushed: 90,138
• Columns: 18

Licence and redistribution

• Licence as found: other (ct-open-data-public-domain)
• Redistribution verdict: open
• Basis: Connecticut Insurance Department, public domain by Connecticut's own open-data policy. THE PORTAL FIELD IS NOT THE BASIS: data.ct.gov view t64r-mt64 carries licenseId PUBLIC_DOMAIN / license {"name": "Public Domain"} / attribution "Department of Insurance" / provenance official, but US STATE works are not public domain by statute -- 17 U.S.C. 105 reaches only works of the UNITED STATES Government -- so that dropdown is an agency assertion, and this table's audits overturned seven rows for citing a portal field in place of the rights holder. WHAT RAISES IT ABOVE THE FIELD is written Connecticut policy that names the field as an act of dedication. CT Office of Policy and Management, Open Data Publication Guidelines, https://portal.ct.gov/-/media/ct-data/data-publication-guidelines-v1.pdf (HTTP 200, 940,859 bytes, 11 pages, read and re-verified 2026-11-10), verbatim: "Explicitly identify the license of your data. Generally, data should be identified as 'Public Domain.' [...] If applicable law prevents you from DEDICATING your dataset as public domain, you may have to apply for an Open Data license, but first contact the Chief Data Officer." That sentence is load-bearing twice: it calls the field a DEDICATION -- the operative legal act, since no statute performs one for a state -- and it supplies the documented exception path for when law prevents one, so a view carrying Public Domain is an agency that FOLLOWED the policy rather than accepted a default. The same document opens "Open data is data that can be freely used, reused, and REDISTRIBUTED without legal or financial restrictions" and quotes CGS 4-67o, under which open data "(C) does not have restrictions governing use". AND NOTHING RESTRICTS IT. The portal's real Terms of Use is https://data.ct.gov/terms (HTTP 200, 604,490 bytes) and is entirely a warranty/liability disclaimer -- it reserves no rights and says nothing about copying. CT.gov's disclaimer reserves copyright in exactly two things, the CT.gov logo and the state seal under CGS 3-106a, a carve-out structure that would be pointless if everything else were reserved. 17 of 17 sibling data.ct.gov datasets carry the same value, consistent with the policy. A CORRECTION RECORDED ON PURPOSE: this entry was first assessed UNVERIFIED on the premise that "no prose licence exists anywhere findable". That premise was FALSE and the probes were simply at the wrong paths -- /about, /terms-of-use and /dataset/terms-of-use are all HTTP 404, and portal.ct.gov/opendata serves a 404 page under HTTP 200 (a soft 404). The footer links /terms and portal.ct.gov/datapolicy, and prose exists at both. A 404 on a guessed path is not evidence of absence. RESIDUAL WEAKNESS, stated so a stricter reader can downgrade: the prose is OPM's (the Chief Data Officer's unit, which owns the policy, the statutory duty and the portal), not the Insurance Department's own page. The chain is 'state policy directs a dedication -> the rights-holding agency recorded one', not 'the rights holder wrote a grant'. One line from CID or the CT Chief Data Officer would settle it outright. NO PERSONAL DATA, checked rather than assumed: the complainant is represented by no column -- no name, address, policy number, age or narrative. All 822 company values are corporate or fraternal insurers, and the longest value in any text column is 71 characters.

Citation

Connecticut Insurance Department, Insurance Company Complaints, Resolutions, Status, and Recoveries, https://data.ct.gov/d/t64r-mt64, accessed 10 November 2026. Public domain; attribution: Department of Insurance.

Loading

from propensity.data.hf_mirror import load_mirrored

result = load_mirrored("ct_doi_complaints", org="SMVaddadi")
result.frame.head()

load_mirrored falls back to the original loader (origin, then its own offline replica) if this mirror is unreachable, so it is safe to depend on.