Who Audits the Estimate
Short answer: Nobody has the right to audit it. The number on a collision estimate comes out of a commercial database, and no American instrument we located gives anyone the right to inspect that database, test it, or make its owner answer for it. That is not the same as saying it never gets tested. It does get tested, by the repair industry that pays for the testing itself, and the vendors answer when they choose to. Both of those are voluntary, and neither is a right.
The claim about the right is an absence claim, so the rest of this page is mostly about what we actually read and where the search stopped.
A note on quotations. Everything in quotation marks below was re-fetched from its primary source on 2026-09-02 and matched against the source text before it was published here. Curly quotation marks in a source are rendered as straight ASCII quotes. An em dash inside a quotation is rendered as a spaced hyphen. Nothing inside a quotation is added, removed or reordered, and where a source contains a defect it is quoted as found and the defect named. Where a document prints the same testimony twice in differing words, and we could not pin which version a sentence came from, we cut the quotation and gave the substance instead.
The finding: one regulator, one product, two opposite answers
California is the only state we found whose rules name the collision estimating software as such and attach legal consequences to its output. It does so twice, in the same chapter, pointing in opposite directions.
First, the software is a floor the insurer may not write below. From 10 CCR 2695.8(f):
"An insurer shall not prepare an estimate that deviates from the standards, costs, and/or guidelines provided by the third-party automobile collision repair estimating software used by the insurer to prepare the estimate, if such deviation would result in an estimate that would not allow for repairs to be made in accordance with accepted trade standards for good and workmanlike automotive repairs by an auto body repair shop, as described in this subdivision."
The conditional clause at the end is doing real work and the sentence means much less without it. It bites only where the deviation would drop the estimate below the workmanlike-repair standard that the same subdivision sets, which requires an estimate "of an amount that will allow for repairs to be made in accordance with accepted trade standards for good and workmanlike automotive repairs by an "auto body repair shop" as defined in section 9889.51 of the Business and Professions Code".
Second, the same software is inadmissible as evidence of what labour costs. From 10 CCR 2695.81(d)(7), the rule governing the Standardized Auto Body Repair Labor Rate Survey:
"(7) Only direct responses provided by an auto repair shop on a survey questionnaire that complies with subdivision (d)(9) of this section shall be used in a Standardized Labor Rate Survey. Labor rates obtained from the following sources shall not be used in a Standardized Labor Rate Survey: (A) Estimates written by the insurer to estimate repair claims, (B) Third-party automobile collision repair estimating software used by the insurer to prepare estimates, (C) Subrogation reimbursement, or (D) Any source other than direct responses provided by an auto repair shop on a survey questionnaire."
Put the two together. In one subdivision the platform's standards, costs and guidelines are a floor the insurer may not go below. Four subdivisions later in the same chapter, the platform's labour rates are barred outright, and the only admissible evidence of what labour costs is a shop's own written answer on a questionnaire. The same regulator treats the same product as authoritative for what a repair requires and as unreliable for what a repair is worth.
Neither rule audits the database. One defers to it and one excludes it. In neither place does California ask whether the software is right.
The counterweight, in the same breath. "Nobody can look at it" would be too strong, and the same regulation is where it fails. 10 CCR 2695.8(b)(4), in the total loss provisions, says: "The insurer shall take reasonable steps to verify that the determination of the cost of a comparable vehicle is accurate and representative of the market value of a comparable automobile in the local market area. Upon its request, the department shall have access to all records, data, computer programs, or any other information used by the insurer or any other source to determine market value." That is a real access right, it reaches computer programs, and it reaches past the insurer to "any other source". Its limit is its subject: it is about total loss market value, not about the parts and labour estimating database, and it is exercisable by a department, not by the shop or the owner holding the estimate.
The stronger of the two sits in the labour rate survey rules, at 10 CCR 2698.91(h):
"Upon request, the Department shall have access to all survey responses, other survey records, data, computer programs, or any other information used by the insurer or any other source to develop the survey and/or determine geographic area labor rate information. The fact that information, data, or records used or relied upon by a licensee are obtained through a third-party source shall not absolve the licensee of its responsibility to provide the Department with the requested information. The licensee shall ensure that any contracts it enters into with a third party to conduct these surveys does not conflict with this requirement. This information will be held confidential pursuant to the provisions of subdivision (f) of Section 2695.81. All survey records, questionnaires, and all other information regarding the survey shall be maintained by the insurer for a minimum of five years."
("any contracts it enters into ... does not conflict" is the grammar of the regulation as published. It reads the same way in the Department's own final text of the regulation and in Cornell's reproduction, both read on 2026-09-02.)
That third sentence is the closest thing to a mandated vendor-contract audit term we located in American auto physical damage regulation. Note how far it actually reaches. It binds the licensee, meaning the insurer, and through the insurer's contract to a third party's material. It applies to labour rate surveys. It is not a licence, not a registration and not an examination of the vendor, and it puts no duty on the vendor toward anyone. A vendor that refused would put its customer in breach, not itself.
Scope of the claim above. We read in full text, on 2026-09-02, California 10 CCR 2695.8, 2695.81 and 2698.91, and Washington WAC 284-30-391 and 284-30-392. We did not run a fifty-state survey and we did not search any state's bulletins, circular letters or enforcement records. "The only state we found" means the only one inside that boundary. We are not claiming that no other state names the collision estimating software in its rules. Five instruments in two states is the whole search, and a bulletin or a circular letter in a state we never opened would not have reached us.
Why the survey rule is weaker than it looks
Of the five instruments we read in full, the California labour rate survey is by a wide margin the most detailed methodology written for a collision claim input. We did not read the other forty-eight states' rules, so this is a comparison within that boundary and not a national ranking.
The survey must go to every registered shop, under 2695.81(d)(2): "an insurer must send a survey questionnaire to all auto body repair shops registered with, or licensed by, the Bureau of Automotive Repair as an auto body and/or paint shop". The prevailing rate is defined arithmetically at (d)(5) as "the labor rate at or below which a simple majority of surveyed shops charge in a specific Geographic Area". Direct repair program discounts are excluded at (d)(6): "No Standardized Labor Rate Survey shall use any discounted rate negotiated or contracted for with members of the insurer's Direct Repair Program, or any other Direct Repair Program".
Four limits travel with all of that.
It is optional. 10 CCR 2698.91(j): "Nothing in this section shall be construed to require an insurer to conduct an auto body repair labor rate survey." The opening recital of 2695.81 says the Commissioner "recommends insurers use" the standardized survey "if the insurer elects to use a survey".
Compliance buys a presumption, not a rate. Under 2695.81(c), a complying survey "shall result in a rebuttable presumption by the Commissioner that the insurer has attempted in good faith to effectuate a fair and equitable" labour rate component of a settlement.
Submission is not approval. 10 CCR 2698.91(k): "The fact that an insurer has submitted a labor rate survey pursuant to this section shall not be construed to indicate, and an insurer shall not represent, that the Department has approved the survey for use in settling and paying claims or that the Department has made a determination that the survey data is accurate and reliable."
Most of it is confidential. The regulation splits the survey into public and non-public halves: 2698.91(d)(1) lists the data that "will be made available to the public upon request", and 2698.91(g) lists the "non-public information" that must be filed separately from it. The individual shop rates, the questionnaire and the definitions of the geographic areas are all in the non-public list. One public item is worth marking: 2698.91(d)(1)(F) requires "A description of the formula or method the insurer used to calculate or determine the specific prevailing auto body rate reported for each specific geographic area." The one claim-input methodology in these rules that must be published is the insurer's labour rate formula, and it is the insurer's formula, not the vendor's.
Why there is no lever: the enforcement tool runs through the rate filing
The only vendor-oversight framework American insurance regulators have drafted is the NAIC's Third-Party Data and Models (H) Working Group framework for property and casualty pricing and underwriting. Its materials for the Summer National Meeting of 2026-08-12 were read on 2026-09-02.
The framework's sanction is this:
"Failure on the part of a third-party data or model vendor to provide a regulator with requested data and/or model information may result in the regulator prohibiting insurers from using the data and/or model for pricing or underwriting any property and casualty insurance product offered in the state."
And the Working Group's minutes record why a voluntary registry was thought to work at all:
"Some members supported a voluntary process, as it would not require legislative changes and could start faster. States could still effectively compel participation by conditioning the use of vendor models on registration status. Members noted that the enforcement of registration requirements may not require legislation in many states, as regulators could simply decline to approve rate filings that rely on unregistered vendors. Several states indicated that they already apply this approach informally."
That is the whole enforcement theory, and it runs through the rate filing. A regulator makes a vendor answer by threatening to refuse the carrier's rates.
There is no rate filing for a collision repair estimate. A parts price and a labour time are not rating factors; they are claim payment inputs. Nothing is filed with any department before an estimating database is used to write an estimate, so there is nothing for a regulator to refuse. We do not claim that no regulator has ever obtained data from an estimating vendor. We located no instrument giving one the right to demand it, and a request a vendor chose to answer would leave no trace in the instruments we read.
This is why the framework stops where it stops, and the minutes say so without hostility to claims oversight:
"He said the current draft represents phase one of a broader effort, as the Working Group originally considered a broader scope, including claims handling, utilization review, marketing, and fraud detection."
Claims was dropped because pricing and underwriting is where a filing process already exists to hang oversight on. That is exactly what makes claims the hard case.
The framework was not adopted on 2026-08-12. The meeting summary for that date records the Working Group adopting its Spring National Meeting minutes and its July 16 minutes, and discussing the draft framework, which it "exposed July 8 for a 28-day exposure period ending Aug. 5". The framework remains an exposed draft and is not law.
The counterweight. The framework is explicit that it does not move responsibility off the insurer: "This framework maintains the longstanding principle that insurers are responsible for their own compliance obligations." The regulators' theory is that the licensee is accountable for what it buys. That theory is coherent, and it is the same theory Washington's total loss rule embodies. And California's deference in 2695.8(f) exists to stop insurers writing below a proper repair. In that rule it is the estimating database, not the regulator, that gives shops and consumers their protection.
Why no existing licence category catches these vendors
The framework's registry design contemplates that a vendor might already sit inside an existing licensed category. In California, the category that looks closest is the insurance-support organisation, defined in Insurance Code section 791.02(l), read on 2026-09-02:
"(l) "Insurance-support organization" means: (1) Any person who regularly engages, in whole or in part, in the business of assembling or collecting information about natural persons for the primary purpose of providing the information to an insurance institution or agent for insurance transactions, including either of the following: (A) The furnishing of consumer reports or investigative consumer reports to an insurance institution or agent for use in connection with an insurance transaction. (B) The collection of personal information from insurance institutions, agents, or other insurance-support organizations for the purpose of detecting or preventing fraud, material misrepresentation, or material nondisclosure in connection with insurance underwriting or insurance claim activity."
The definition continues at paragraph (2) with a list of persons who are excluded from the category, being agents, governmental institutions, insurance institutions, medical care institutions, medical professionals and peer review committees. A data vendor is not on that exclusion list, so the vendor's position turns entirely on paragraph (1).
The subject matter of paragraph (1) is information about natural persons. A parts and labour database is information about vehicles and repair operations. A total loss valuation database is information about vehicles. On the face of the statute neither is inside the category, and the reason is definitional.
Scope: this is California's enactment, read on its own text. We did not read other states' versions, and we located no agency or court reading of the phrase.
What the vendors say about their own numbers
Two of the three estimating systems named on this page, CCC and Audatex, are sold by companies that have filed with the Securities and Exchange Commission, and both have told the Commission, in their own filings, that the priced content of an estimate is bought in rather than measured. We read no comparable filing for the third system, Mitchell, in this pass, and we make no claim about whether one exists. A company's filing is strong evidence of what the company says. It is not evidence that what it says is true.
CCC Intelligent Solutions Holdings Inc., in its Form 10-K for the fiscal year ended December 31, 2025 (accession 0001193125-26-067448, filed 2026-02-24), states that it obtains much of its data about vehicle parts and components and "collision repair labor and costs through license agreements with third parties" who "may be sole-source suppliers" of that data. The same filing elsewhere refers to materials licensed from "sole-service suppliers", against "sole-source suppliers" in the passage above. Both spellings are in the filed document and neither is a transcription error of ours.
Solera Corp.'s Form S-1 filed 2024-06-28 (accession 0001193125-24-172270) contains the phrases "substantial portion of the data used in our repair estimating software", "licensed from sole-source suppliers", and "has historically provided us with vehicle glass data". Each was confirmed present in that accession. We do not reproduce the sentences around them, for the reason given immediately below.
How these company statements were verified, and what that method cannot do. We did not load these filings into the working context, because a 10-K of this size is exactly the kind of document that wrecks a research pass. We verified the wording instead by exact-phrase search against the SEC's own full-text search index, confirming each phrase in the specific accession named. Before trusting it, we proved the instrument in both directions: the phrase "collision repair labor and costs through license agreements with third parties" returned 32 filings including the CCC 10-K, while the same phrase with "purchase agreements" substituted returned zero hits and no error, across 50 of 50 index shards. A search that can return a clean zero is a search whose hits mean something. The limit of the method is that it proves a phrase is present in a filing. It does not prove the shape of the sentence around it, and it cannot support a paraphrase wider than the phrases actually returned. That is why the Solera passage above is now given as phrases rather than as a narrative of what the company says: two elements of our earlier draft of it, a list of supplier types and the description of the glass data supplier as a primary competitor, returned zero hits on the same instrument and were cut.
The counterweight: the database does get tested, and the vendors do answer. An earlier draft of this page stated that we had located no published accuracy measurement of any deployed commercial estimating or valuation database, by any vendor, regulator, standards body or academic. That was wrong. Two mechanisms exist, both built and paid for by the repair side of the industry rather than by the vendors or by any regulator.
The Society of Collision Repair Specialists published a measurement of one of the most heavily used labour values in the three systems. From its account of the study, dated 2022-11-01 and read on 2026-09-02: "SCRS worked with AkzoNobel Vehicle Refinishes, Axalta Coating Systems, BASF Automotive Refinish, PPG Industries and Sherwin-Williams Automotive Finishes to perform the research at the Global Finishing Solutions (GFS) Center for Excellence, at the GFS headquarters in Osseo, Wisconsin." The method was documented and externally witnessed: "The study utilized 45 new OEM parts, donated by Ford Motor Company, including hoods, fenders and front door shells for the 2018 Ford F-150, and was monitored and audited by DEKRA North America". SCRS states that each of the three systems "establishes the labor for blend on a two-stage refinish by utilizing a formula of 50% of the full refinish value", and reports: "The conclusions of this study indicate that blending, on average, took 31.59% more time than a full refinish, rather than the 50% less time allocated in the three IP estimating systems."
That is a published, externally audited measurement of a deployed commercial estimating database value, and on SCRS's own account it found all three systems understating the same operation by a wide margin. The finding and its interpretation are SCRS's, made by a trade association with an interest in the answer; we did not obtain the underlying data and we are not in a position to confirm the figure independently. Read as evidence about auditability, it cuts hard both ways: it is the strongest evidence on this page that these databases can be tested from outside, and the strongest evidence that testing them is worth doing. We are not claiming that the blend finding generalises to any other labour value in any of the three systems. One operation was measured, on parts for one vehicle line, by a party with an interest in the answer.
There is also a standing channel. The Database Enhancement Gateway describes itself as "an initiative created and funded by the Alliance of Automotive Service Providers (AASP) and Society of Collision Repair Specialists (SCRS)", and says it "helps improve the information in collision repair estimates through your feedback about vehicle-specific errors, omissions, and inaccuracies in the database and labor times". It connects users "directly to the databases of CCC, Mitchell, and Audatex products", forwards issues to the vendor, and records the traffic on "an easy to navigate public database". Submitters are told to expect confirmation at milestones including "Response/resolution received from the Information Provider".
Now the limits, which are the reason none of this amounts to an audit right. Both mechanisms are funded by the shops, not by the vendors and not by any regulator. Participation and response are voluntary throughout. The DEG's own page states the boundary exactly: "Information Providers are all solely responsible for the times they publish." A vendor that declines to change a number is not in breach of anything, and SCRS exercises no authority over any vendor at all. The blend study is evidence that the numbers can be checked. It is not a power to make anyone check them, and it is not a power to make anyone act on the result.
The consequence for the person holding the estimate is narrow and worth stating precisely: the number on the line is a licensed data product, assembled from suppliers the estimate does not name. Nothing on this page establishes that any particular number on any particular estimate is wrong, high, low or biased.
Whether anyone has to tell the vehicle's owner
No instrument we read requires that a vehicle owner be told that the estimate or the valuation for their claim was produced, populated or priced by a third-party system, or which system it was. An absence claim is worth nothing without a control that shows the search can find such a requirement when one exists, so here are two.
Federal law knows how to write this disclosure. 15 U.S.C. section 1681m(a), the Fair Credit Reporting Act, read on 2026-09-02:
"If any person takes any adverse action with respect to any consumer that is based in whole or in part on any information contained in a consumer report, the person shall - (1) provide oral, written, or electronic notice of the adverse action to the consumer;"
and, skipping paragraph (2), which concerns credit scores, the part that matters here:
"(3) provide to the consumer orally, in writing, or electronically - (A) the name, address, and telephone number of the consumer reporting agency (including a toll-free telephone number established by the agency if the agency compiles and maintains files on consumers on a nationwide basis) that furnished the report to the person; and (B) a statement that the consumer reporting agency did not make the decision to take the adverse action and is unable to provide the consumer the specific reasons why the adverse action was taken; and"
Name the third-party supplier, and tell the consumer the supplier did not make the decision. American law has written that architecture, for credit. It has not written it for the number on a repair estimate.
California knows how to compel a sentence onto the face of an estimate. This is the tighter control, because it is in the same regulation, about the same document. 10 CCR 2695.8(g)(3) requires that where an insurer specifies non-original-equipment crash parts, "The insurer must disclose in writing, in any estimate prepared by or for the insurer, the fact that it warrants that such parts are at least equal to the original equipment manufacturer parts in terms of kind, quality, safety, fit, and performance". California has compelled a disclosure onto the estimate about aftermarket parts. It has compelled none about who or what produced the estimate.
What the rules do require is disclosure of the comparables and the arithmetic, not of the system. Washington's WAC 284-30-392 requires the total loss valuation report to include "A list of the comparable motor vehicles used by the insurer to arrive at the actual cash value", and that list must include "The source of the information used". That phrase sits inside the list of comparable vehicles and identifies where a listing came from. It does not name the valuation product or its vendor. The report itself is disclosed only "if requested", under WAC 284-30-391(4)(d). California requires under 10 CCR 2695.8(b)(4) that the cost of a comparable automobile, "once determined, shall be fully itemized and explained in writing for the claimant at the time the settlement offer is made", and under 2695.8(f) that where a partial loss is settled on a written estimate prepared by or for the insurer, "the insurer shall supply the claimant with a copy of the estimate upon which the settlement is based". You are entitled to the estimate. You are not entitled to know what wrote it.
The database Washington did regulate, and the one it did not
Washington wrote substantive content requirements for the total loss valuation database. WAC 284-30-391(2)(b)(iv), read on 2026-09-02:
"(iv) Computerized source: The insurer may use a computerized source to establish a statistically valid actual cash value of the loss vehicle. The source used must meet all of the following criteria: (A) The source's database must produce values for at least eighty-five percent of all makes and models for a minimum of fifteen years taking into account the values of all major options for such motor vehicles. (B) The source must produce actual cash values based on current data within a reasonable distance of the principally garaged area, not to exceed one hundred fifty miles. (C) The source must rely upon the actual cash value of comparable motor vehicles that are currently available or were available in the market place within ninety days prior to or after the date of loss. (D) The source must provide a list of comparable motor vehicles used to determine the actual cash value. If more than thirty comparable motor vehicles are located, the insurer need list only thirty but may list more."
Read the grammar carefully. Sub-paragraphs (A) to (D) are written as duties on "the source", but the operative sentence is "The insurer may use a computerized source", and the chapter's enforcement runs to the insurer as the licensee. If the database fails the eighty-five percent test, the insurer is in violation and the vendor is not.
The comparison is the point. Two databases, sold by the same handful of companies, in the same claim, to the same carrier. The valuation database has an eighty-five percent coverage floor, a fifteen-year depth requirement, a hundred-and-fifty-mile locality rule, a ninety-day currency rule and a comparables-listing duty. The repair estimating database has none of that.
The one precedent, and three reasons it does not transfer
There is one American case of a regulator forcing a claims-pricing database out of an interested party's hands. It comes from health insurance, and it worked.
In 2009 the New York Attorney General reached an agreement with UnitedHealth Group over the Ingenix databases, whose data insurers used in calculating out-of-network reimbursement. On 2009-03-31 both the parent's and the subsidiary's chief executives appeared before the Senate Committee on Commerce, Science, and Transportation.
Andy Slavitt, then Chief Executive Officer of Ingenix, told the committee that during the Attorney General's review his office had raised a concern that Ingenix's ownership of the databases presented an inherent conflict of interest, and that under the agreement the databases would be transferred to an independent non-profit. He drew the distinction the estimating desk would also draw, between an inherent conflict and any actual practice of bias, and described his company's role as a limited one, collecting, organising and keeping current the charges physicians bill for their services.
Those statements are given in substance rather than in quotation marks, and the reason is a defect in our own retrieval worth naming. The printed hearing record carries each witness twice, once as spoken testimony and once as a separately submitted prepared statement, and the two versions differ in wording. Three retrieval routes returned three different renderings of the same Slavitt sentences, differing on whether the transfer sentence contains the word "soon", on "nonprofit" against "non-profit", and on "has been" against "have been". One quotation in our earlier draft matched neither version, because it combined the prepared statement's opening with the spoken version's wording. We cut every Slavitt quotation rather than print a sentence we cannot pin to one version of the record.
The denial belongs in the same passage as the remedy. Two sentences from Stephen J. Hemsley, then President and Chief Executive Officer of UnitedHealth Group, came back character-identical on every route and read the same in both versions of the record. He testified: "First, the Ingenix databases did not set the reimbursement rates used by any health insurer." And: "Our recent agreement with the New York Attorney General did not relate to the manipulation of data, and we disagree with any suggestion or allegation of fraud." No finding of manipulation was made. We will not describe this settlement as a finding of manipulation, because there was none and the company denied it on the record. The remedy went to who owned the instrument, not to any individual number.
The successor exists. FAIR Health's own account of its origin, read on 2026-09-02:
"Established in October 2009 as part of the settlement of an investigation by New York State into conflicts of interest involving the adjudication of claims, FAIR Health was formed to create a conflict-free, trusted and transparent source of data to support claims adjudication and to meet the healthcare cost and utilization information needs of all participants in the healthcare community."
Seventeen years on it describes itself as "an independent nonprofit that collects data for and manages the nation's largest database of privately billed health insurance claims", as "A conflict-free, nonprofit organization that qualifies as a public charity under section 501(c)(3) of the Internal Revenue Code", and it says of its data: "Our data are incorporated into statutes and regulations around the country".
Three reasons it does not transfer cleanly to collision.
Ownership. Ingenix was a wholly owned subsidiary of the payer, and vertical integration between the payer and the pricer was the specific conflict the New York office identified. That form is not present here. CCC is a public reporting company that files annual reports with the SEC, which is how we verified its own statements above. We did not trace the ownership of Mitchell in this pass, and beyond the Solera registration statement cited below we did not trace Audatex's. We assert nothing about either. Whatever conflict exists in collision estimating is a different one and would need to be established on its own evidence.
Jurisdiction. The New York Attorney General is a state law officer acting under state powers, not an insurance department examining a vendor. Nothing in the Ingenix record establishes that an insurance regulator can examine a claims-data vendor, and the section above shows the enforcement lever the regulators themselves designed does not reach claims.
Funding. Whoever proposes an independent successor for collision estimating has to say who pays for it. We could not verify how FAIR Health was funded at its founding, and we cut the testimony we had on that point rather than publish it unverified.
What we could not verify, and the walls
Everything in this section is a limit on the page above, stated so a reader can discount accordingly.
Cut for failure to re-verify. A comment letter from the National Association of Mutual Insurance Companies, stating that departments' regulatory authority does not extend to third parties, was to have been the industry's own admission on the central point. The NAIC materials PDF is retrievable, but the fetch route returned the document truncated part-way through the comment letters, and the NAMIC letter was not in the retrieved content. We cut the quotation rather than print it with a hedge. For the same reason we cut the list of the ten organisations that commented and the claim that no collision estimating vendor was among them. That is an absence claim, and an absence claim drawn from a truncated document is worth nothing. We refuse to publish that list until the whole document has been read.
Also cut, all for the same reason, from the Senate hearing: the share of revenue the divested databases represented, the exchange about how the successor non-profit would fund itself, the testimony that the transfer was not otherwise planned, and the parent's "reference database only" characterisation.
Cut because two versions of the same testimony disagree. Every Andy Slavitt quotation was cut, for the reason given in the precedent section: the hearing record prints spoken testimony and a separately submitted prepared statement in differing words, our three routes disagreed about which words belong to which, and one drafted quotation matched neither. The substance survives and is stated without quotation marks. The two Hemsley sentences retained are the two that came back identical on every route and in both versions.
Corrected, not cut. The claim that no published accuracy measurement of a deployed commercial estimating or valuation database existed was false, and it was published in an earlier draft without any instrument being run against it. The SCRS blend study of 2022-11-01 is exactly such a measurement, and it is now on the page at full strength together with the Database Enhancement Gateway. This was the largest single defect in the draft, and it ran in the direction of our own argument, which is the direction an absence claim is most likely to run.
Cut for scope. MOTOR Information Systems publishes marketing language describing its labour times as built with proprietary methodologies. That page describes a mechanical work-times product for customer pay operations, not the collision guide, and we will not use it as evidence about collision body labour. We also cut the 2010 Allstate consent order over Colossus, the sealing of a valuation methodology as a trade secret in Watson v. Progressive, the New York vendor-contract circular letter, and CCC's revenue split between insurers and repairers, none of which we re-fetched today.
Cut for instrument failure. We ran no string counts over the CCC 10-K, and we publish none. A count is only meaningful against the extraction it was run on, and we did not produce a fresh full extraction. Two elements of our earlier Solera paraphrase returned zero hits on the SEC full-text index and were cut, as described above.
Cut as our own arithmetic. An earlier draft said the Senate hearing came six weeks after the New York agreement. We never established the agreement's date from a primary source, and the interval was in any event wrong. The hearing date is given above because it is printed on the hearing record; no interval is now asserted.
Walls and failures, with URLs, tested 2026-09-02.
Direct command-line fetching to any host is blocked in this session by the egress proxy, which answers 403 to CONNECT. This was re-tested today against www.govinfo.gov and failed the same way. This is an environment constraint, not a site refusing us. All retrieval here went through browser fetch routes.
ag.ny.gov is robots-disallowed. The 2009 New York settlement agreement itself is therefore still not in our hands, and we did not request it. The precedent above is reached through the congressional record and the successor organisation's own founding statement instead, which are primary routes to the same facts but are not the instrument. This is also why no date is asserted for the agreement.
https://content.naic.org/sites/default/files/national_meeting/081226-materials-tpdmwg-su-nm.pdf returns the document, but our route truncated it part-way through the comment letters. Passages we reproduced from it matched exactly, and each was located in a named attachment; passages we could not find in it are unverified rather than absent, and are treated that way above.
https://www.govinfo.gov/content/pkg/CHRG-111shrg50467/html/CHRG-111shrg50467.htm returns the hearing, truncated. The PDF at https://www.govinfo.gov/content/pkg/CHRG-111shrg50467/pdf/CHRG-111shrg50467.pdf reaches further and is what established that the record carries two differing versions of each witness.
https://www.sec.gov/Archives/edgar/data/1818201/000119312526067448/ccc-20251231.htm returns the 10-K, truncated mid-document. Its content ended in the middle of a sentence, which is how we caught it. This is why the company statements above were verified by exact-phrase search instead.
The SEC full-text search endpoint rejects phrase queries beyond roughly one hundred characters with "Blank search not valid", which is an error and not a zero result. One query returned a 500. Both were retried with shorter phrases.
Cornell's California regulation URLs need a dot before the subsection number, not a hyphen. https://www.law.cornell.edu/regulations/california/10-CCR-2695-8 returns HTTP 200 and a list of the titles of the California Code of Regulations, not section 2695.8. A citation built on the hyphen form would look retrieved and be wrong.
A wall that is not one, corrected below. The California Department of Insurance's own PDF of the labour rate survey regulations returned extractable text today, not an image-only scan.
Corrections
This log opens with this publication and will record every correction made to this page.
2026-09-02. Correction on first publication, to our own prior research note rather than to a published page. That note recorded the California Department of Insurance PDF at https://www.insurance.ca.gov/0250-insurers/0500-legal-info/0500-gen-legal-info/upload/Labor-Rate-Survey-Regulations-Effective-Nov-30-2016.pdf as a 20-page image-only scan with zero extractable text, and gave that as the reason two earlier passes failed to obtain the California regulations. Re-tested on 2026-09-02, the file returned extractable text beginning "STATE OF CALIFORNIA / DEPARTMENT OF INSURANCE", including the heading "FINAL TEXT OF REGULATION / Auto Body Repair Labor Rate Surveys / Date: November 30, 2016 / CDI Regulation File: REG-2012-00002". The document has a text layer on this route. We have used it as an independent second source for 10 CCR 2695.81(d)(7) and 10 CCR 2698.91(h), and both matched Cornell's reproduction exactly.
2026-09-02. Correction to the same note, which reported a sentence from the NAIC meeting summary as a direct quotation. Our re-fetch of that summary returned different wording for the same item. We have not published it as a quotation and have stated its substance instead.
2026-09-02. Correction made in adversarial review before publication. A draft of this page asserted that no published accuracy measurement of any deployed commercial estimating or valuation database existed, by any vendor, regulator, standards body or academic. No instrument had been run against that claim, and the claim was false. The SCRS blend study of 2022-11-01 and the Database Enhancement Gateway are both now on the page, with their limits.
2026-09-02. Correction made in the same review. Four quotations attributed to Andy Slavitt before the Senate Commerce Committee were cut. The hearing record prints spoken testimony and a separately submitted prepared statement in differing words; one drafted quotation matched neither version, and the others could not be pinned to one. Two elements of the Solera paraphrase, and an interval of six weeks between the New York agreement and the hearing, were cut for the same class of reason.
Sources
Every source below was re-fetched and checked on the date given before anything from it was published here.
Cal. Code Regs. tit. 10, section 2695.8, Additional Standards Applicable to Automobile Insurance. https://www.law.cornell.edu/regulations/california/10-CCR-2695.8 , read on 2026-09-02, and pulled twice by two independent routes. The sentence in subdivision (f) about the third-party estimating software was added by an amendment filed 12-31-2012, operative 1-30-2013.
Cal. Code Regs. tit. 10, section 2695.81, The Standardized Auto Body Repair Labor Rate Survey. https://www.law.cornell.edu/regulations/california/10-CCR-2695.81 , read on 2026-09-02. New section filed 11-30-2016, operative 1-1-2017. Subdivision (d)(7) was verified three times: twice at this URL by two routes, and once against the California Department of Insurance's own final text of the regulation, cited below. All three were character-identical.
Cal. Code Regs. tit. 10, section 2698.91, Reporting Auto Body Repair Labor Rate Surveys. https://www.law.cornell.edu/regulations/california/10-CCR-2698.91 , read 2026-09-02.
California Department of Insurance, Final Text of Regulation, Auto Body Repair Labor Rate Surveys, dated November 30, 2016, CDI Regulation File REG-2012-00002. https://www.insurance.ca.gov/0250-insurers/0500-legal-info/0500-gen-legal-info/upload/Labor-Rate-Survey-Regulations-Effective-Nov-30-2016.pdf , read 2026-09-02.
California Insurance Code section 791.02, Insurance Information and Privacy Protection Act, 2025 California Code, amended by Stats. 2021, Ch. 190, Sec. 5 (AB 1184), effective January 1, 2022. https://law.justia.com/codes/california/code-ins/division-1/part-2/chapter-1/article-6-6/section-791-02/ , read 2026-09-02.
Wash. Admin. Code 284-30-391 and 284-30-392. https://app.leg.wa.gov/wac/default.aspx?cite=284-30-391 and https://app.leg.wa.gov/wac/default.aspx?cite=284-30-392 , both read 2026-09-02. Statutory authority RCW 48.02.060 and 48.30.010, WSR 09-11-129, filed 5/20/09, effective 8/21/09.
15 U.S.C. section 1681m, Requirements on users of consumer reports. https://www.law.cornell.edu/uscode/text/15/1681m , read 2026-09-02.
NAIC Third-Party Data and Models (H) Working Group, Summer National Meeting materials of 2026-08-12, including the framework exposure draft of 2026-07-08 and Working Group minutes. https://content.naic.org/sites/default/files/national_meeting/081226-materials-tpdmwg-su-nm.pdf , read 2026-09-02, retrieved truncated. Meeting summary: https://content.naic.org/sites/default/files/national_meeting/2026-sunm-summary-h-tpdmwg.pdf , read 2026-09-02. The framework is an exposed draft and is not law.
CCC Intelligent Solutions Holdings Inc., Form 10-K for the fiscal year ended December 31, 2025, accession 0001193125-26-067448, filed 2026-02-24. Wording verified by exact-phrase search against the SEC full-text search index on 2026-09-02, with a negative control returning zero hits.
Solera Corp., Form S-1 filed 2024-06-28, accession 0001193125-24-172270. Wording verified the same way on the same date. Two elements of an earlier paraphrase returned zero hits and were cut.
Society of Collision Repair Specialists, SCRS Study Concludes Blend Time is Greater than Full Refinish, dated 2022-11-01. https://scrs.com/scrs-study-concludes-blend-time-is-greater-than-full-refinish/ , read 2026-09-02.
Database Enhancement Gateway. https://degweb.org/ , https://degweb.org/about-deg/ and https://degweb.org/inquiry-process/ , all read 2026-09-02.
S. Hrg. 111-37 Part 2, Deceptive Health Insurance Industry Practices: Are Consumers Getting What They Paid For? Part II, hearing before the Senate Committee on Commerce, Science, and Transportation, March 31, 2009. Witnesses Stephen J. Hemsley, President and Chief Executive Officer, UnitedHealth Group, and Andy Slavitt, Chief Executive Officer, Ingenix. https://www.govinfo.gov/content/pkg/CHRG-111shrg50467/html/CHRG-111shrg50467.htm and https://www.govinfo.gov/content/pkg/CHRG-111shrg50467/pdf/CHRG-111shrg50467.pdf , both read 2026-09-02, the HTML retrieved truncated.
FAIR Health, Mission and Origins, https://www.fairhealth.org/mission-origin , and About Us, https://www.fairhealth.org/about-us , both read 2026-09-02.
General consumer information: not legal, insurance, or financial advice. Requirements, coverage, and practices vary by state, policy, and manufacturer.