The Cheapest Thing You Can Do Is Nothing

Short answer: Every route that gets you a decision-maker costs more than the dispute. Small claims court is the exception, and it has no lawyer in it.

By Anthony Braswell for Quorum Industries LLC, The Autobody Directory · Updated 2026-09-03 · How this was written, and what the machine may not do

How this page was produced: Researched, drafted and checked with AI assistance under human direction, and signed off by the named author. How this site is written

That is the corrected version of a claim we got wrong the first time. We had written that every route out of a disagreement with your insurer costs more than the disagreement is worth. Then we priced small claims court in three states and found filing fees of $15 to $150, which against a median recovery of $1,946.32 is under one per cent to about eight per cent. Small claims is not expensive. Every other route in this article that reaches somebody with authority over the number starts at $375. The sentence that survives is narrower and harder: every route that supplies somebody with the authority to put a number on your car costs more than the number is usually worth.

The one number this rests on

Connecticut and Texas are the two state complaint files we queried directly, and Connecticut is the only one of the two that records how much money a complaint recovered. Its open-data endpoint was queried directly on 2026-09-03. Here is the query, so you can run it yourself:

https://data.ct.gov/resource/t64r-mt64.json
  ?$select=count(1),count(distinct file_no),sum(recovery),min(closed),max(closed)
  &$where=subcoverage in('Collision','Comprehensive','Physical Damage')

The row is not the complaint, and this is the trap in this file. Connecticut writes one row per reason, subreason and conclusion attached to a complaint, so a single complaint can appear many times with its recovery repeated in every row. File number 7050758, one Allstate collision complaint closed in February 2023, occupies six rows, each carrying the same $60,280.70. Counting rows overstates the number of complaints by a factor of more than two and inflates any sum of recoveries. Everything below is counted on distinct file_no, not on rows.

Our arithmetic on those inputs: the department recorded no recovery in 87.0 per cent of Connecticut auto physical damage complaints (1,956 divided by 2,248), and a recovery in 13.0 per cent (292 divided by 2,248). Because far more than half the file is zero, the median recovery recorded across all 2,248 complaints is $0.00. Counted on rows rather than complaints the zero share is 79.1 per cent, so the row count understates how often a complaint records no recovery. We are not claiming that 87.0 per cent of these complainants walked away with nothing. That is what the department's money column records, and the section below measures how far wrong that column can be.

Exactly one complaint, file 7053225, carries both a zero row and an $805.87 row; we count it among the 292 that recovered something. That is why 292 plus 1,957 complaints with at least one zero row exceeds 2,248 by one. That file is also a warning about the unit of analysis: its rows name two different carriers, GEICO and Allstate, under one file number.

Complaining to the Connecticut Insurance Department is free. In money recorded by the department, its median result is the same as the median result of doing nothing at all.

What the recovery field actually records, and how far wrong it can be

This was the defect that kept this article off the press, and it is worth saying plainly what it was. Everything above rests on one column, recovery, and that column has no published coding rules. If the department records only money it extracted itself, and not money a carrier paid after being prompted, the zeros are overstated and the sentence above is too strong.

We tried to find the rules and they do not exist in public. The dataset's own metadata defines the column as nothing more than "Recovery Amount". The department's complaint-process page walks through the steps of a complaint, and tells you what it cannot do, in these words: "Please note, we cannot make the insurance company approve or cover services that fall outside the contract or if no violation has occurred." It never says what goes in the money column. The department's Consumer Affairs page offers one sentence about money, that the unit "recovers approximately $4 million a year on behalf of consumers", which is a total and not a rule. The federal catalogue entry for the same dataset carries no data dictionary. The department's own Complaint Activity Reports page returned a 404 on 2026-09-03. All read on 2026-09-03.

So we closed it the second way, by measuring from the data how far wrong the field can be. Connecticut codes a conclusion on every row, and that vocabulary is specific enough to bound the error.

The same outcome code sits on both sides of the money line. Among complaints recorded as recovering nothing, 67 carry the conclusion "Claim Paid". Among the 292 complaints that did record money, 64 carry that same conclusion "Claim Paid". A code that appears 67 times against a zero and 64 times against a number is not a code the money column follows. Money moved in at least 67 complaints this field records as zero. That is a count, not a worry.

Here is every conclusion code among the 1,956 zero-recovery complaints that implies the complainant obtained something, queried on 2026-09-03:

That sum is an upper bound rather than a count of complaints, and deliberately so. One complaint carries several conclusion codes across its rows, so the categories overlap: taken separately they add to 2,290 against 1,957 complaints. The number of distinct complaints behind that table is therefore at or below 320, never above it.

Our arithmetic on that bound. If every one of those 320 complaints in truth delivered the complainant something, then at least 1,636 of the 2,248 complaints, or 72.8 per cent, delivered nothing, and at most 612, or 27.2 per cent, delivered something. Read at its most generous, by also counting the 180 complaints concluded "Justified", the 129 sent to arbitration and the 51 referred to appraisal, the bound rises to 680, and at least 1,276 complaints, or 56.8 per cent, still delivered nothing.

The headline survives the uncertainty, and this is the point of doing the bound. For the median complaint to have recovered anything at all, at least 1,124 of the 2,248 would have to have got something. The most generous reading of every outcome code in the file reaches 972. That is 152 complaints short of moving the median. So the median recorded result of a Connecticut auto physical damage complaint is $0.00 whether the recovery field counts only the department's own extractions or every dollar that ever moved.

What the bound does not do is rescue the precise figure. The honest form of the finding is that the department recorded no recovery in 87.0 per cent of these complaints, and that the true share who got nothing is somewhere between 56.8 and 87.0 per cent. We have not established which, and no email to the department has been sent.

Among the 292 complaints that did record a recovery, taken by ordered offset on distinct complaints on 2026-09-03:

The median needed no interpolation beyond the usual: 292 is even, so it is the mean of the 146th and 147th ordered values, which the endpoint returned as $1,942.64 and $1,950.00. That mean is our arithmetic.

Seven instrument checks, because a query that silently fails looks exactly like a query that succeeds. First, the filter bites: unfiltered the same endpoint returns 86,041 rows and 24,018 distinct complaints, against 4,984 and 2,248 filtered. Second, asking for a subcoverage that does not exist, Collisionn, returns a count of 0, so the filter is capable of returning nothing. Third, sorting recovery in descending order returns $60,280.70 six times before $60,000 and $48,483.37; a text sort would have put values beginning with a 9 on top, and the dataset's own metadata types the column as a number, so the ordering behind those percentiles is numeric on two independent proofs. Fourth, the three coverage codes reconcile against the whole: Collision 3,970 plus Comprehensive 945 plus Physical Damage 69 is 4,984 rows exactly, and no row in the filtered set has a null recovery. Fifth, the money split reconciles both ways: 3,942 zero rows plus 1,042 paying rows is 4,984, and 1,957 plus 292 complaints is 2,249, one more than 2,248, which is the single file that appears on both sides. Sixth, the conclusion tables above reconcile at row level on both sides of that split, to 3,942 and to 1,042. Seventh, the "Claim Paid" figure of 67 was counted a second way, one subcoverage at a time, returning 34 for Collision, 20 for Comprehensive and 13 for Physical Damage, which sum to 67 exactly.

We record two failed controls, and the second is a trap worth naming. The row-level sum of recoveries, $5,505,450.21, is unchanged when the zero rows are dropped, which proves the filter bit. It is not a total of money recovered, because it adds the same recovery once per row. The nested query that would total one recovery per complaint was refused by our own network layer, so we do not publish a corrected total at all rather than publish the inflated one.

The second failure is worse, because it does not announce itself. Above roughly 200 characters of URL, our own network layer silently discards the $where clause and the endpoint answers the unfiltered question instead, returning HTTP 200 and the whole-file total of 24,018 complaints. A query written to count a narrow subset came back with the entire dataset and looked exactly like a success. We caught it because 24,018 is a number we already knew the shape of. Every count published above was therefore re-run short enough to survive, and checked against a total it had to reconcile with. Anyone re-running this work should assume the same trap.

What this number is not

It is one state. It is one coverage group. It is four years and eight months. We refuse to let Connecticut stand in for the country, and no national figure appears anywhere in this article, because none exists that we could find.

The date window is a property of the coverage codes, not of the file. We got this wrong in the first version and are correcting it here. The file itself begins on 2018-01-09. These three subcoverage codes begin on 2022-01-04, because Connecticut did not code auto complaints at this level of detail earlier. We checked what that discards: only 4 private passenger auto complaints in the entire file closed before 2022. The filter is therefore a coverage filter that behaves like a date filter, and almost nothing falls through it.

It is also not a distribution of the amounts people were arguing about. It is a distribution of amounts recorded as recovered through a state insurance department, which is a different quantity, and the section above measures how different it can be. For a complaint that recovered some amount, the disagreement was at least that large. For the 1,956 that recorded nothing, the size of the disagreement is simply not in the file.

The absence of a national figure is a property of what regulators collect, not of how hard we looked, and we tested that before saying it. The same kind of query run against Texas on 2026-09-03 returned 104,649 automobile records covering 99,434 distinct complaints, across sixteen record fields, and not one of them is a dollar amount:

complaint_number, respondent_name, complainant_role, reason, complaint_confirmed_code, disposition, received_date, closed_date, complaint_type, coverage_type, coverage_level, involved_party_type, respondent_id, respondent_role, respondent_type, complainant_type

Texas has row duplication too, though much milder than Connecticut's. Its automobile file counts every kind of auto coverage including liability, so it is not scope-matched to Connecticut's physical damage subset and we draw no ratio between them. The point is narrower and holds: a file many times Connecticut's size records no money at all.

Two other measurements of the same order of magnitude, both read at their own source on 2026-09-03. In the Reynolds settlement with Progressive, the claims administrator's FAQ states "The estimated average recovery is $533," says "The Total PSA Impact Amount that is available to be claimed by the Settlement Classes is estimated to be $30,748,947," and defines the per-person amount mechanically: "the PSA Impact Amount is 4.31% of the Actual Cash Value, as determined in Progressive's records." Our arithmetic on two of those figures, $533 divided by 0.0431, implies an average class vehicle worth about $12,367; that division is ours, not the settlement's. And in the one recent appraisal case with a published opinion, Ohio's Eighth District Court of Appeals recorded that "Farmers issued a check payable to Stewart for $1,393.29, representing the difference between the original amount and the appraisal award."

For scale: ISO figures published by the Insurance Information Institute put 2021 average claim severity at $5,010 for collision and $2,042 for comprehensive. Our arithmetic: the Connecticut median recovery of $1,946.32 is about 39 per cent of an average collision claim ($1,946.32 divided by $5,010).

What each way out costs

The department of insurance complaint

Free, and structurally unable to decide the thing you are arguing about. California's statute says so in terms. Insurance Code section 12921.4 requires the commissioner to acknowledge a complaint within ten working days and give notice of final action within thirty days of taking it, then closes: "This section shall not be construed to give the commissioner power to adjudicate claims." Only California's statute was re-read for this point on 2026-09-03; we have not re-checked what any other state's complaint statute says.

Price: zero. Median money result recorded in Connecticut: zero. Results it can still produce without recording a dollar: 175 complaints concluded "Corrective Action" and 67 concluded "Claim Paid" among the zero-recovery complaints, as counted above.

Small claims court

The cheap exception, and the only route here that is both cheap and supplies a real decision-maker. Filing fees read at the instrument in three states on 2026-09-03.

New York, by statute, charges a "filing fee of fifteen dollars for claims in the amount of one thousand dollars or less and twenty dollars for claims in the amount of more than one thousand dollars". California's Code of Civil Procedure, for a filer who has brought twelve or fewer claims in the previous twelve months, sets "Thirty dollars ($30) if the amount of the demand is one thousand five hundred dollars ($1,500) or less.", "Fifty dollars ($50) if the amount of the demand is more than one thousand five hundred dollars ($1,500) but less than or equal to five thousand dollars ($5,000)." and "Seventy-five dollars ($75) if the amount of the demand is more than five thousand dollars ($5,000)." Massachusetts publishes a schedule on its own judiciary site: "Small claims of $500 and under= $40", "Small claims of $501 to $2,000 = $50", "Small claims of $2,001 to $5,000 = $100", "Small claims of $5,001 to $7,000= $150".

Our arithmetic against the Connecticut median recovered amount of $1,946.32: the New York fee of $20 is 1.0 per cent, the California fee of $50 is 2.6 per cent, and the Massachusetts fee of $50 for a claim in that band is 2.6 per cent.

Massachusetts also caps the downside of a large claim. Its own page adds that "The filing fee for claims of property damage of more than $7,000 arising from an automobile accident is also $150", so the fee stops rising exactly where a car claim gets serious.

Two limits, and the second is the one that matters. Ceilings vary by state, and we did not re-read them. And a small claims court can order money; it cannot order a carrier to change how it values cars, and there is no lawyer in the room.

The people who sell the expensive alternative recommend this one anyway. In a trade report read on 2026-09-03, Kristen Felder of Collision Hub said "I love small claims court", the report records that "Small-claims court would be better, he and Montanez indicated", and Mark Olson of Vehicle Collision Experts described the appraisal alternative as a case where "You spent a grand to chase $1,500".

The public adjuster

Priced as a share of the recovery, which means it is the only route that cannot by construction cost more than the dispute. Texas caps it by statute: a public insurance adjuster's "total commission received may not exceed 10 percent of the amount of the insurance settlement on the claim." Our arithmetic on the Connecticut median: 10 per cent of $1,946.32 is $194.63.

Two things break the route anyway. Florida, which has the broadest definition we have read, switches its own cap off for exactly this claim. Section 626.854(19) reads in full: "Subsections (5)-(18) apply only to residential property insurance policies and condominium unit owner policies as described in s. 718.111(11)." Subsections (5) to (18) contain the entire fee cap, including the "Twenty percent of the amount of insurance claim payments or settlements, exclusive of attorney fees and costs, paid to the insured by the insurer" limit at (11)(b)2. On a Florida car claim there is no percentage cap at all.

And $194.63 is not a cost to you. It is a reason nobody licensed takes the work. That is an inference, not a finding, and nothing we read measures whether public adjusters accept auto claims at any price.

The appraisal clause

The route that actually reaches the valuation, and the expensive one. Every policy we have read allocates cost the same way: each side pays its own appraiser and the two split the umpire. You pay one appraiser plus half an umpire whether you win or lose.

Two cost sources, five years apart, both read on 2026-09-03. The American Society of Certified Auto Appraisers, a certification body that also directs work to its own members, publishes: "Your appraiser: $250-$450 (paid by you) / Insurer's appraiser: Paid by the insurer / Umpire fee: $250-$500 (typically split equally between you and the insurer) / Your total out-of-pocket: $375-$700". That $375 to $700 total is the page's own arithmetic, not ours.

The practitioner figures are roughly twice that. The trade report records that Felder "said the process could involve an appraiser's fee of $495-$600 and half of an umpire's $600-$750 bill". Where a judge has to pick the umpire, that is "expensive, with the customer paying $1,000-$2,000 for their half of such a selection", according to Olson.

The two sources differ by roughly a factor of two and are five years apart. We refuse to average them. Here is each, kept separate, with our addition shown:

Our arithmetic throughout, on the quoted inputs above. We are not claiming these are what an appraisal costs. They are one certification body's published schedule and one practitioner's figures given to one trade reporter in 2021, and the two disagree by roughly a factor of two.

Set against the Connecticut median recovered amount of $1,946.32, that is 19.3 per cent at the cheapest and 133.6 per cent at the dearest. Set against the 25th percentile of $804.00, the cheapest route is 46.6 per cent, the cheapest practitioner route consumes 98.9 per cent of it, and every dearer practitioner route exceeds the whole amount at stake.

And the counts, queried on 2026-09-03, are what make this concrete. Of the 292 Connecticut complaints that recorded a recovery: 62 recovered less than $700, 91 recovered less than $975, 124 recovered less than $1,495, and 169 recovered less than $2,600. Our arithmetic on a denominator of 292: 21.2 per cent, 31.2 per cent, 42.5 per cent and 57.9 per cent respectively.

Three further things travel with those percentages, and none of them helps.

They all assume you win the entire disputed amount. The same trade report records that "Montanez said he'd only truly won about a dozen appraisals out of 2,000-3,000," and that "He's never saw a case end with less than an umpire who 'cut the baby in half.'" Every percentage above is therefore a floor.

The mechanism runs both ways. In the Ohio case, the insurer was the party that made the appraisal demand, moved to compel appraisal, paid the award, and thereby ended the litigation; the Eighth District reversed and remanded, holding that the trial court "erred in failing to find the claims moot and dismiss the action" and that after the binding award and payment "there was no longer a live controversy between the parties", which undid a grant of class certification. A route your carrier can force you down is not only an exit.

And the same certification body that publishes the fee schedule also asserts on the same page a "Typical additional recovery: $2,000-$8,000+ above the insurer's original offer", with no source of any kind. We do not use that figure and neither should you.

The lawyer

We could not price this one, and it is the largest hole in the comparison. No contingency rate for one individual with one car on an auto physical damage claim was located at any source. Class action fee awards exist and run from roughly a fifth to a third of a fund, but a class action is a different market from your dented door, and we refuse to present the one as a proxy for the other.

Doing nothing

Cost: zero. Recovery: zero. On Connecticut's own file, zero is also the median recorded money outcome of the free complaint route, though not its only outcome.

Where the arithmetic reverses

All of the above assumes the money you recover has to pay for the person who recovers it. In a few of the states we read, a statute breaks that link, and where it does, the size of your dispute stops governing whether the case is worth bringing. This is the most actionable part of the article and it is not a caveat. If you are in one of these states, the arithmetic above does not describe your position.

Massachusetts writes the objection into the statute and disposes of it. General Laws chapter 93A section 9(4), read on 2026-09-03:

"If the court finds in any action commenced hereunder that there has been a violation of section two, the petitioner shall, in addition to other relief provided for by this section and irrespective of the amount in controversy, be awarded reasonable attorney's fees and costs incurred in connection with said action; provided, however, the court shall deny recovery of attorney's fees and costs which are incurred after the rejection of a reasonable written offer of settlement made within thirty days of the mailing or delivery of the written demand for relief required by this section."

"Irrespective of the amount in controversy." A legislature wrote down the exact objection this article is built on and legislated it away. The same section, at 9(3), sets recovery at "actual damages or twenty-five dollars, whichever is greater; or up to three but not less than two times such amount" where a refusal to grant relief upon demand was made in bad faith, after a written demand at least thirty days before filing. Our arithmetic on the Connecticut median: not less than two times $1,946.32 is $3,892.64, and three times is $5,838.96, plus fees. Massachusetts also publishes on its judiciary site that its $7,000 small claims limit bends for this exact injury: "If your case is based upon property damage sustained in an automobile accident, the award may exceed $7,000."

Washington has commanded fees since a referendum in 2007, and the statute names expert witnesses. RCW 48.30.015(3), read on 2026-09-03:

"The superior court shall, after a finding of unreasonable denial of a claim for coverage or payment of benefits, or after a finding of a violation of a rule in subsection (5) of this section, award reasonable attorneys' fees and actual and statutory litigation costs, including expert witness fees, to the first party claimant of an insurance contract who is the prevailing party in such an action."

"Shall", not "may". The statute's own footer records its provenance: "2007 c 498 s 3 (Referendum Measure No. 67, approved November 6, 2007)." An expert witness is precisely what a valuation argument needs and precisely what a $1,946.32 dispute cannot fund, and this is the only statute we read that names them. What we did not establish is whether Washington's total-loss valuation rule falls inside the list of rules that trigger subsection (3); subsection (5) names WAC 284-30-330, 284-30-350, 284-30-360, 284-30-370 and 284-30-380, the valuation rule is not among those five, and we did not test the reach of the unreasonable-denial limb.

Georgia sets a floor, and it is the only instrument we found that grows as the dispute shrinks. O.C.G.A. section 33-4-6(a), read on 2026-09-03, makes an insurer liable in addition to the loss for a penalty of "not more than 50 percent of the liability of the insurer for the loss or $5,000.00, whichever is greater", and for all reasonable attorney's fees for the prosecution of the action. Our arithmetic on the Connecticut median: the fifty per cent limb is about $973, so the $5,000 floor governs, at about 2.57 times the amount at stake. Three limits travel with it and cannot be dropped: it requires a refusal to pay within sixty days of a demand, it requires a finding of bad faith, and whether paying part of a claim counts as a refusal to pay under that section is something we did not research and do not claim.

Texas attaches a rate to delay. Insurance Code section 542.060(a) makes a non-compliant insurer liable, in addition to the amount of the claim, for "interest on the amount of the claim at the rate of 18 percent a year as damages, together with reasonable and necessary attorney's fees." A 2017 amendment must travel with that quotation: subsection (c) replaces the 18 per cent with the rate determined under Section 304.003, Finance Code plus five percentage points in an action to which Chapter 542A applies. We did not read Chapter 542A's scope provision and we make no claim about whether an auto physical damage claim falls inside or outside it.

New York's version is discretionary, and that is the whole difference. General Business Law section 349(h) says "The court may award reasonable attorney's fees to a prevailing plaintiff." and caps its treble enhancement: "The court may, in its discretion, increase the award of damages to an amount not to exceed three times the actual damages up to one thousand dollars". May, not shall, and a thousand-dollar ceiling on the enhancement.

And it is not a one-way ratchet: Florida took its statute away. Florida had a general fee-shifting section covering every line of insurance including auto. It is gone, and we established that by absence with the instrument controlled twice. The section page for 627.428 returns "The statute you have selected cannot be found." The identical URL pattern, on the same run, served section 627.70152 in full, with its title "Suits arising under a property insurance policy." and a history line reading "s. 12, ch. 2021-77; s. 16, ch. 2022-268; s. 17, ch. 2022-271." And the chapter index lists "627.427 Payment of judgment by insurer; penalty for failure" and "627.429 Medical tests for HIV infection and AIDS for insurance purposes" with no 627.428 entry between them. The instrument finds a Florida section when the section is there. This one is not.

The seven states we actually read

Seven states were read at an instrument for this table on 2026-09-03. Every cell below is a reading of a specific document. Every blank says so. Nothing in this article ranks or counts states beyond these seven, and no range in it is drawn from more.

We refuse to print the eighteen state table this article was drafted from. Only seven of its states were re-read at an instrument, and a cell nobody has re-opened is not evidence. We do not claim that the other forty three states lack a fee-shifting statute, or that they have one. Seven states were opened, and a blank cell above means the instrument was not read rather than that it says nothing. The $15 to $150 filing fee range quoted at the top of this article is the range across the three states in the first column, and nothing wider.

What we could not verify, and the walls we stopped at

Connecticut's own small claims filing fee is missing from this article, and that is the gap that stings. Connecticut supplies every number in the first section, and its judiciary page at https://www.jud.ct.gov/faq/smallclaims.html refused on 2026-09-02 with a robots disallow, reported as "robots.txt fetch failed: [SSL: SSLV3_ALERT_HANDSHAKE_FAILURE]". This is the second time we have tested it and had the same answer. We did not route around it and we did not go to a mirror or an archive, so the Connecticut fee is simply absent.

Three more walls. California's own legislature site at leginfo.legislature.ca.gov refused with a robots disallow, so the California Insurance Code and Code of Civil Procedure sections quoted above were read at a secondary code publisher instead of at the state. The Texas legislature's statute site at statutes.capitol.texas.gov refused the fetcher we were using that day, so the Texas sections were also read at a secondary publisher. Retested on 2026-09-03 it served full statutory text to a different fetcher, so that refusal was a fact about one instrument and not about the host. In both cases the fallback is a code publisher, not the state, and that is a weaker source than we would like. Third, on 2026-09-03 our own network layer refused the nested Socrata query that would have totalled one recovery per complaint, so no corrected sum of recoveries appears above.

One thing that is not a wall but is worth recording: the Connecticut Insurance Department's Complaint Activity Reports page at https://portal.ct.gov/cid/consumer-resource-library/cid-reports/complaint-activity-reports returned a 404 on 2026-09-03. Nothing blocked us. The page was not there.

Four things we could not establish at all:

What a lawyer charges one person with one car. Not located at any source. Everything available is class action fee data, which is a different market.

Whether anyone will take these cases. Nothing here measures supply. A ten per cent contingency on $194.63 is not a barrier the legislature intended; it is a reason a professional declines the file. That is reasoning, not evidence.

The published enacted text of the Florida repeal. The bill that carried the repealing words and the bill that passed are not the same bill, and the enacted text would not retrieve on either path we tried on flsenate.gov. That is why the Florida finding above rests on controlled absence from the current code rather than on the repealing sentence. A bill that was laid on the table is not a law, and an earlier draft of this research nearly cited one as though it were.

Any distribution of the amount in dispute. None exists in anything we located, for any state. The Connecticut file is a proxy and we have said so three times.

One arithmetic problem we are recording rather than solving. In the Ohio opinion, the court states an appraisal award of $11,564.08, an adjusted vehicle value of $10,295, an initial payment to Stewart of $9,795 representing that value plus fees minus a $500 deductible, and then a check for $1,393.29 described as the difference between the original amount and the award. No subtraction we can perform on the court's own figures reproduces $1,393.29: the award less the adjusted value is $1,269.08, and the award less the initial payment is $1,769.08. We reproduce the court's figures as the court stated them and we do not compose another from them.

Sources

Everything below was fetched and matched against the sentence it supports before that sentence was written.

• Connecticut Insurance Department complaint records, https://data.ct.gov/resource/t64r-mt64.json, read on 2026-09-03. All counts, distinct-complaint counts, percentiles, threshold counts, disposition and conclusion counts, plus the unfiltered, negative, ordering, reconciliation and subcoverage-split controls.
• Connecticut complaint dataset metadata, https://data.ct.gov/api/views/t64r-mt64.json, read on 2026-09-03. Column descriptions, including "Recovery Amount" as the whole of the definition of the recovery field, and the number datatype on that column.
• Connecticut Insurance Department, The CID Complaint Process, portal.ct.gov, read on 2026-09-03. Describes the handling of a complaint and says nothing about the recording of money.
• Connecticut Insurance Department, The Consumer Affairs Unit, portal.ct.gov, read on 2026-09-03. The approximately $4 million a year sentence.
• Federal catalogue entry for the same dataset, catalog.data.gov, read on 2026-09-03. No data dictionary.
• Texas Department of Insurance complaint records, https://data.texas.gov/resource/ubdr-4uff.json, read on 2026-09-03. 104,649 automobile records, 99,434 distinct complaints, sixteen fields, no dollar amount.
• Massachusetts General Laws c. 93A s 9, malegislature.gov, read on 2026-09-03.
• RCW 48.30.015, app.leg.wa.gov, read on 2026-09-03, including the referendum footer and the subsection (5) rule list.
• O.C.G.A. s 33-4-6, law.justia.com, read on 2026-09-03.
• Tex. Ins. Code s 542.060 and s 4102.104, codes.findlaw.com, read on 2026-09-03.
• N.Y. Gen. Bus. Law s 349 and N.Y. City Civ. Ct. Act s 1803, nysenate.gov, read on 2026-09-03.
• Cal. Code Civ. Proc. s 116.230 and Cal. Ins. Code s 12921.4, codes.findlaw.com, read on 2026-09-03.
• Fla. Stat. ss 626.854 and 627.70152, and the chapter 627 part II index, leg.state.fl.us, read on 2026-09-03, with the 627.428 absence controlled twice.
• Massachusetts judiciary small claims page, mass.gov, read on 2026-09-03.
• American Society of Certified Auto Appraisers umpire process page, certifiedautoappraisers.com, read on 2026-09-03.
• Repairer Driven News, 13 February 2021, appraisal clause process, repairerdrivennews.com, read on 2026-09-03. Felder, Olson and Montanez quotations.
• Stewart v. Farmers Ins. of Columbus, Inc., 2026-Ohio-1451, Court of Appeals of Ohio, Eighth Appellate District, Cuyahoga County, supremecourt.ohio.gov, read on 2026-09-03. Judgment reversed and remanded.
• Reynolds settlement administrator FAQ, altotallossclaim.com, read on 2026-09-03.
• ISO figures published by the Insurance Information Institute, iii.org, read on 2026-09-03. The page names the source with registered trade mark symbols, dropped here.

Corrections

2026-09-03. Our earlier version rested its headline on Connecticut's recovery field while recording, in a closing paragraph, that the field has no published coding rules and that the finding might therefore be too strong. That was a defect, not a caveat, and it has now been closed. We looked for the coding rules first, at the dataset metadata, the federal catalogue entry, the department's complaint-process page and its Consumer Affairs page, and they do not exist in public. We then bounded the error from the data instead. At most 320 of the 1,956 zero-recovery complaints carry a conclusion code implying the complainant obtained something, and at most 680 on the most generous reading of every code in the file. The median recorded result of $0.00 survives both bounds, because moving the median would require at least 1,124 complaints and the generous bound reaches 972. The article now states the route it took, the bound, and the range the true zero share lies in, which is 56.8 to 87.0 per cent.

2026-09-03. Our earlier version reported disposition counts of 106 "Claim Settled" and 14 "Company Position Overturned" among the zero-recovery complaints as the counterweight to the headline. Those counts are correct, but the conclusion field is the more informative one and is now used instead: 67 zero-recovery complaints are concluded "Claim Paid", against 64 among the complaints that did record money. The same code sits on both sides of the money line.

2026-09-03. We are recording a new failed control. Above roughly 200 characters of URL, our own network layer silently drops the $where clause and returns the unfiltered whole-file total of 24,018 complaints under HTTP 200. One query in this run came back that way and was caught only because the wrong answer was a number we recognised. Every count in this article was re-run inside the limit and reconciled against a total.

2026-09-03. Our earlier version counted rows in the Connecticut file as though each were a complaint. Connecticut writes one row per reason, subreason and conclusion, so a single complaint can occupy many rows carrying the same recovery. Every figure drawn from that file has been recounted on distinct complaints: 4,984 rows are 2,248 complaints, 3,942 zero rows are 1,956 complaints that recorded no recovery, 1,042 paying rows are 292 paying complaints, the median recorded amount falls from $2,013.09 to $1,946.32, and the share recording nothing rises from 79.1 per cent to 87.0 per cent. The direction of the finding is unchanged and its magnitude is larger. The published sum of recoveries has been withdrawn as double counted.

2026-09-03. Our earlier version said the January 2022 window was "the span of the file itself, not a date filter we applied". That was wrong: the file begins in January 2018, and the three coverage codes begin in 2022. Only 4 private passenger auto complaints in the file close before 2022, and that control is now shown.

2026-09-03. Our earlier version attributed Stewart v. Farmers Ins. of Columbus to the Ohio Supreme Court. It is a decision of Ohio's Eighth District Court of Appeals, and the judgment was reversed and remanded. Both now appear wherever the case is cited.

2026-09-03. Our earlier version stated the California small claims fees as "$30", "$50" and "$75" inside quotation marks. The statute reads "Thirty dollars ($30)", "Fifty dollars ($50)" and "Seventy-five dollars ($75)". The quotations now match the statute. The Florida twenty per cent cap quotation, which had been cut mid-phrase after "attorney fees", now runs to a clean boundary.

2026-09-03. Our earlier version said $15 to $150 was "one to five per cent" of the median and that small claims was "one to two orders of magnitude cheaper than everything else here". Against the corrected median of $1,946.32 the range is 0.8 to 7.7 per cent, and $150 against the $375 floor of the appraisal route is a factor of 2.5, not an order of magnitude. Both claims have been restated at their true size. The Massachusetts fee applicable at the median is $50, not $100, because the corrected median falls below the $2,000 band boundary. That $15 to $150 range is now labelled as covering the three states whose fees we read, and no wider.

2026-09-02. Our own earlier framing said every route out of a claim dispute costs more than the dispute. That is false of small claims court. The claim has been narrowed to routes that supply a decision-maker, and the small claims exception is now stated in the first line rather than buried.

2026-09-02. Our earlier note that the department of insurance complaint "is free" was accurate and incomplete. It is free, and on Connecticut's file its median recorded money result is $0.00, which is what you get for not complaining. Both halves now appear together.

2026-09-02. Our earlier field list for the Texas complaint record named fifteen fields while calling them sixteen. The sixteenth is involved_party_type, which appears only in records that carry it and was dropped from the first sample. The list above is now complete. The finding it supports, that no field is a dollar amount, is unchanged.

2026-09-02. Earlier research in this series described Washington as giving a consumer no private remedy, on the strength of a case about the total-loss valuation rule. Washington has had a first-party fee-shifting statute since a 2007 referendum, and it commands fees and expert witness costs. The two findings are reconcilable and both now appear.

General consumer information: not legal, insurance, or financial advice. Requirements, coverage, and practices vary by state, policy, and manufacturer.

More guides