Measured
Repair or replace, on 1,012 real decisions
Every estimator has an opinion about whether a carrier will pay to replace a quarter panel. This is what they actually did, counted off the documents they sent, panel by panel, with the sample size and a 95% interval on every row.
- Sample
- 1,012 decisions
- across 20 panels, 6 observations minimum
- Documents dated
- January 2024 to August 2026
- The same documents as the corpus. The panel table was last rebuilt on 8 August 2026, five days before the parts index, so it reflects a slightly smaller corpus than the parts page does.
- Source
- Primary documents
- adjuster estimates and supplements, counted line by line
- Last built
- 8 August 2026
- page last checked against the data 14 August 2026
Every published panel, with its interval
The dot is the measured replace rate. The bar through it is the 95% Wilson interval: the range of true rates that would produce a sample like this one reasonably often. A long bar means the panel was not seen many times, not that carriers are erratic about it. Read the bars before you read the dots.
The panels worth arguing about
These went both ways in the corpus. The closer to the middle, the more the outcome depended on the estimate rather than the panel, which is precisely where writing a defensible line is worth the effort.
| Panel | Replaced | Repaired | 95% interval | Times seen |
|---|---|---|---|---|
| quarter panel | 46% | 54% | 32% to 60% | 44 |
| deck lid/liftgatethin | 50% | 50% | 29% to 71% | 18 |
| rockerhint only | 55% | 45% | 28% to 79% | 11 |
| door shell | 64% | 36% | 54% to 73% | 103 |
| roofhint only | 67% | 33% | 35% to 88% | 9 |
| bumper cover | 72% | 29% | 64% to 78% | 144 |
| hood | 73% | 27% | 57% to 85% | 37 |
| fender | 77% | 23% | 66% to 85% | 73 |
| windshield/glass | 78% | 22% | 63% to 89% | 37 |
| moulding/trim | 92% | 8% | 86% to 96% | 127 |
| bumper impact bar | 95% | 5% | 82% to 99% | 37 |
| wheel | 95% | 5% | 85% to 99% | 43 |
| taillamp | 96% | 4% | 86% to 99% | 49 |
| mirrorthin | 96% | 4% | 82% to 99% | 27 |
| headlamp | 98% | 2% | 88% to 100% | 43 |
| radiator/condenser | 98% | 2% | 88% to 100% | 43 |
Read the interval column before quoting a row. The closest thing here to an even split is the deck lid/liftgate, replaced 50% of 18 decisions on an interval of 29% to 71%, which straddles half: on that panel the corpus cannot call it either way. Every other row leans further. A rate is not a rule, and an interval that sits wholly above or below half is the corpus telling you the panel has already decided.
The ones that never went the other way
Replaced every single time they appeared in this corpus. Mostly bolt-on and mostly not repairable in any meaningful sense. The number in brackets is how many times each was seen, and the figure after it is the low end of the 95% interval, which is the honest floor on that claim. A panel seen 6 times and replaced 6 times is consistent with a true rate well short of certainty, so treat the short rows as a direction rather than a rule.
- emblem/nameplate (54, at least 93%)
- grille (50, at least 93%)
- fender liner (35, at least 90%)
- bumper absorber (28, at least 88%)
What the shape says
Overall, 83% of these decisions went to replacement. The split is not random: welded structural panels are the ones carriers push to repair, and bolt-on components are the ones they replace without argument. A rocker or a quarter panel is a labor decision about a skilled repair. A grille is a part number.
Which is why the middle of that first table is where an estimate has room to argue at all. On the deck lid/liftgate the split is close to even, so two estimates on similar damage can reasonably land on different sides of it. At the top of the table they cannot: a panel replaced four times in five is not waiting to be persuaded.
What this page cannot tell you is WHY. It records what carriers decided, not how severe each loss was, and severity is the obvious competing explanation for every rate on it. A document that shows the crease at the rear edge and the panel contacting the door plainly argues a different case than one reading "fender damaged" — but that documentation changes the outcome is a claim this corpus cannot test, and it is not made here.
How this was counted
Primary documents. Adjuster estimates and supplements, counted at the line level. These are decisions carriers actually paid for, not opinions about what they would pay for.
- 1Only lines whose operation is R&R (remove and replace) or Rpr (repair) are counted. Refinish, blend and sublet lines are ignored: they are not a repair-or-replace decision.
- 2Each line is matched to a panel by its description, using an ordered pattern list where the specific case wins. A fender liner is not a fender and a wheel opening moulding is not a wheel.
- 3Replace rate is replace decisions divided by all decisions for that panel.
- 4A panel seen fewer times than the publication floor is left off the page entirely rather than shown with a warning.
- 5The 95% interval on each row is a Wilson score interval, which is the reason a panel replaced every single time still shows a range rather than a flat 100%.
Every panel in the file cleared the 6 observation floor, so none is withheld. The published file holds a rate and a sample size per panel and nothing else. No line from any individual claim is in it, and none is on this page. The same table is what the estimator itself is grounded on, so what you are reading here is what the product uses.
What the words mean
- Panel decision
- One estimate line whose operation was replace or repair, matched to a panel. One estimate usually carries several.
- R&R
- Remove and replace. The panel comes off and a new one goes on, so the line carries a part.
- Rpr
- Repair. The panel stays on the car and is straightened or filled, so the line carries labor and no part.
- Replace rate
- Replace decisions divided by all decisions counted for that panel.
- n
- How many decisions the row is computed from.
- 95% interval
- A Wilson score interval. The range of true replace rates that would produce a sample like this one reasonably often. A wide interval means the sample is small, not that the panel is unpredictable.
What this table cannot tell you
- Panel matching is done on the description text, so a line described in a way no pattern recognises is not counted at all. The direction of that error is unknown.
- Decisions are not weighted by severity. A lightly creased fender and a folded one both count once, and the mix of severities is whatever came through the shop.
- There is no per-carrier breakdown here. A rate that pools Progressive with USAA hides any difference between them, and the per-carrier samples are too small to publish separately.
- A rate is not a rule. Every one of these decisions was made against photographs and a car, and the corpus cannot see the car in front of you.
- Every document came through one collision and mechanical shop in Orlando, Florida. Rates, part prices and adjuster behaviour are regional, and nothing here is a national figure.
- The vehicle mix is the shop’s mix, which runs heavily to European and exotic cars. A dataset built in a shop that sees mostly domestic sedans would not look like this one.
- Tested in public against Progressive’s own estimates of record: from photographs alone it landed within 1.2% on a 2023 Nissan Versa and 3.4% on a 2019 Nissan Rogue Sport, and all 13 cars are published at /measured/accuracy.
In plain English
What bolts on gets replaced and what is welded gets argued. The contested middle, fenders, door shells, quarter panels and rockers, is where the estimate itself decides the outcome, and it is also where the intervals are widest.
Table last built 8 August 2026 · page last checked against it 14 August 2026
Run it against a job you already argued
Bring the photographs and the repair order that finally got paid. That is the only test worth anything, and it is what requesting access is for.