Platform
Repair or replace, by what actually happened
The single most consequential line in an estimate, and most software guesses at it.
What was measured
912 decisions across 20 published panels, read out of carrier estimates. Overall, 83% of the measured decisions were replace rather than repair. The rate varies sharply by panel, which is the useful part.
Why the minimum sample is what it is
A panel is published only once it has at least 6 decisions behind it. Dropping that threshold would multiply the panel count several times over immediately, and it is the most tempting and most dishonest lever available in this dataset. It stays where it is and the panel count grows by the sample growing.
What this changes in practice
Writing repair on a panel that carriers replace nine times out of ten produces a supplement and a delay. Writing replace on a panel that is routinely repaired produces an argument. Knowing the actual rate before the first write is worth more than either.
Before you ask
Does it decide for me?
No. It shows what carriers did on that panel with the sample size attached, and you decide. The car in front of you is not a statistic.
Why are some panels missing?
Because they have fewer than 6 decisions behind them so far. A missing panel is an honest gap rather than an oversight.
Where can I see the table?
The full measured table with a rate and a sample size on every row is published at /measured/repair-or-replace.
Run it on damage you have already written
Accounts are created by hand, usually within a business day. Bring your own photographs and a repair order you have already closed, so you can see how close it lands before you decide. No card, no trial timer.