Refutation Cost
What it costs to check a claim relative to what it cost to make it, which breaks systems when only one of those numbers falls.
Reviewed July 16, 2026Stable
When not to use it
- Where a cheap machine check exists, which puts the system in the favourable regime and makes automation the right answer.
- Where the submissions are genuinely malicious, which is an abuse problem with different remedies.
- Where the checking party is paid and resourced, since the asymmetry is uncomfortable rather than destabilising when the cost sits with someone who consented to it.
Reach for something else instead
- Staged trust — cost of submission falls as contribution history accumulates, keeping the barrier high for bulk and low for the committed.
- Paying the checker — moves the burden to a party who consented to it, which does not remove the asymmetry and does make it sustainable.
- -
Read more on the blog
- Worse than chance means bias, not noisePeople identify high-quality synthetic video 24.5% of the time. A coin would do better, and the reason it beats them is the finding.
- One bug revoked every photo those cameras signedProvenance is the serious answer to synthetic media, it is now an ISO standard shipping in consumer hardware, and the gap between signing and verifying is wider than the adoption figures suggest.
- Fraud doubles in 18 months. Retraction takes 40.The scientific literature is the one place in this territory with a real record, and the record shows the correction machinery growing at less than half the rate of the thing it corrects.
- Slop scores premium 70% of the timeThe first rigorous measurement of machine-generated content in ad buying found it passes every quality check the industry uses, and passes them better than real inventory does.
Further reading
- Stenberg (2026), The end of the curl bug-bounty — the primary account: 87 confirmed vulnerabilities, over $100,000 paid, and a valid rate falling from roughly one in six to 5%.
- Wong et al. (2021), External Validation of a Widely Implemented Proprietary Sepsis Prediction Model — alert volume as a cost borne by the party who did not generate it.
Primary sources, listed so you can check the claims on this page rather than take them on trust.
Where people go wrong
- Reading a falling acceptance rate as declining contributor quality rather than a falling cost of submission.
- Reaching for detection, which classifies origin and leaves the cost exactly where it was.
- Assuming the problem requires bad intent, when reasonable behaviour on all sides is sufficient.
At a glance
FieldFoundations
Testwhat does a claim cost to make, and to check
Distinctionfrom cheap-verification problems, where a machine check exists
Symptomvolume, not quality
Remedyfriction priced to the honest case, not detection
DifficultyBeginner → Intermediate
Where this sits
A starting point. Nothing needs to come before it.
0Levelstarting point
0Needs firstnothing
0Opens upnothing further
1Areastays here
Computed from the prerequisite graph, not assigned. How this works