Independent estimators will disagree. That is the point of having them. A single number hides its uncertainty; several numbers side by side show it. The question is how to turn them into one figure that can be paid, without letting anyone choose the answer after seeing the estimates.
The figure: the median
The paid figure is the median of the counted estimates: the middle one when they are lined up from lowest to highest. The rule is fixed in advance and cannot be changed for a particular cycle.
The median is used because it is hard to move. To shift it, a party would have to move several estimates on the same side, not just one. An estimator far out on its own changes the median little, which protects against a single bad or bent estimate. If the outlier is right, the later audit will show it, and its track record will carry the weight next time.
What is published with every figure
- every accredited estimate, counted or not, with the estimator’s name;
- which estimates were drawn into the counted set, and the record of the draw;
- the median, the full range, and the middle half of the range;
- any merges of correlated estimators, and the evidence for them;
- each estimator’s method notes, and what changed since the last cycle.
Anyone can recompute the figure from what is published. If the published figure and the recomputed one differ, that alone is grounds for review.
A wide spread
When the estimates disagree by more than a set amount, the figure is still the median, but it is flagged as wide. A wide flag sets three things in motion:
- Each estimator explains its position in a short published note: what it sees that the others may not.
- The audit body prioritizes that figure in its next field check, so the facts arrive sooner.
- The accreditor checks methods, never results, for errors that could explain the gap.
A wide spread often means something real is happening that the methods disagree about: a housing market turning, a new kind of care need, a region changing fast. Flagging it brings that to the surface instead of averaging it away.
No sudden swings
The floor is something people plan their lives around, so a change in the figure is phased in. A rise reaches people promptly, because waiting would leave them short. A fall is spread over several months, so no one is hit by a sudden cut. Lowering what a floor-standard home must provide is a separate matter: that weakens the guarantee itself, and only the hardest process can do it (how the guarantees change).
Too few independent estimates
The median needs at least five independent estimates after correlated estimators are merged. If fewer arrive, because estimators were driven out, defunded, or failed the independence tests, the figure is flagged as unreliable and the last sound figure stands, adjusted only for the general change in prices. It stays that way until enough independent estimates return.
The fallback is deliberately unattractive to anyone hoping to exploit it. Driving estimators out does not hand the figure to whoever remains; it freezes the figure where it was and draws attention to the cause.
An estimator right alone
Sometimes one estimator sees what the others miss. The system is built to reward that, not to punish it. The outlier’s estimate is published beside the median, its explanation is published if the spread was wide, and when the audit proves it right, its accuracy reward and its track record reflect it. Over a few cycles, other estimators adopt what it saw, or keep losing to it.
That is how the set learns. Not by forcing agreement, but by keeping every estimate visible long enough for reality to settle who was right.