Every market, read as one book.
Indexes, relative strength, microstructure and the engine’s own science — 1,129,691 lots, 0 sub-markets tracked.
The index laboratory
Not enough settled history to chart an index for this market yet — it lights up as quarters accrue.
The reads the engine stands behindVerified moversprice movement · 95% confidence
No maker in the market clears the 95%-confidence bar yet — we only print a move the data resolves.
Every lot on the book
The deep pools
The doctrine
Honesty, built in
Every figure on this desk names its method and base, and the same laws bind the engine underneath: where the data cannot clear its own bar, the desk abstains — and the abstention is printed. Three of the gates, drawn.
The abstentionBelow the floors — too few sales in a cell, a confidence interval that will not resolve the sign — the engine prints a dash rather than dress up a number it cannot back. Suppressed cells are the method working.
The calibrationWhat a flag is worth is refit from every nightly replay, recency-weighted and shrunk. The 10×+ bucket drops on purpose: extreme comp gaps under-deliver, and the curve says so instead of extrapolating.
The printed-bid gateA bidding read rides only a bid that was actually printed on an exposed book — never inferred from an ask. Where a house shows no live bidding, the desk prints the ask and says so.
Methods
Price movementhedonic log-price regression per market (reference/form/size/house controls, IRLS-weighted, quarterly time coefficients) and same-object repeat-sale fits where pairs allow; both abstain below density, dominance and CI gates, and a ±plausibility ceiling drops implied compound moves the controls cannot defend.
The recordevery flag replayed against realized results at both hammer and all-in bases — never backfilled.
Calibrationbeat-rates refit nightly, recency-weighted and shrunk.
Figuresprint their n and basis; suppressed cells mean the data did not clear the bar.
Datapublic auction results across 16 houses.