Every takeoff we can check against a public agency's own published quantities,
including the lines we got wrong.
We publish the misses because a record without them is an advertisement. Each source
below is a public document — the same numbers are available to you, and the comparison can be re-run.
State DOT bridge replacement
Compared 2026-08-02 · 32 pay items scored of 55 in the tabulation
21 of 32 within ±15% — of those, 18 within ±1%
The full 119-sheet set was taken off blind — the sheets carrying the agency's own quantities were withheld from the agents — then compared line by line against the published bid tabulation. The item mapping is ours and is open to argument: a different mapping gives different numbers.
Matched
Drilled shafts, all three diameters (18/24/36 in) — 124 / 216 / 186 LF, exact
Guardrail transitions, end treatments and downstream anchors — counts exact
Missed — and why
RIPRAP-75% — Two of the three sites reported the source as insufficient for the riprap extent: thickness and slope are printed, the boundary is not dimensioned. This is a coverage gap wearing the clothes of a measurement error.
SET (TY II)-50% — We found two; the agency pays for four. The 48 in pipe and its end treatments sit on sheets outside the list we allowed the site to read.
REMOV STR (PIPE)-38.8% — 131 vs 214 LF. The site that measured it had already flagged that the drawings contradict themselves — the removal note says 131 LF while the profile shows 49.4 LF. Neither figure matches the agency's.
BARRICADES-93.8% — 1 vs 16 months. No sheet in the set states the contract duration, so we carried one month as a placeholder and said so rather than guessing a number.
What this does not prove
It proves nothing about pricing — this job was measured, not priced. It does not prove completeness: 16 of the 48 mapped items were quantities the drawings themselves printed and we copied, so they are excluded from the score rather than counted as wins.
Limits we created ourselves
Seven items where we produced no number came mostly from a decision of ours, not a failure of measurement: we restricted which sheets each site could read, to control cost and to keep the answer sheets out of view. The pier, cap and column concrete dimensions live on standard sheets outside that list, so all three sites correctly reported the source as insufficient rather than inventing a dimension.
State agency outdoor pavilion
Compared 2026-08-27 · our total against 7 real bids at a public opening
Against the median of 7 real bids: -45.4% — below every bid (bids $497,000 to $743,000)
Blind by construction of time: our workbook was produced 21 Jul 2026, the state opened bids 11 Aug 2026. Our total is compared against the range of the seven real bids, not against any single number, because the bidders' own disagreement is the honest yardstick.
Matched
The screening band we declare (-50/+100) contained 4 of the 7 real bids, including the median.
Missed — and why
total (direct-only) vs 7-bid market range-45.4% vs median; below every bid; 0.74x the bidders' own spread below the range — Our direct-only figure sat below every one of seven bids, 45% under the median. Two causes were identifiable and independent of the bids: the total carried no overhead/site-running cost at all, and labour was priced at a private-work base while Missouri pays state prevailing wage on public jobs.
What this does not prove
It does not verify any line quantity (the tabulation is lump-sum). It does not tell whether the winner profits. The figure tested was our DIRECT cost only — no overhead, no prevailing-wage labour — which is exactly what the miss measures.
Limits we created ourselves
The total we tested carried no indirect cost and priced labour at a private-work base on a prevailing-wage state job — both our own modelling choices, and both named before any correction was applied.
State agency building repairs
Compared 2026-08-27 · our total against 8 real bids at a public opening
Against the median of 8 real bids: -42.3% — below every bid (bids $2,599,400 to $3,590,998)
Same blind-by-time construction as the pavilion entry: workbook 21 Jul, bids opened 11 Aug. Compared against the eight responsive bids; one non-responsive partial-scope submission is excluded for the reason the state itself printed.
Matched
The screening band we declare (-50/+100) contained 7 of the 8 real bids, including the median.
Missed — and why
total (direct-only) vs 8-bid market range-42.3% vs median; below every bid; 0.93x the bidders' spread below the range — Our figure sat below all eight of the bids, 42% under the median — nearly the same offset as the pavilion, which is the finding: a consistent, systematic price-level gap with two named causes, not measurement noise.
What this does not prove
No line quantities verified (lump-sum tab). Base-bid scope alignment with our alternates split is assumed, not proven. The figure tested was direct-only, as above.
Limits we created ourselves
Same two self-made limits as the pavilion entry: no indirect layer, private-work wage base on a prevailing-wage job.
State agency outdoor pavilion, re-measured
Compared 2026-08-27 · our total against 7 real bids at a public opening
Against the median of 7 real bids: -9.1% — inside the bid range (bids $497,000 to $743,000)
The two corrections have sources independent of the bids: the prevailing-wage factor comes from Missouri's own Annual Wage Order 33 for Phelps County, and the preliminaries allowance from a published size-tier band. Rebuilt through the real pipeline and computed by Excel itself.
Matched
Against seven real bids of $497,000-$743,000 our $522,350 lands inside the range, 9.1% under the median, with two real bidders below us.
Missed — and why
claiming this landing as accuracycalibration caveat, not a numeric miss — Recorded as a caveat rather than a numeric miss: applying corrections after seeing a comparison is how overfitting starts, so the parameters are frozen and the next unseen tabulation is the real test.
What this does not prove
This round is calibration, not a second blind test: the parameters have independent sources, but the decision to apply them came after seeing the first comparison. The proof is the next unseen bid tabulation, run with these parameters frozen. It also does not validate the factor per trade — an unweighted seven-trade mean was used.
Limits we created ourselves
The unweighted trade mean and the choice of correction moment are both ours; the entry says so rather than presenting the landing as accuracy.
State agency building repairs, re-measured
Compared 2026-08-27 · our total against 8 real bids at a public opening
Against the median of 8 real bids: -24.2% — below every bid (bids $2,599,400 to $3,590,998)
Identical corrections and pipeline as the pavilion re-measure (Cole County wage order, same allowance model).
Matched
The declared band (-50/+100) contained all 8 real bids including the median; the gap narrowed from -42.3 to -24.2 points against the median.
Missed — and why
total vs 8-bid range after corrections (renovation)below the range of eight bids by 0.40x their own spread — The same sourced corrections put the new-build job inside its bid range but left this renovation short of the range by 24% against the median. The difference between the two jobs is the measurement: what remains unpriced is the cost of working in an occupied existing building (demolition sequencing, protection, phasing), which new-construction rates do not carry.
What this does not prove
Same calibration caveat as the pavilion re-measure. Additionally, the remaining gap is attributed to renovation-difficulty by reasoning, not yet by evidence.
Limits we created ourselves
We did not add a third factor to close the remaining gap, precisely because the only number available at that moment would have been the gap itself. A renovation-difficulty layer enters only with a source (P-405 Table 4-1) and gets tested on the next unseen renovation tabulation.
Checks that proved nothing
Listed because leaving them out would make the record look better than it is.
School district bid results (totals only) — The published document gives bid totals per bidder and no line-item quantities, so a difference in total cannot tell a measurement error apart from a pricing one. Listed here because leaving it out would make this record look better than it is.
Barracks renovation — blind prediction — Line-level quantities (MO tabs are lump sum); winner's margin. One job is one sample either way.
Park cabins — blind prediction — See above.
Laboratory HVAC replacement — blind prediction — See above.
Dining facility — blind prediction — It does not prove our quantities are right. The award proves what the market charged for this building, and our cost band comes from a published per-SF rate, not from our own measured lines — 103 of 236 of which carry no rate at all. A band that lands could land for the wrong reason: the rate could be right while a quantity is wrong, and the two errors cannot be separated from the outside. What the quantities themselves are worth can only be shown by somebody re-measuring the same drawings, which has never happened yet.