How Our Numbers Have Changed

Why anything moved at all

No table anywhere says what incorporating Niwot would cost a household. It has to be built — from the county assessor’s parcel roll, Census survey data, the sales-tax receipts of a comparable town, and a budget nobody has run yet. That takes months. The vote does not wait.

So we published early and kept working. Each revision below is a figure that was estimated becoming one that is derived: an assumption replaced by a measurement, or a number that could not be reproduced from its own inputs rebuilt so that it could.

We are not claiming the numbers have stopped moving. We are claiming something narrower and checkable: they are derived rather than asserted, each is reproducible from a published command, and a figure cannot quietly change on one page while staying put on another. The full derivation, with its error bars sets out how uncertain each input is, and which one carries most of that uncertainty. It is the Census margin on Niwot’s own household income, and it is not close.

The revisions

In date order. Some raised the cost we publish and some lowered it; they are listed the same way either way, because a page that sorted them by whether they suited us would not be worth reading.

1. 12 September

   What incorporation costs a typical household

   $60 a month→$65 a month

   Our model assumed Niwot households buy less telephone and internet service because Niwot has fewer offices than the town we benchmarked against. That is a claim about businesses, applied to households. Removing it raised the figure by about $4 a month. We found it and published the higher number the same day.

2. 14 September

   What incorporation costs a typical household

   $65 a month→$60 a month

   Four more, found in the same week the checks that catch them were built. The largest: we had been treating a per-person spending figure as per-household. We were also taxing groceries, which Niwot would exempt, and charging the full 2.5% downtown where the municipal tax replaces an existing 1% district tax.

3. 14 September

   The range across households

   $41 to $92 a month→$44 to $92 a month

   The two ends of that range had been computed on different assumptions — the floor from a more pessimistic spending estimate, the ceiling from the central one. Put on one basis, the floor rose by $3. Nothing was recomputed and the typical figure did not move; the range stopped answering two questions at once. This one came from outside: a group that disagrees with us had reproduced our table and could not get our floor out of it. As published, it could not be derived; now it can.

4. 16 September

   Household spending, by type of home

   one figure for every dwelling→$473 a year for a house, $244 for a condominium

   Until this date every home in Niwot carried the same household spending tax, whether a $340,000 condominium or a $3.2 million house — overstating the smallest households and understating the largest. The split is measured: Census household size by structure type and the county voter roll independently agree that a condominium household is 0.48 of the town average. The typical figure rose to $63, because the typical dwelling is a house and a house holds more people than average.

5. 21 September

   Share of town revenue paid by Niwot households

   48.3%→50.1%

   One budget line — use tax — was split between households and businesses at a ratio nobody had derived. Deriving it from the line’s own components moved households up, so we now say households carry slightly more of the town’s revenue than we had been. It does not change what anybody pays.

6. 30 September

   Share of town revenue paid by Niwot households

   50.1%→about 49%

   The calculation of what Niwot households buy was written on 12 September and never revised, so it kept two things we had already corrected elsewhere: it scaled a household’s purchases by a per-person income comparison instead of a per-household one, and it counted groceries, which this measure exempts and which the town would therefore never tax. Both made households look like bigger buyers than they are. This correction lowers the figure, so the part it does not fix belongs next to it: those same grocery dollars are still inside the $80 million sales figure we measure the share against, which understates households by roughly half a point. We publish about 49% rather than a decimal for that reason. It does not change what anybody pays — the monthly figure comes from a different calculation and did not move.

7. 27 September

   How far sales tax can fall before the budget is in trouble

   28% before the annual surplus reaches zero→19.8%, permanently, before accumulated reserves run out

   A change of measure, not a correction of arithmetic. The 28% was exact when it was written: in 2030 the budget modelled $2,319,000 of sales tax against a $649,000 surplus. The budget then moved to setting road reserves aside every year as an expense rather than letting them accumulate as surplus — the more cautious treatment — which takes that surplus to about $12,000 and leaves any measure built on it meaningless. The replacement is a harder test: how far sales tax would have to fall and stay fallen, every year to 2040, before the town had spent its accumulated reserves. The smaller number is the tougher test.

8. 28 September

   The range across households

   $28 to $94 a month→$28 to $108 a month

   Two defensible methods spread the same town-wide spending total across homes, and they disagree about the top of the range far more than the bottom. We now publish the central estimate where they agree and the higher figure where they genuinely disagree, which raises the ceiling by $14 and leaves the floor and the typical figure alone. The rule applies at every point rather than end by end — otherwise it becomes a way to pick the convenient number.

9. 28 September

   Private-road costs, Countryside and Cottonwood Park Condos

   $398 and $1,015 per home→$756 and $590 per home

   The two associations share Countryside Drive, and the county’s roads workbook can attribute a road to only one owner, so it gave the whole drive to one of them. Now split by length, as the committee’s own survey does. The total pavement is unchanged to the square foot — this moved who carries the cost, not the cost.

10. 28 September

    Chance the town’s reserves go negative at some point by 2040

    3.4%→2.6%

    The pro forma rebuilt its stress-test simulation and its results table moved, but the summary beneath the table went on quoting the older figure. Nothing about the budget changed; a paragraph was left behind when the table above it moved.

11. 29 September

    Tax on business equipment in the two business parks

    $13,000 a year→$13,000 a year — the county’s roll says $21,500

    Colorado businesses declare their equipment to the County Assessor and pay property tax on it. Our figure for the two business parks came from an earlier estimate that named no source, so we asked Boulder County for the roll. Two things came back. The estimate was ambiguous about whether it meant the equipment’s market value or its taxable value, and one of our own road-district calculations had read it the wrong way, making the tax base for a possible road district look larger than it is. That is corrected; the published 15-to-20-mill range for a road district is unchanged, because every case still rounds to the same whole mill. The second thing is that the county records more equipment than we had counted — $20.6 million against our $13.1 million, worth about $21,500 a year rather than $13,000. We have left our lower figure in the budget. The difference is two tenths of one percent of revenue, and equipment values fall as equipment ages, so a single business installing a new production line can lift the total in a way that will not last.

12. 29 September

    Chance the town’s reserves go negative at some point by 2040

    2.6%→1.93%

    This is the second time in two days that this figure has moved down, so here is plainly what happened. The town’s building department would be run under contract, and the contractor is paid 85% of the permit fees it collects — so when permits fall, its bill falls with them. Our downturn simulation did not know that. It applied one rule to every expense, so it cut permit revenue in a recession and raised the contractor’s bill in the same year, charging the town for both sides of a fee that cannot move that way. We found it by reading the model rather than by running it, published it on 19 September before we had fixed it, and corrected it on 29 September. Every figure it touches moves in the town’s favour, which is exactly why we would rather you heard it from us.

13. 8 October

    County road money the town would receive

    $32,000 a year (2028)→$17,000 a year (2028)

    Boulder County levies a small property tax for roads and bridges, and Colorado law gives each town half of what that levy raises on property inside it. Our budget had counted all of it, and used an older county rate. It now counts half, at the county’s current rate. This one raises the risk figures, so here is everything it moves: reserves in 2040 fall from $6.85 million to $6.61 million; 2030 goes from a $12,000 surplus to a $4,000 shortfall, covered from a reserve of $2.8 million; the chance reserves go negative at some point by 2040 rises from 1.93% to 2.15%; and sales tax could now fall 19.1% for good, rather than 19.8%, before reserves ran out. What a household pays does not change.

14. 9 October

    All-in monthly cost, including a car purchase and remodeling

    $77 a month→$79 a month

    A resident checking our household figures against the budget asked us to reconcile the two, and doing that turned this up. The all-in figure spreads the 2.5% tax on a car purchase over seven years, and our example car was $35,000. The budget’s own vehicle use tax works out to about $150 a household a year, which is a $42,000 car every seven years, so the example now matches the budget. This one raises the figure. The $63 a month for a typical home does not include cars and does not change, and the tax calculator still lets you enter your own car price.

15. 11–12 October

    What a townwide road district would cost

    15 to 20 mills→12 + 6 + 9 + 4 = 31 mills, about $2,160 a year on the median home

    A townwide district is not on offer: no proposal exists, one was defeated at the ballot in 2013, and the 2025 proposal was withdrawn before the ballot. This is what one would cost if it were ever formed. We rebuilt the figure over two days. First the range rose from 15 to 20 mills to 17 to 22 mills. The county’s 2025 method spread a district’s bond over a 25-year road cycle, which only works if the district collects for about four years before construction starts. A district formed now would collect from 2028 and build in 2029, so it has to meet its whole bond payment, the way our own town budget and every neighborhood district are already priced. Then we stopped quoting a range at all. It assumed every property inside the boundary pays, and quoted on its own it read as a price nobody would pay. Every page now builds up from the 2025 proposal’s 12 mills: + 6 for the same roads paid for the way a district has to, borrowing and repaying within twenty years on roads two years older; + 9 once the homes and business parks with the plainest case to be drawn out of the boundary are left out; + 4 once the neighborhoods that would be cheaper on their own leave too. This raises the cost of a district, so here is what it does not change: the neighborhood-district figures, 14 to 47 mills, never used the county’s method and do not move, and our count of the town’s tax base is still about 4% below the budget model’s, which if anything makes every townwide figure slightly high. Corrected along the way: two pages still gave the old figure as a flat number, which our check for superseded figures missed and now catches; a typical neighborhood district was described as about 32 mills, where the published figure is typically 33; the senior homestead exemption under a district was put at $214 a year where its own arithmetic gives $224; and the road information paper said since May that the 2025 proposal would have been over 14 mills on a 20-year cycle, which we could not reproduce and have withdrawn. Our first wording of the + 9 step said the law requires leaving those properties out. It does not: the boundary is drawn by the petitioners and approved by the commissioners.

16. 14 October

    Reserves in 2040 if all three property tax assumptions go wrong at once

    $1.26 million→$1.39 million

    Our stress-testing paper tests three assumptions behind the property tax line: how fast home values grow, how much incorporation lifts them, and the state’s assessment rate. When the budget was corrected on 8 October, the dollar figures in that section were adjusted by hand instead of being rerun, and its probabilities had not been rerun since September. We have now rerun all of them from the model. Most moved in the town’s favour: with all three wrong at once, the chance reserves go negative is 18.55%, not 20.2%; a cut to a 5.5% assessment rate costs $1.86 million of 2040 reserves, not $1.96 million; and values would have to fall 5.4% a year, not 5.1% a year, before reserves ran out. No conclusion in the section changes: reserves stay positive in every case it tests.

17. 14 October

    How much a downturn would ask a council to cut, in the median case

    $463,000 over twelve years→$561,000 over twelve years

    This table on the stress-testing paper had not moved since early September. It now includes the 8 October budget correction and the Sheriff margin added to the spending a council could cut, and its percentages are measured against the $20.6 million the page names, not an older $17.4 million. The median cut is larger. As a share of what a council could cut it is still 2.7%, and in 39% of simulated futures no cut is needed at all. One smaller figure moved the same way: in our stress case, reserves at their lowest are 4.4 times the minimum a town should hold, not 4.5. Every model figure on that paper is now generated from the model itself, and the site will not build if the two disagree.

18. 15 October

    Chance the town’s reserves go negative at some point by 2040

    2.15%→5.10%, about one in twenty

    We found an error in our downturn simulation. It is meant to draw on more of the budget’s contingency as the economy weakens. It was doing the opposite, drawing it down in strong years and holding it back in weak ones. Corrected, the chance that reserves dip below zero at some point by 2040 is about one in twenty, and in the worst one-in-twenty paths reserves fall to about zero at their low point before recovering. The budget itself does not change: every line, the tax rates, the reserves in the budget tables and how far each line could be wrong before reserves run out are as they were. Every simulated figure on the stress-testing paper has been rerun with the correction, including the downtown and events line, which is now never reduced in about 92% of simulated futures.

What stops this happening quietly

A figure moving is not the problem. A figure moving in one place and not another is — that is how an organisation ends up with two numbers for the same thing and no idea which one a resident is holding.

Every published figure that has ever been superseded is recorded, with the date and the reason, in a file that runs against every page, paper, handout and letter each time the site is built. If a retired figure reappears anywhere a reader can reach, the site does not build — not as a warning, as a failure. Where a document has to quote an old figure — a change record like this one, or a filing that is frozen as submitted — that exception is written down with its reason rather than waved through.

The same applies to the models. The household cost, the road costs, the district levies and the budget stress test each have a script that regenerates them, and a check that fails if a published figure has drifted from what the model produces. The budget model itself can be downloaded and run: it reproduces every figure the pro forma and its stress tests publish.

What we have not changed

The two tax rates on the ballot — a 2.5% sales tax and a 4-mill property levy — have never moved. Neither has the structure of the budget they fund. What has moved is our estimate of what those rates cost a particular household, which is a different question and a harder one.

Figures filed with the Election Commission are frozen as submitted. Where an estimate has since moved away from a filed figure we say so rather than quietly restating it; the divergence is set out in the methodology paper.