Methodology

How are comparables selected?

Four hard filters, applied in order, each rejection recorded:

1. Same property class (a condo never comps a single-family home).
2. Same neighborhood or assessment area (or within a radius when coordinates are available).
3. Living area within ±20% of the subject's square footage.
4. Sold — or assessed, for uniformity arguments — within the county's evidence window (18–24 months for the launch counties).

Survivors are ranked by similarity (square footage first, then year built, beds, baths, lot) and the 8 most similar are kept. Fewer than 3 survivors means no verdict: we refuse to compute a number from a sample that thin.

A rejected comparable is not discarded quietly. Each one is returned with the reason it failed — it is the subject itself, a different property class, outside the area, outside the size window, no recorded sale where a sale was required, or evidence older than the window — and the check displays them. That list is the part a hearing officer will push on hardest, because the fastest way to attack a comparables set is to show which homes were left out of it. A list dominated by wrong-class rejections says something different about your sample than a list dominated by homes that missed the recency window by a month, and a bare median tells you neither.

What is an assessment ratio?

For a market-value argument (New Jersey's standard), each comparable's ratio is its assessed value ÷ its recent sale price. Your home's market value is estimated as the median comparable sale price per square foot × your square footage. The implied fair assessment is the median ratio × that market estimate.

For a uniformity argument (common in Illinois), each comparable's ratio is its assessed value per square foot, and the implied fair assessment is the median of those × your square footage. No sales are needed — the claim is unequal treatment, not wrong market value.

The over-assessment is your assessed value minus the implied fair assessment. Negative means you are assessed below what comparables suggest, and we tell you an appeal could backfire.

What is the coefficient of dispersion (COD)?

COD = 100 × mean(|ratio − median ratio|) ÷ median ratio. It measures how much the comparables disagree with each other. It is the standard uniformity statistic in IAAO ratio studies, where COD ≤ 15 is generally considered acceptable for single-family residential. High dispersion weakens any median, so it lowers the confidence score.

How is confidence scored?

Three components, summing to 0–100:

• Comparable count (0–40): 3 comps score 15, each additional comp adds 5, full marks at 8.
• Dispersion (0–40): COD ≤ 5% scores 40, minus 2.5 points per COD point above 5, floor 0.
• Recency (0–20): median evidence age ≤ 180 days scores 20, tapering to 0 at the county's window edge.

HIGH is 70+, MEDIUM 45–69, LOW below 45.

The recency taper is scored against the county's own window rather than a fixed calendar, so the same eighteen-month-old sale is worth more in a county with a two-year window than in one with an eighteen-month window. That is deliberate: an evidence age only means something relative to what the jurisdiction will look at. It also means the confidence score inherits any weakness in the window itself, and the window is a modelling default in both launch counties rather than a published standard.

When do we say “not worth it”?

A gap under 5.0% of the implied fair assessment is inside normal appraisal noise — boards rarely adjust it. LOW confidence blocks a filing recommendation no matter the gap. Estimated first-year savings that don't clear the filing fee block it too. “Strong case” requires all three: a gap of 10.0% or more, HIGH confidence, and estimated annual overpayment of at least twice the filing fee (minimum $200).

The fee in that comparison is the fee you would pay, not a headline figure. Where a county charges a statutory schedule banded by assessed value, the band is selected from your assessment before the comparison runs — quoting a middle band as though it were the whole schedule understates the fee at the top of the range and lets through a recommendation the fee should have stopped. Each county's schedule is on its county page.

We expect most homes to come back “looks fair.” A tool that tells most users not to file is the tool worth believing when it says the opposite.

Where a statute overrides the gap: New Jersey's Chapter 123

Those thresholds are ours. New Jersey's are the legislature's, and they win. Under Chapter 123 (N.J.S.A. 54:1-35a, 54:3-22(c), 54:51A-6) the Director of Taxation publishes an average ratio — the “Director's Ratio” — for every municipality on 1 April, and a common level range of 15% either side of it, multiplied not added: an average ratio of 78.00% gives a range of 66.30% to 89.70%, because 78 × 0.85 and 78 × 1.15.

Your assessment is divided by your home's true value to give a subject property ratio, and where it lands is the whole answer:

• Inside the range — the board may grant no reduction at all, however over-assessed the comparables make you look. A 13% gap can be entirely non-actionable.
• Above the range — a reduction is due, and it is average ratio × true value. Not the comparables' implied fair assessment. That is the number we show.
• Below the range — the statute raises your assessment to the same figure. Filing actively harms you. This is the outcome, not a risk.

We do not currently publish a New Jersey verdict, because no municipality's Director's Ratio has been read from a primary source and we will not invent one. Every Bergen County check returns “cannot determine” with the reason, rather than falling back to the 5%/10% thresholds above — thresholds that bear no relationship to the test the county board actually applies.

The dollar figure is the softest number on the page

Everything above produces an assessment gap. Turning that into the figure people actually care about — money off next year's bill — takes one more step, and it is the step with the least support behind it. The estimated annual overpayment is the relief multiplied by a single county-level effective rate applied to assessed value: 20.0% in Cook County, 2.3% in Bergen County. Neither has been verified against a primary source, and neither is really one number.

Cook County's burden is the assessment level multiplied by the state equalization factor the Illinois Department of Revenue re-strikes every year, multiplied again by the composite rate of every taxing district your parcel sits in. Three separately-published factors are collapsed into one constant here. New Jersey's general tax rates are struck per municipality, per year, and Bergen County contains roughly 70 municipalities. Both figures are labelled unresolved in the rules files that carry them, and they are the direct multiplier behind the “not worth it” comparison against the filing fee. Read the dollar estimate as an order of magnitude. The gap itself, and the ratio arithmetic behind it, are exact.

Money is integer cents, and rounding happens once

Every currency value in the engine is an integer number of cents, and every rate is basis points — 2.30% is stored as 230, never as 0.023. A float multiplication is rounded to a whole cent at the moment it becomes money and never before, half away from zero, through a single shared rounding function. Ratios in basis points go through the same function, which is why the common level range limits come out as the two-decimal percentages the Director publishes rather than as something ending in a trail of digits. One rounding rule, applied once per derived figure, is what makes a hearing officer's hand-arithmetic agree with ours instead of landing a cent or two away and inviting an argument about which of you is wrong.

Four places the engine refuses to guess

A refusal is a result. Each of these returns nothing rather than something plausible:

• Fewer than 3 surviving comparables. The analysis throws rather than producing a median from one or two homes, and the page shows the rejections instead of a verdict.
• A missing Director's Ratio. In a county whose relief runs on a statutory corridor, no ratio means no verdict — not a fallback to the gap thresholds above, which would recommend appeals the board is required by statute to deny.
• A corridor test with no true value. A uniformity analysis compares assessments to floor area and never produces a market value, so it cannot feed a test that divides assessment by true value. Where the two are mismatched the answer is “cannot determine”, not an approximation.
• Today's date. The engine never reads the wall clock. The as-of date is passed in, which is what makes a check reproducible — the same inputs and the same date always produce the same verdict, including a year from now when the deadline countdown would otherwise have changed the answer.

The engine has no dependencies and makes no network calls, so there is nowhere inside a calculation for a model, a service or an estimate to sit. You can re-run any of it by hand from the numbers shown on the check.

Known limitations

• The estimated annual overpayment applies a county-level estimated effective tax rate — your actual levy varies by municipality, exemptions, and year.
• The current demo runs on a clearly labelled synthetic neighborhood; verdicts on your real parcel require your county's parcel data.
• Condition, renovations, and location factors inside a neighborhood are not modelled. A board may weigh them; bring photos.
• New Jersey's Chapter 123 corridor is modelled, but no municipality's Director's Ratio is encoded yet, so every New Jersey check returns “cannot determine.” Other states with their own statutory relief tests are not modelled at all — Cook County currently runs on the gap thresholds above.
• New Jersey's deadline can extend past 1 April when a municipality mails its notices late; we have no bulk-mailing date input and always show 1 April. Weekend and holiday rollover is not modelled either.

The engine is open to inspection: pure TypeScript, zero dependencies, zero network calls, fully unit-tested. See sources and the changelog. This methodology awaits review by a credentialed property-tax consultant before launch — a launch requirement, not an option.