How to use Valuate Everything

The short version

1. 1Enter the listings you have seen

   On Market data: one row per listing, its price and the features that drive it, such as the year, kilometres or land size. Or start from a Quick Start.

2. 2Add what you want to buy

   On Items to buy: each item's features and its asking price. Assumptions sets the year, inflation and the model. Press ▶ Valuate.

3. 3Read the fair price

   The sentence at the top names the best value and how far each asking price sits from its fair price. The chart and the tables show the model behind it.

Nothing is uploaded. The model is fitted in your browser, on your listings only. This guide is general information, not financial advice.

The screen

Inputs on the left, the answer beside them, the chart and the tables under it. Point at a part in the list to see where it is.

Top right, Guide opens this page. The tool has no tour: this page is the walk through it.

Quick Start

Four worked examples, built from real listings and sales gathered in October 2026: a used Camry Hybrid (18 dealer listings), a Perth house (23 sales in Morley, 2023 to 2026), a Sony A7 Camera body (24 listings, in pounds) and a Hotel night in Perth (22 rooms). Each brings two items to check.

On a phone, press Full screen under a demo and turn the phone sideways to read it.

• The page opens on the Camry. Its first car, a 2021 Ascent Sport with 52,771 km, asks $32,369 against a fair price of $34,765: 6.9% under.
• Prices move. The examples are a snapshot of the market on the day they were gathered, not a live feed.

Market data

The listings the model learns from, one row each: a price and the features that drive it. Columns says what each one is; Listings holds the rows.

• More rows and columns. + Add listing under the table, ⤢ to enlarge it, + Add column for another feature, the bin to delete one.
• A blank is never guessed. A row with a blank or unreadable cell is left out, and the warning at the foot of the inputs counts it. A blank Yes/No reads as No, a blank data year as the current year.
• No data year? Delete that column and every listing is taken as seen this year, at its listed price. + Add data year brings it back.

Items to buy

The listings you are weighing up. Each is priced by the model and judged against its fair price, but never added to the model.

• Every feature, or no price. An item missing a feature is listed in the table as missing it, and left unpriced.
• Edits wait for Valuate. The results dim and the button is ringed until you press it, so you can enter every listing and item first.

Assumptions

The money the prices are in, and the shape of the model. Current year and Inflation show only when the listings have a Data year column.

• Which model? Assumptions, in the maths says what each one assumes about how a price falls with age or kilometres.

Read the answer

One sentence, four figures, then the tables under the chart, each behind its chevron.

• A minus is good news. A negative gap means the asking price is under the model's fair price.
• Warnings sit at the foot of the inputs: rows left out, a feature that never changes, too few listings for the terms, an item outside the listings' range.

The chart

The fitted price along one feature, with every other feature held at one value. The options beside it change how it reads, never the answer.

• Each dot keeps its own miss. A listing is moved to the held values before it is drawn, so it sits off the line by exactly what the model missed it by. The chart, in the maths shows how.
• Zoom. Drag across the 2D chart to zoom, and the price axis follows; ⟳ puts it back, or turns the 3D view back.

Save, export and share

The tool keeps your listings, items and settings in this browser between visits. The rest makes something to keep or send.

• Your own rows. CSV under Listings saves them as you entered them (⬇ CSV) or as an empty ⬇ Template with your columns.
• A link carries your listings. Nothing is uploaded, but anyone with the link can read them.

The model in one picture

The tool fits one regression to your listings: the price explained as the sum of what each feature is worth, the way an appraiser prices a house by its rooms and land. It then prices each item you want with it. The headline is the gap: how far an asking price sits from that fair price.

• P̂the price model, fitted to the listings only
• x_itemthe item's features, a Year as its age now
• Athe item's asking price
• Fits fair price, in the current year's money
• RMSE the typical error: how far listings sit from the model
• sthe same typical error, on the log of the price

• A fair price is the market's average, not a valuation. It is what an item with these features typically lists for among your listings. A listing below the line may be below it for a reason the columns cannot see.
• Read a gap against the typical error. The Camry's −$2,396 is 1.3 typical errors: more than listings usually scatter, but not by much. The house's −$39,996 is a quarter of one: inside the scatter, so the model cannot call it a bargain.
• How well features explain price varies. A camera body is priced almost entirely by its release year, shutter count and condition (R² 0.98). A house is not: street, outlook and renovation are in none of the columns (0.43).

The range is not a confidence interval. It is the fair price give or take one typical error: how far real listings scatter around the line. It does not allow for the line itself being uncertain, which it is with few listings.

Market data: what the model learns from

Each listing becomes one point: a price in today's money and a row of features. The model is the line, or plane, that sits closest to all the points at once: least squares.

• P_ithe price of listing i, as listed
• Y_iits data year (blank: the current year)
• Cthe current year
• πinflation, a year
• x_ijfeature j of listing i
• μ_j, σ_jthat feature's mean and spread over the listings
• nthe number of complete listings
• λthe ridge penalty; 0 for every other model
• βthe fitted effects, turned back into price per unit

• Why ages, not years. Buyers pay for how old an item is, not for the number on it. Turning a year into an age at the data year lets a 2024 sale and a 2026 listing of different cars be compared.
• Why one scale. Kilometres run into tens of thousands and a Yes/No is 0 or 1. Standardising puts them on one footing for the fit (and for Ridge's penalty); the tool then turns every effect back into a price per km, per year of age, per Yes.
• Today's money. The Perth house's first sale, $515,000 in 2023, counts as $515,000 × 1.03³ = $562,754 in 2026 at 3% a year, as Listings against the model shows in Today.
• What cannot be fitted. A feature the same in every listing is dropped, as it explains nothing. Two features that move together exactly cannot be told apart: remove one, or use Ridge.

Inflation is one rate, not the market's own growth. In the Perth house, three of the six 2023 sales sit 23% to 29% under the line, a sign these prices rose faster than 3% a year. At 6% the fit improves (R² 0.43 to 0.52); at 10% it is 0.63, and 11 Pitt Court turns from 5.8% over its fair price to 6.1% under it.

Items to buy: priced, never fitted

An item goes through the same features as the listings, aged from the current year, and the fitted model gives its price. Its asking price is then compared with that.

• Cthe current year
• x_itemthe item's features
• Aits asking price
• Fits fair price

• Outside the listings, the model guesses. When an item's feature lies beyond the lowest or highest in the listings, the fair price is an extrapolation and the tool says so. A straight line runs on for ever: far enough out, Linear can give a car no positive price, which the tool flags too.
• Gap and Gap %. The Camry's two cars ask $2,396 and $1,928 under their fair prices: −6.9% and −5.8%. The percentage ranks them, so a dearer item is not favoured for its bigger dollar gap.

Assumptions: money, time and the model's shape

The modelling approach is a statement about how a price responds to a feature: by a fixed amount, by a fixed percentage, by an amount that bends, or by an amount pulled toward the average.

• αthe intercept: the price with every feature at zero
• α_jthe effect of feature j, per unit
• α_jjits bend, per unit squared
• β_jthe same effect on the standardised scale
• λthe ridge penalty

• Linear or Log-linear for depreciation. Linear takes the same dollars off for each year, so a dear item and a cheap one lose alike. Log-linear takes the same share off, so the dollars shrink as the price falls, as resale prices tend to. On the Camry they agree within $60 on the first car.
• Overfitting, in two numbers. Quadratic fits the Camry's 18 listings a little closer (R² 0.68 to 0.69), but the typical error, corrected for the two extra terms, rises from $1,863 to $1,972: the bends follow the noise in these listings, not a pattern the next listing will share. The tool warns when there are fewer than three listings per term, plus three.
• Why Ridge exists. With few listings, or features that move together (older cars have more km), least squares can split one effect between them unsteadily. Ridge gives up a little fit for steadier effects. On the Perth house at λ 0.10 it lifts Bathrooms from $2,786 to $12,477 a bathroom and trims Bedrooms from $119,273 to $103,839.

A higher R² is not a better model. More terms always fit the listings you have more closely. Judge a model by its typical error, which pays for each term, and by whether its effects make sense.

The fit, the range and the chart

How the four figures and the chart are worked out from the fitted model.

• P̂_ithe model's price for listing i
• mthe average listing, in today's money
• kthe terms fitted: 3 for the Camry, 5 with Quadratic
• others held every feature not on an axis at the Hold others at value

• R² is a share. 0.68 means the features explain 68% of how much the Camry listings differ in price. The rest is what the columns do not capture.
• The typical error pays for every term. Dividing by n − k − 1 rather than n makes a model with more terms show a larger error unless they earn their place. For the Camry it is ±$1,863, 5.7% of the average listing.
• Standardised weight. In Price model terms, the effect of one typical spread of a feature, so features in different units compare: for the Camry, age −$1,816, Above Ascent grade +$1,646, Odometer −$892.
• A dot is like for like. Each listing is moved from its own features to the held ones by the model, keeping its miss, so its distance from the line is exactly its Miss in Listings against the model (for Log-linear, its Miss %).

Worked example: Camry Hybrid

The Quick Start as it loads, worked by hand: 18 listings, Linear regression, current year 2026. The first car to check is a 2021 Ascent Sport with 52,771 km, asking $32,369.

• The tool shows the same. Fair price $34,765, range $32,903 to $36,628; Best value −6.9%, −$2,396 against fair; Model fit (R²) 0.68 from 18 listings and 3 terms.
• What the coefficients say. Among these listings, each year of age takes $1,033 off, each 10,000 km $362, and Above Ascent grade adds $3,675, with the other features held where they are.
• The second car. The 2020 SL with 61,444 km is a year older and has 8,673 km more: $33,418 fair, asking $31,490, −5.8%. So the 2021 Ascent Sport is the better buy on these listings.

What moves it most

Each bar changes one thing by a step worth arguing about, keeps the rest of the Camry example where it was, and records how far the first car's fair price moves.

• The grade decides the verdict. Untick Above Ascent grade and the fair price falls to $31,091, so the same asking price is $1,278 over it. Only 5 of the 18 listings have the grade, so check the car is the grade it says before trusting the discount.
• The model choice barely matters here; the penalty does. Log-linear and Quadratic move the fair price by $60 and $247. Ridge at λ 1.00 pulls the grade's worth toward the average and cuts the discount to $1,040, inside one typical error.
• One listing can move it by $500. Leaving out row 1, a 2021 SL with 82,984 km at $35,980, takes $532 off. With 18 listings, a single odd one counts.
• Inflation only acts on older listings. Every Camry listing is from 2026, so it moves nothing. On the Perth house, whose sales start in 2023, a point either way moves 14 Netley Street's fair price by −$13,270 or +$13,459. And the same Camry listings a year on (current year 2027) are lifted 3% while the car ages a year: −$21.
• For a decision. The discount survives every step but the grade. It is real on these listings, but at 1.3 typical errors it is a reason to look closer, not proof of a bargain.

What it leaves out

A regression on the listings you entered, nothing more. What this assumes, under the tool's answer, lists what applies to yours.

• Only the columns. Condition, history, colour, location quirks and anything else you did not enter are invisible to it.
• Your listings are the market. The model only knows these prices. A handful of listings, or ones picked from one end of the market, give a fair price for that sample only.
• Asking and sold prices are different things. The Camry example uses dealer asking prices, the house sold prices. A fair price from asking prices is what similar items are listed at, not what they sell for.
• Effects add up; they do not interact. Each feature is worth the same amount (for Log-linear, the same share) whatever the others are: km cost a 2019 car what they cost a 2024 one. Quadratic terms bend a single feature, never a pair.
• Whole years. An age is the data year less the model year; the month is not seen. Inflation is one rate for every year and every listing.
• The current year follows the calendar. Open the tool next year and it restates the same listings in that year's money and ages the items a year, so a figure can shift a little.
• No outlier is removed. A typo or a one-off sale pulls the line toward it. Listings against the model shows each miss, so a very large one is worth a second look.

When not to lean on it. Few listings for the features you use, an item outside the listings' range, a gap inside the typical error, or a market that moved faster than your inflation rate. Then treat the fair price as a starting point for negotiation, not a valuation.