Data policy

1. Sources

Torn's official API is the source for prices. Completed auction sales are read from api.torn.com and stored as they arrived. Nothing is scraped from Torn's website, guessed at, or filled in from another site, and a field the API did not return is shown as missing rather than estimated.

The Elimination board has a second source, and it is a third party. Its battle-stat figures are estimates from FFScouter, requested with a key belonging to Cortex. They model a player's stats rather than measure them. That is why the board labels them estimates, and why a fair fight figure derived from them is shown to two decimals and no further. Everything else on that board, including scores, lives, rosters, status and levels, comes from Torn's API as the prices do.

On the database and the companion script, your browser talks to exactly two places: this site, and the Cortex price service that answers it. There is no analytics service, no tracking pixel, no advertising network, and no font loaded from another host. The Elimination board differs in one way: opening its Activity tab fetches a charting library from a public CDN, pinned to an exact version and checked against an integrity hash. Its target list fetches nothing external.

2. Coverage

Over 300,000 completed sales, across 145 ranked-war items, drawn from more than 220,000 individual copies. The tracked set is 101 weapons and 44 armor pieces.

Most items have recorded history from 2023 onward. Twenty-seven of the 145 reach back further, the earliest to February 2016, which is further than Torn's own auction history goes.

New sales usually land within minutes. The database prints the exact totals and the age of its newest sale on its front page, and those live figures supersede the rounded ones above.

The Elimination board covers one competition and about 22,000 enrolled players, sampled every minute while it runs. It prints its own coverage the same way: how many rosters it holds against how many exist, how old the oldest is, and whether Torn currently has the roster endpoint open. A board that is behind says so instead of looking current.

Its activity figures come from two methods with different reach. A sample of sixty players a team is measured through FFScouter, which costs one request per player. Every roster sweep already reports when each member last acted, so differencing consecutive sweeps covers the whole field at no extra cost. The board states which method is behind a figure, because the sampled one carries a much wider error bar.

3. How collection behaves

Paced well under the limit

A background job runs every two minutes and makes at most forty requests per run, spaced a second apart. Torn allows a hundred a minute, and collection stays near a fifth of that.

The companion script does not poll

It prices only the page you opened, spacing its own requests 700 milliseconds apart, and only while that tab is visible. Prices are reused for six hours, so a second visit asks for almost nothing.

Traffic is labelled

Every request the collection job sends carries comment=Cortex, so Torn can identify it. None of it uses your key, so none of it appears in your API log.

Reads only, from Torn

The collection job only ever reads. It cannot spend money, send items, start a crime, or change anything in any player’s account.

4. How a price is produced

A published price is a median over a group of comparable sales, not the last sale and not an average of everything. These are the steps, in order.

Grouped exactly

Sales are grouped by item, perk, rarity and the exact perk roll they carried. A 25% Conserve and a 29% Conserve are different markets, and are never averaged together.

Thin groups stay blank

A group needs at least four recorded sales before any price is published for it. Below that the page shows nothing rather than a guess.

Adjusted for quality

Copies sell at different quality levels, so each sale is normalised to a reference quality for its group before the median is taken.

Weighted towards recent sales

A sale from last week counts for more than one from last year, on a 45-day half-life anchored to the group’s own most recent sale. A market that stopped trading holds still instead of drifting.

Ordered so a better roll is never cheaper

Prices across the rolls of one perk are reconciled so a stronger roll cannot read lower than a weaker one, which thin evidence would otherwise produce.

Published with a range

The figure describes a group, not one sale, so a range is shown beside it. The range narrows as evidence accumulates and stops narrowing at a measured floor, because some spread is real rather than a shortage of data.

Outliers are removed before the median is taken. The number of sales behind any figure is shown next to it, so a price resting on four sales is distinguishable from one resting on four hundred.

5. What this data cannot tell you

History, not the live market

Torn’s API publishes completed sales and never open auctions, so nothing here shows what is listed right now or what the current bid is. That is why the companion script reads the auction page instead: the live side has to come from a page you opened yourself.

Ranked-war gear only

About 145 weapons and armor pieces are tracked, not the whole of Torn. An item with no perk, or below yellow rarity, is out of scope deliberately.

Two-perk sales are priced apart

A copy carrying two perks sells for more than either alone, so those sales are kept out of single-perk prices and shown as their own market. Folding them together once inflated prices by more than twenty per cent.

Copy counts are a floor

A sale recorded before Cortex tracked individual item ids counts as its own copy. Any figure for distinct copies understates resale by construction, and never invents it.

An estimate, not a valuation

Every price here is what copies have sold for at auction. It is not an appraisal, not a guarantee of what yours will fetch, and there is no record of auctions that failed to sell, because Torn does not publish those.

6. Who pays for the data

Not you. There is no credential field anywhere on this site or in the companion script, and neither carries the code to use one.

One key, belonging to Cortex, runs the collection job that reads Torn and fills the auction database. Every visitor then reads that same database, so the cost of gathering the data is paid once rather than once per person. That is why it can be free and ask nothing of you.

The Elimination board additionally uses Torn API keys donated by a few players, so its rosters refresh faster than one account's rate limit permits. How those keys are stored and removed is set out in the privacy policy.

7. Corrections

If a price here contradicts a sale you know happened, report it and it will be checked against the underlying rows. Message Shield [3516052] in Torn.

8. Changes to this policy

The date at the top of this page is the date it last changed. When the sources, the method or the limits change, this document changes in the same release, and the change is listed in the public changelog.