How prices work

The short version

• Estimates come from completed trades that players report, and from deals completed on Trochus Exchange that both traders confirmed. Asking prices are shown separately and never change an estimate.
• Newer, independently supported reports count more. One extreme report can't move the price much, and repeated or suspicious reports count less.
• Every estimate is a range with a confidence level, the number of supporting reports and when it was last updated. Without enough evidence we show "Not enough data" instead of guessing.
• Nobody can type in a price — not players, not moderators. These are community estimates, not official prices, and they can be wrong.

What counts

Completed trades of a single item type for currency feed the estimate: the unit price is the total paid divided by the quantity. Reports awaiting moderation (pending) and accepted ones count; flagged and rejected ones don't. Only trades from the last 120 days are considered, and you can report trades up to 365 days old.

Asking prices — from active sell listings and from asks players observed in the last 30 days — are what people hope to get, not what trades completed at. Item pages show them as a separate "What sellers are asking" range once at least 3 different people are asking. Each person's latest ask counts once.

Bundles are never split. When a trade included several unknown items (or items on both sides), there's no honest way to know what each one was worth, so the report is kept for the record and excluded from every calculation.

Site-confirmed vs one-sided. A deal that both traders confirmed as completed on Trochus Exchange is stronger evidence than a one-sided report. It is still not verified in the game — the site can't see what happened there, and two accounts could collude — so it gets more weight, not unlimited trust.

Corrections (name, nicknames, category, description, image or price) go to moderators and never change data by themselves.

Weighting

Each eligible report gets a weight. The weights multiply:

Screenshots help moderators accept a report. They stay private: only the person who uploaded them and moderators can open them. You can attach up to 3 images of up to 4 MB each.

Robust statistics

Prices are compared on a logarithmic scale, so "twice as much" and "half as much" count as equally far from the middle. Then:

1. We take the weighted median of the reports — the middle of the evidence, not the average.
2. We measure the typical spread around it (the median absolute deviation, never less than 0.08 on the log scale so a tight cluster doesn't flag every small difference).
3. Reports more than 3.5× that spread away from the median are excluded as outliers. Item pages show which recent reports were excluded.
4. The estimate's typical value is the weighted median of what's left, and the range runs from the 25% to the 75% point of the weighted evidence — the middle half.

Because the median ignores how extreme the far ends are, one wild report can't drag the price around.

Duplicates & manipulation

For site-confirmed deals, the pair of accounts counts as one reporter, so ten deals between the same two friends decay together. New submissions are also checked automatically. Flagged reports are kept out of the estimate until a moderator accepts them:

• Outlier: a unit price more than 3× above or below the current estimate.
• Duplicate: the same person, item and kind on the same day, within 1% of the same price.
• Burst: more than 15 reports by one account in 24 hours. (Accounts can submit at most 30 reports a day in total.)
• Repeated pair: a site deal between two accounts that already have 2 such reports for the item.

Two more checks are informational because the weighting already handles them: reports from accounts younger than 14 days, and someone reporting 3 or more trades with the same trading partner.

Enough data & confidence

An estimate is shown only when the evidence is worth at least 3 independent reports (the "effective sample size", which discounts heavy and near-zero weights), comes from at least 2 different reporters, and its newest report is under 90 days old. Otherwise the item shows "Not enough data".

The spread is the width of the range relative to the typical value. Confidence, sample size and the last update are always shown next to an estimate.

Freshness

When the newest supporting trade is more than 30 days old, the estimate is marked stale and its confidence drops one level. After 90 days without new trades there is no estimate at all. This is applied whenever a page is shown, so an estimate never looks fresher than it is.

Movement review

Every recalculation is stored. If the typical value would move by more than 35% and the trades from the last 14 days are worth fewer than 8 effective reports, the new estimate is held for a moderator and the public one stays as it was. Item pages say when a newer estimate is waiting for review. Dropping to "Not enough data" is the cautious direction and is never held.

What moderators can do

Moderators can:

• accept, reject or flag a report (accepted reports count fully);
• publish or reject an estimate that was held for review;
• freeze an item's estimate while they investigate manipulation, and unfreeze it.

Moderators cannot type in a price, edit the numbers in a report, or decide on their own reports or deals. Every decision is recorded in an audit log, and every published estimate stores which reports it used and how much each one weighed.

Price history

Charts use only estimates that were actually published. A chart appears once there are at least 3 of them spanning 7 days or more. Each estimate holds until the next one — we never draw invented or interpolated history.

Trade checker

The Trade Checker values each side as currency (exactly) plus each item's quantity × its estimated range. The net difference is what you receive minus what you give, using the cautious ends: the low end of what you receive against the high end of what you give, and vice versa.

• Likely favorable / unfavorable only when even those cautious ends agree, and the difference at the typical values is at least 10% of the larger side.
• Roughly fair otherwise — overlapping ranges are never presented as certain.
• Insufficient data when any item has no reliable estimate. That item is never counted as worth zero; the parts that can be valued are shown as partial.

A comparison is as confident as its least certain item, one level lower if any estimate is stale. "Ease of selling" is shown separately from value, and only for items with at least 6 completed trades reported in the last 30 days.

Listing badges

Sell listings get a "Below estimated range" badge only when the price per unit is under the low end of an estimate with at least medium confidence. It's context, never a promise of a bargain.

Known limits

• Several established accounts reporting consistent fake prices could still move an estimate slowly. Movement review, freezes, pair decay and the audit trail are the defences.
• Asking-price context refreshes when an item is recalculated, not the instant a listing is posted.
• Items worth less than one trochus each are rounded up to 1.