The most common complaint about the limiteds trade checker is that it hedges. Load two sides of a deal, and if they land within ten percent of each other, the verdict is "even" - not "you're up 4.2%," not a decimal, just even. A few users have asked for the exact edge. The refusal is not a missing feature. It's the most carefully considered decision in the tool.
The number you want doesn't exist
To referee a 4% difference, the underlying values would need to be accurate to better than 4%. Ours aren't. Nobody's are. The checker prices items from our own tracker data, a blend of recent-sale prices and RAP, and every input in that blend carries error you cannot wish away.
Start with the lag problem. RAP is an average of sales that already happened, so even when it is current it trails a market that has kept moving; and until September 11, 2026 our own figures were worse than that, read from a legacy Roblox feed that had stopped recording in early 2025. A banner on the data doesn't shrink the error bar - it just tells you the bar exists. Then add the ordinary noise: thin trading on mid-tier items, single sales that swing an average, the spread between what a patient seller and a desperate one accepts in the same week. Stack those up and the real uncertainty on any individual valuation is comfortably wider than the margins people want adjudicated.
A tool that answers "you're up 3.8%" from inputs like that isn't measuring 3.8%. It's laundering noise into a number with a decimal point, and the decimal point is doing all the persuading.
Precision is a costume
The interesting part is that the sharper the verdict, the less trustworthy the tool, and almost everyone's instinct runs the other way. A verdict with two decimals reads as rigor. A verdict of "even" reads as a shrug. But the two-decimal answer requires precision the data cannot supply, so displaying it is a claim the tool can't back - a costume rigor wears. The shrug is the measurement.
This is the same reasoning that keeps value-pending items out of the checker's totals entirely. The modern UGC collectibles have no trustworthy price signal (the listed floors are troll numbers as often as not), so the checker excludes them and says so, rather than guessing and folding the guess silently into your total. You can see an excluded item and allow for it; you cannot see a guess folded into a total.
What the band actually does for you
The practical effect of the ten-percent band is that the checker only speaks when the data can carry the claim. When it says you're winning a trade, the gap is big enough to survive every error source we know about. That verdict means something precisely because the tool stays quiet on the close ones. A checker that calls every trade is like a friend who has a strong opinion on everything - you learn to stop asking.
And the close calls it declines to referee? Those are judgment calls, and the judgment isn't ours to make. Inside the band, the deciding factors are the ones no valuation captures: which item you'd rather hold, which one moves faster when you want out, whether the other trader is someone you'll deal with again. Demand and liquidity beat a two-percent paper edge every time the paper edge is smaller than the paper's own margin of error.
How the ten percent is measured
The checker divides the gap between the two sides by the larger side's total. Take two hypothetical totals: 50,000 on one side and 45,000 on the other. The gap is 5,000, exactly 10% of 50,000, so the verdict is even. Drop the smaller side to 44,999 and the gap is a hair over ten percent, so the checker names the heavier side. Measuring against the larger side keeps the verdict symmetric: loading a deal as A-for-B or as B-for-A gives the same answer. It also means the bigger side can be up to about 11% larger than the smaller one (50,000 is 11.1% more than 45,000) before the checker speaks. Value-pending items sit outside both totals.
Ten percent, checked against our own data
When we wrote this in August, ten percent was a judgment. The tracker now has four weeks of snapshots from the current Roblox source to test it against. Between September 11 and October 9, 2026, the RAP of the median tracked limited moved 6.6%. Of the 172 items, 59 moved more than 10%, and a quarter moved 13.8% or more. Moderately traded items (50 to 499 units sold in 90 days) had the widest median swing, 9.9%. Items in our Low bucket (under 50 units) moved a median 3.4%, and the 16 Dormant items didn't move at all, because nothing sold. That stillness isn't precision. A RAP that can't move is just old.
So ten percent is about one month of ordinary drift for a typically traded limited. Inside the band, the two sides of a deal could swap places within weeks without anyone doing anything. Our own two figures disagree, too: the value estimate and raw RAP sit more than 10% apart on 18 of the 172 items, which is why the checker shows both columns.
Roblox runs a coarser version
Roblox's own Trade Quality Filter works on the same instinct from the other end. Its help page on trading scams (read October 17, 2026) says the filter "automatically checks incoming trade offers and flags ones that look unfair or one-sided" and hides them from your inbox. You set it to Low, Medium, High or None under Account Settings > Privacy & content restrictions > Trading & inventory. Roblox doesn't publish where those thresholds sit. Like the even band, it is built to catch lopsided offers rather than referee close ones, and the same page says plainly that "no filter can catch every bad deal."
The general rule
There's a version of this essay for every tool we build: the tracker labels its estimates as estimates with dates attached, and the checker rounds its verdicts to the width the data supports. The rule underneath is the same one: a tool should know its own error bars, and it should never let its interface imply less uncertainty than its inputs contain.
Sites in this niche compete on confidence. Confidence is cheap to render - it's a font choice. We'd rather compete on the other thing, the one you can only notice after a few months of use: whether the tool's strong claims kept turning out to be true. Refusing to split hairs finer than the data supports is how a strong claim stays strong. The even band isn't the checker being timid. It's the checker telling you the truth about itself.
Corrected September 11, 2026: what changed.