Leaderboard and Comparison

What the Leaderboard Ranks

The Leaderboard has two tabs, Standard and Stress limit. Each ranks its own kind of run on its own scale, Standard scoring or Stress-limit scoring: a stress limit total never sits next to a standard one. Under the title, the number of reports ranked and What is ranked? sum up the rules below.

To put one of your reports on the leaderboard, make it public: see Reading a Report. A report's rank badge (SS to F) compares its score with recent public reports; the rules are in How Scoring Works.

When a newly published report enters the top 4 above one of yours, you get a notification, and an email unless you turned off Leaderboard overtake (see Account and Security).

Reading a Row

Two columns are always shown, first: #, the position, and Configuration: the CPU, its cores, the RAM, the server software and Minecraft version, the name of the server when its owner verified it and shows it on their profile, and the Verified and Certified badges. When a report is attributed to a hosting provider, the provider's logo replaces the CPU icon, and a certified report also names the provider. At the end of the row, one button adds the report to the comparator and another opens it.

The other columns can be shown, hidden and reordered. By default the table shows User, Total score, Single-core, Hardware, Gameplay, Date, the scoring version (headed Standard scoring or Stress-limit scoring) and Rank. User reads Anonymous when the report belongs to no account.

On a narrow screen, each report becomes a card with the same information.

Choosing and Ordering Columns

Click Columns above the table, or right-click its header row. The menu groups the columns into Scores, Machine, Server and Report; tick the ones you want. Arrange columns… opens a list where you drag each column, or move it with its arrows, to set the order from left to right. Reset layout brings back the default columns and order.

Every one of these values is already shown on the public page of the report. Your layout and your page size (Per page: 25, 50 or 100) are saved in this browser only: a link you copy carries the ranking you see, not your layout.

Sorting

Click a column header to sort by it: the first click sorts in descending order, the next one reverses it. On a narrow screen, use Sort by and the button next to it. The leaderboard opens sorted by Total score, highest first.

All the scores, Date, the scoring and plugin versions, Cores / threads, Usable CPUs, Frequency, RAM and Plugins can be sorted. Versions sort in version order (1.10 after 1.9), and reports that lack the value always come last, in either direction. User, OS, Java, Minecraft, Server type, Host, Country and Rank cannot be sorted.

Filters

Filters add up: a report must match all of them. Each active filter appears as a chip under Active filters; click it to remove it. Reset filters clears every filter but keeps the tab, the sort and the page size.

The page address holds the tab, the filters, the sort and the page: copy it to share exactly the view you see.

Scoring and Plugin Versions

The scoring rules have versions, and the two scales are numbered separately: version numbers of the standard scale mean nothing on the stress limit scale. A report's scoring version appears in its row and on its page, as Standard scoring or Stress-limit scoring followed by the number.

• Under Versions, From and To limit the view to a range of scoring versions of the current tab, and a range of plugin versions. The menus list only the versions present in that tab. Bounds given in the wrong order are put back in order.
• Switching between Standard and Stress limit clears the scoring version filters, since they belong to the other scale.
• When the view mixes several scoring versions, or reports with no recorded version, a warning says so: scores computed with different rules are not directly comparable. The button next to it shows only the latest version.
• A scoring version that belongs to the other scale, from an old link for instance, is ignored, and the page tells you.

Comparing Reports

Add to Compare, on a leaderboard row or on a report page, puts the report in the Comparator, a panel at the bottom of the screen. It holds up to 5 reports, kept in this browser; Compare opens them side by side once there are at least two. Compare with others, in a report's Share menu, opens the same page with that report.

The Compare Reports page works like this:

• Add Reports takes a report ID (vxb_…) or a CPU name, which searches the leaderboard. Up to 5 reports.
• You can add any report you can open: public, unlisted or certified reports, and the private reports you may read, your own and those of a server you hold and have verified. A report that has expired or cannot be found is listed with the reason.
• The first report is the Reference: the others show their gap to it in percent. Click another card to change the reference.
• Once there are two reports, Highlights colours the best and worst value of each row, and AVG, MIN, MAX and MED add computed columns.
• Score Comparison, Performance Profile (each report's strengths, normalised to 100 %) and Score Decomposition chart the scores; the Scores, Hardware, Tests and Server tabs detail them.
• Copy link copies the address of the comparison.

Each report is one run. Before drawing conclusions from a small gap, look at how stable each run was on its report page.

Comparing Your Own Servers

Dashboard → Compare, the Multi-server comparator, puts your linked servers side by side on up to 12 metrics: score components and unit test measurements. Top performer counts the metrics each server wins, the Comparison matrix colours each row from best to worst, and a click on a row shows its trend over the chosen period. It needs a Pro or Enterprise plan or a hosting provider account (see Plans & Limits).

Before and After a Change

To know whether a change (a Java flag, a Paper setting, a new host) helped, the API and the Claude connector compare runs taken before and after it. This is not on a page of the site: call GET /api/v1/reports/before-after with a token that has reports:read (see API and Tokens), or ask Claude through the connector (see Connect Claude to Your Account).

• Give 1 to 10 reports on each side, and name the tests the change targeted, for example mobAI or hopper. The other tick tests are watched for collateral damage, and the hardware tests serve as a control.
• Every report must be one you can open. A verdict needs both sides to come from the same server, the same VoxelBench version and the same benchmark mode; declare the other differences you intended: platform (server software), Minecraft version, Java version or hardware.
• The verdict is improved, degraded, no_detectable_effect, undecidable or not_measured, and collateral_regression flags another test that got worse.
• It often refuses to conclude, by design: two runs of the same server never give exactly the same figures. Noise is estimated from the runs themselves, which needs at least three runs on each side.
• The platform does not know what else changed between the runs: the verdict assumes the targeted change was the only one.

The API also compares two reports directly (GET /api/v1/reports/compare): it says which one leads on the total score, whether the gap is within the noise, or that the two cannot be compared, for example a standard run and a stress limit run.