Methodology & licence

Rankings and the app leaderboard measure adoption — the volume of activity a model, provider or app sees on the unoblox gateway — never quality, safety or a recommendation. Figures are computed from platform-observed usage on fixed 1-, 7- and 30-day UTC windows, refreshed on a schedule, not on every page view. These windows NEST (the 1-day window is fully contained in the 7-day window, which is fully contained in the 30-day window) rather than being independent, non-overlapping periods; the release boundary below is applied to the difference between a window and the next narrower one, not only to each window on its own, precisely because they nest.

• Wallet-charged, not free-tier. Only the ordinary paid path counts — never free-tier/BYOK/comped traffic, and never a workspace that has never completed a real cash top-up. A wallet balance can still mix real top-ups with unspent signup-bonus or referral-reward credit (the platform does not track which specific paisa funded which request), so a ranked figure is best read as "wallet-charged usage from workspaces that have topped up at least once," not a guarantee that every counted token was paid for in cash. Any future synthetic probe traffic is excluded structurally, since it is never billed.
• Contributor floor. A figure is published only once at least 100 requests and 5 distinct workspaces contribute to it in the window.
• Dominance cap. If a single workspace would make up more than half of EITHER the request count or the token count for a figure, that figure is withheld for that window rather than published — checked on both published numbers independently, not only the one a page happens to sort by.
• No platform or supplier self-traffic. Platform-owned workspaces, and a provider's own workspaces calling its own endpoints, are excluded before ranking. On the app leaderboard, a provider operator's workspaces are excluded regardless of which provider actually served the traffic.
• Coarse rounding. Published counts are rounded, never exact, and a published trend ("up/down since last window") is rounded to the nearest 5 percentage points, so a figure cannot be used to infer a single request's exact timing or size.
• Nested-window guard. A wider window (7- or 30-day) is only published if the PART of it not already covered by the next narrower published window also clears the contributor floor and dominance cap on its own — otherwise someone could recover that excluded middle range by subtracting one published figure from another. A residual risk we have not yet closed: comparing a model's figure against an app's figure for the same period can, in principle, narrow things down further; we mitigate this with the same coarse rounding above and are tracking a stronger fix.
• Minimum account age (app leaderboard only). A newly created workspace's traffic does not count towards an app's row until the workspace has existed for a minimum period.

No workspace, organization, API key, user, or individual request is ever identified in a public figure. Model and app rankings are platform-level aggregates only; the underlying per-tenant data is never exposed through this surface, with or without consent.

An app or agent appears on the app leaderboard only because it identified itself on its own requests, via the standard X-Title / HTTP-Referer request headers — the same convention used elsewhere in the industry. Not sending those headers keeps a caller's traffic off this board entirely; this is the caller's own act, not a consent flow unoblox operates on the app's behalf. Only the app's total request and token counts across all models are shown — never a breakdown by model, and never by the app's own users, workspaces or keys.

Draft wording below; not final until the owner approves it for the Terms of Service and this page.

1. Aggregate statistics published here (minimum-contributor thresholds and the dominance cap applied) are made available under a CC BY 4.0-equivalent licence: reuse and redistribution are permitted with attribution to unoblox.
2. Application rankings, computed from the optional X-Title / HTTP-Referer headers, are published under the same licence, conditioned on the calling application confirming it has the right to supply that self-identification.
3. What this licence does not cover: any workspace-, organization-, key-, or user-level data, and any content of a request or response — none of that is published here under any licence.
4. An app or a workspace owner may request removal of their data from this board; removal requests are honoured within 7 business days.
5. Being ranked is not an endorsement by unoblox, and unoblox retains discretion to remove a listing (for example, for suspected gaming of the ranking).

Before any performance or uptime figure is published under an upstream provider's name, unoblox checks that provider's own terms of service for restrictions on publishing measurements of their service — an outstanding item before that specific surface (a separate, consent-free track measuring platform-run synthetic probes) goes live, and independent of the volume rankings and app leaderboard this page otherwise describes.