Open methodology

Every number should explain itself.

The arena is competitive. The calculation, verification labels, privacy boundary, and publication rules stay transparent.

01 · THE INDEX

A reproducible score, not a mystery rank

The AI Maxxing Index adds stack depth to a capped GitHub output term. Every intermediate number is visible on the profile and in its public JSON.

Index = Σ √(min(tool sessions, 1,000)) + min(20, 2 × √(merged PRs + contributions ÷ 25))

02 · QUALIFICATION

Real use earns stack depth

A tool contributes to stack depth after at least 20 sessions or $5 of reported or estimated usage value. Qualifying tool depth is the square root of its sessions, rewarding breadth without letting repetition grow linearly. Sessions past 1,000 per tool still count toward displayed spend but not toward the score, so no single tool can dominate the stack.

Usage is never double-counted: when the same tool, model, and day exists as both a self-reported row and a verified reporter row, only the verified copy counts.

Usage value is displayed as part of the build record but never added directly to the Index. A larger budget cannot buy a larger score by itself.

03 · OUTPUT

GitHub proves output, not AI usage

Merged pull requests and contribution-calendar activity form the output term. Every 25 contributions count as one output unit, and the entire output bonus is capped at 20 points.

GitHub cannot reveal local Claude Code, Codex, or OpenCode consumption. Those tool, model, token, and usage-value aggregates come from a manual report or the privacy-preserving local reporter.

04 · VERIFICATION

Verified and Self-reported stay visibly different

Self-reported rows are entered by the account owner and remain labeled as such. Verified rows are signed by a linked local reporter and accepted only after protocol validation. Mixed profiles disclose that both sources are present.

For verified Codex and Claude rows, dollars are an API-equivalent estimate calculated from mutually exclusive uncached input, output, cache-read, and cache-write buckets using the versioned model rate card. The server recalculates the estimate before storing it. OpenCode imports the cost recorded in its own session database. An API estimate is not a subscription charge or a claim about the developer's bill.

05 · PRIVACY

Only aggregates leave the machine

The reporter sends daily counts for tools, models, sessions, tokens, and estimated cost. It never sends prompts, responses, reasoning, code, commands, file paths, repository names, attachments, or raw local records.

Connecting data and publishing a profile are separate choices. Signing in, importing candidates, or syncing usage never makes an account public.

06 · NEUTRALITY

Sponsored credits are excluded from competitive headline totals. Sponsorship never changes verification, qualification, rank, or the Index formula.