HOW WE KNOW WHAT WE SAY WE KNOW

STRONG CLAIMS.
NARROW SCOPE.

The methodology exists so an impressive number cannot quietly change meaning when it reaches a headline. Different evidence classes stay different. Different maturity states stay earned.

Evidence classes

repository-derived: Git history, pull requests, files, contracts, and other facts directly attributable to a versioned repository.
runtime/test-derived: Results produced by running a specified runtime, test suite, acceptance path, build, sanitizer, or verification harness.
user-reported operating fact: An operating fact supplied by the founder, such as the current $136/month external cash burn. It is labeled rather than silently promoted into repository evidence.
user-authored thesis/statement: Purpose, interpretation, philosophy, and founder language. These can be important without pretending to be empirical measurements.
external comparator: Context from outside Inversion Labs. It does not become a productivity multiplier unless definitions and accounting boundaries are harmonized.

Three different views of output

Gross branch graph: All reachable refs in the measured 30-day graph. Useful for actual activity; includes parallel and superseded work.
Conservative converged: Mainline activity counters plus substantially converged main + terminal-candidate churn. A tighter convergence reference, not a hidden replacement.
PR-local overlapping: The 90-day explorer preserves the work visible inside every captured pull request. Shared ancestry, stacked branches, supersession, and convergence mean the totals overlap and are not unique-code output.

True 1.0 is a release doctrine, not a version string.

01
EXPERIMENT
A bounded idea or prototype with questions still larger than its proof.
02
TOOL
A real usable artifact exists, but verification or release obligations remain incomplete.
03
VERIFIED
Important behavior has reproducible evidence and stated limitations.
04
RELEASE CANDIDATE
The product surface is substantially converged while explicit release gates remain.
05
TRUE 1.0
Implementation, verification, dogfooding, documentation, release obligations, public truth claims, and the Inversion Labs Promise all survive their gates.

Promise gate: Any project that violates the Inversion Labs Promise cannot graduate to TRUE 1.0, regardless of technical performance, popularity, revenue, or version number.

Passing tests do not equal release authorization.

Tests answer whether specified behavior passed specified checks.

Release authorization asks whether the wider security, provenance, operational, and product obligations have enough evidence to ship.

CAPT's current public case deliberately shows both states at once: green ordinary suites and a blocked release-security gate.

Operating-cost boundaries stay visible.

The $136/month figure is a user-reported operating fact describing current incremental external cash spend. It excludes founder labor, already-owned hardware, baseline household infrastructure, and future scaling. The $48,000/month anecdotal comparison is an external cash-denominator context only, not evidence of a proportional productivity advantage.