Adoption and compatibility evidence
August 9, 2026 · View on GitHub
This page separates verifiable framework evidence from adoption claims. It is part of the v1.0 release gate: every public statement must be traceable to a repository file, a dated CI run, or a permissioned adopter record.
Current status
As of the latest review, zero independent external adopters are publicly confirmed. The repository maintains three reference consumer fixtures—Atlas, Beacon, and Cinder—to exercise the compatibility boundary. They are useful engineering evidence, but they are not customer outcomes and are not counted as adoption.
The independent-adopter target is three permissioned outcomes. Recruitment is not started until a team opts in through the support channel. No name, metric, logo, download count, or case-study language is published without written permission from that team.
Evidence categories
| Category | What it proves | Source | Public claim allowed |
|---|---|---|---|
| Reference fixture | A maintained consumer shape remains compatible | compatibility/README.md and CI | “The fixture passed.” |
| Compatibility run | A specific revision/version passed a dated matrix | base-cli-compatibility-evidence-<run-id> artifact | “Version X passed run Y.” |
| External adopter | A real independent team completed an agreed outcome | Permissioned adopter record | Only the approved case-study facts |
Reference fixtures must remain separate packages with their own metadata and tests. Base, base-demo, and other adjacent repositories are explicitly excluded from the independent-adopter count.
Adopter program
- Recruit: offer a short evaluation against the adopter readiness checklist; record only a consented contact and the intended CLI shape.
- Onboard: pin a supported minor release, run the installed-wheel smoke test, and capture migration friction without receiving private credentials or production data.
- Verify: run the adopter's agreed compatibility and contract tests; keep the framework revision, package versions, platform, and dated result.
- Publish (optional): request written approval for an anonymized metric or named case study. A refusal or incomplete pilot remains a valid private outcome and is never presented as adoption.
- Review quarterly: re-confirm consent, compatibility status, and the exact wording of every public claim.
The support issue template is the intake path. Security reports and private
reproductions must follow SECURITY.md.
Reproducible compatibility records
Every successful Reference consumers workflow can emit a JSON record named
base-cli-compatibility-evidence-<run-id>. It contains the source revision,
framework version, fixture names, Python/Typer matrix, timestamp, and the
commands needed to repeat the check. Download a record from a run with:
gh run download <run-id> \
--repo basefoundry/base-cli \
--name base-cli-compatibility-evidence-<run-id> \
--dir compatibility-evidence
The record is an engineering result, not a customer claim. Before publishing an external outcome, maintainers must link its permissioned source record and make the approved scope clear.