10-source-archive-release-boundary.md

# Source archives define the release validation boundary

Status: accepted. Date: 2026-10-24.

Validate the four Heni source distributions outside the checkout, with released
non-Heni dependencies. A core-only project must work independently; inspect the
reachable component graph rather than every node in Cabal's elaborated plan.
Aeson brings transitive QuickCheck, while Tasty/Hspec and Heni reporters remain
optional. Keep framework test APIs outside the core.

Run core tests with a restricted tool path, then separately require positive Git
patch integration and optional framework reporters. Archive tests also prove
complete Haddock, schema agreement, the standalone library/CLI exit contract and
representative migrations. Preserve failed extractions and bounded validation
logs. Do not let absence of a capability silently satisfy its positive gate.

Advertise only successfully tested OS/compiler combinations and dependency sets.
GHC 9.12.4 is both ends of the initial admitted compiler range; widening base
bounds is a compatibility task. Keep -Werror in owned CI policy. Version public
Haskell APIs under PVP independently from wire schemas.

The maintainer selected BSD-3-Clause and supplied metadata by reference to their
other packages. Name lookups are neither reservations nor ownership checks.
Source/docs hashes and readiness evidence are repository-only records excluded
from source distributions to avoid self-reference. Candidate upload, rendered
Hackage review, account authorization and publication remain separate release
actions after local readiness; local artifacts never imply an upload occurred.

Haddock instance-origin links can target hidden definition modules. Release docs
selectively generate the two implementation definition pages with `--show`,
marking them `not-home` and explicitly outside the supported import API. Public
facades remain documentation homes and installed Cabal module exports do not
change. Check generated module and source links as well as coverage percentages.