Abstract
A method is disclosed for storing the committed prior of a content-drift gate (the recorded hash of every artifact in a shared, version-controlled corpus) as one small file, a shard, per artifact under per-class subdirectories of a lock directory, whose listings are themselves the index, instead of one regenerated aggregate lock file. Each key maps to one filename: short safe-alphabet keys lacking the reserved separator map to themselves, and other keys map to a truncated readable stem plus a digest taken over the whole key before truncation, so keys that agree up to the cut stay distinct; a planner refuses the run if two distinct keys claim one path. The writer builds the complete intended shard set (a plan) and applies its difference against disk: a shard at a path the plan does not assign is an orphan and is removed, and a planned shard absent from disk is missing and is restored; a check mode reports the same difference without writing. Self-stamped provenance is omitted, derived totals are recomputed on read, fields are serialized in a declared order, and the lock directory is excluded from the catalogue it protects, because a store holding its own hash has no fixed point. Authors onboarding different artifacts then overlap only on disjoint paths, the one overlap a server-side three-way tree merge is guaranteed to resolve. A branch-scoped check asserts only the shards the branch touched, directly or through the artifact path each shard records, and fails closed on an unreadable base.
Creative Commons License

This work is licensed under a Creative Commons Attribution 4.0 License.
Recommended Citation
Assuncao, gustavo matthew, "Conflict-Free Sharded Committed-Prior Hash Lock with Injective Slug-to-Shard Naming and Plan-Then-Diff Orphan Reclamation", Technical Disclosure Commons, (November 20, 2026)
https://www.tdcommons.org/dpubs_series/13194