Check-in [d5926b325c]

A structural-horizon clone lacks C's ancestry, so tag propagation has nowhere to start: every check-in in K would derive no branch, and closed markers and names set below C would vanish. The seed of draft-fossil-horizon-00 Section 5.2 restates the dense tag state at each entry of K so the unmodified derivation comes out exact.

internal/horizon/seed mints it at an origin (T cards per row in force at C and each check-in of K whose primary parent is outside K; D the latest summarized event rounded up to a whole second; U the origin's login), and holds it at a clone in a table of its own, outside the artifact store, so nothing that enumerates or serves artifacts can send it. It is applied when held, by Rebuild, and by seed.Apply, which the sync layer calls after each round. Replacing or discarding a seed re-derives the store, since only a re-derivation separates the old seed's rows from those real artifacts wrote. A Verifier hook carries the Section 5.2.4 provenance policy; no signing is implemented.

The crosslink-table clearing moves from verify into manifest.Rederive so Rebuild and the seed share one owner of re-derivation. The cards match zit's seed byte for byte in T and U, and each implementation derives from the other's seed what it derives from its own. check-in: 45fb2c7ed4 user: [email protected] tags: feat/horizon-cones, trunk

The Appendix A step 4 measurement: an opt-in corpus test reporting what a structural horizon saves over a content horizon at several depths. check-in: 5430186930 user: [email protected] tags: feat/horizon-cones, trunk

Opt-in corpus test (LIBFOSSIL_CORPUS) for draft-fossil-horizon-00 Appendix A step 4: for C at 10, 100 and 1000 check-ins back along trunk's primary-parent chain and at the check-in nearest a year before the tip, compare a content horizon's transfer with a structural horizon's under both held-set rules. It works on a copy of the repository. The seed is estimated as one control artifact with one T card per tag in force at each entry of K, uncompressed.

fossil-scm.org clone: 67894 artifacts, 37.2 MiB stored, 8425.2 MiB expanded (21008 manifests, 8.9 MiB stored, 1099.8 MiB expanded; no sub-manifest nodes; 6263 other structure, 691 clusters). Runtime 58s, peak heap 161 MiB. Share = (content - structural) / content.

descendants rule K below (n / MiB exp) content MiB exp structural MiB exp share exp share stored 10 back 12 20996 / 1098.9 1136.5 37.6 96.7% 49.0% 100 back 121 20887 / 1089.9 1262.0 172.1 86.4% 48.5% 1000 back 1746 19262 / 962.2 2341.0 1379.0 41.1% 40.0% 1 year (426 back) 796 20212 / 1036.1 1846.3 810.3 56.1% 44.5%

leaves rule: K = 20907 of 21008 at every depth; share 0.0% expanded, -7.7% stored (the seed outweighs the 101 manifests below). check-in: c6ef4af29f user: [email protected] tags: feat/horizon-cones, trunk