Visitar URL original
F14c on the dense lane: declare six arms, disclose DuckDB VSS and ArangoDB (#68) by tae898 · Pull Request #69 · humemai/arcadedb-embedded-python · GitHub
Skip to content

F14c on the dense lane: declare six arms, disclose DuckDB VSS and ArangoDB (#68) - #69

Merged
tae898 merged 1 commit into
mainfrom
fix/f14c-dense-declarations
Oct 2, 2026
Merged

tae898 merged 1 commit into
mainfrom
fix/f14c-dense-declarations

Conversation

@tae898

@tae898 tae898 commented Oct 2, 2026

Copy link
Copy Markdown

Closes #68.

  • fairness_check.py: PHASE_SPLIT_DECLARED (Chroma, Qdrant fp32 and int8, sqlite-vec fp32 and int8, SurrealDB embedded, each with its reason) counts as declared; PHASE_SPLIT_DISCLOSED (DuckDB VSS, ArangoDB) counts as accounted for and is named in the gate's output.
  • export_web.py: _phase_split_notes prints under the dense table which arms have no separate index phase and that the two disclosed arms' totals include their index build, reading the same maps, so the page and the gate cannot disagree.

Landing-side code only: no lane, runner, or adapter changes, so the running campaign's rows are unaffected.

Checked on October's 1,643 rows: F14c reports the dense lane ok (11 split, 6 declared, 2 disclosed) and 0 failures across every lane; the two sentences generate from the dense arms. The regenerated payload diff is read at the dense landing's dry run, as every landing does.

🤖 Generated with Claude Code

https://claude.ai/code/session_01WWcDr7TwvcfMsQkeYQVHZr

…hase, disclose DuckDB VSS and ArangoDB on the page

Rehearsed on October's rows before the dense landing: F14c failed the dense
lane (11 of 19 arms split ingest from index; 8 neither split nor declared).
Read per adapter: Chroma builds during add, Qdrant (both precisions) builds in
the background with the wait inside the timer, sqlite-vec (both) has no
separate index, SurrealDB embedded defines its index before the load; DuckDB
VSS and ArangoDB build after the load inside one timer. fairness_check gains
PHASE_SPLIT_DECLARED and PHASE_SPLIT_DISCLOSED; export_web prints, under the
dense table, which arms have no separate index phase and that the two
disclosed arms' totals include the index build, reading the same maps so the
page and the gate cannot disagree. Landing-side only: no lane or runner code
changes, so no row changes.

Checked: on October's 1,643 rows F14c now reports the dense lane ok (11 split,
6 declared, 2 disclosed) and 0 failures over every lane; the two sentences
generate from the dense arms (BENCH_SKELETON=1, as the laptop has no overlay).

Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01WWcDr7TwvcfMsQkeYQVHZr
@tae898
tae898 merged commit f0292bc into main Oct 2, 2026
2 checks passed
@tae898
tae898 deleted the fix/f14c-dense-declarations branch October 2, 2026 03:17
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

F14c refuses the dense landing: eight dense arms neither split ingest from index nor declare why

1 participant