Visitar URL original
Stage generator: exact pass order at N=5 (T1 1-3, T2 1-3, then T1/T2 alternating) and the ArcadeDB go marker by tae898 · Pull Request #503 · humemai/arcadedb-embedded-python · GitHub
Skip to content

Stage generator: exact pass order at N=5 (T1 1-3, T2 1-3, then T1/T2 alternating) and the ArcadeDB go marker - #503

Merged
tae898 merged 1 commit into
mainfrom
subset-schedule
Oct 24, 2026
Merged

tae898 merged 1 commit into
mainfrom
subset-schedule

Conversation

@tae898

@tae898 tae898 commented Oct 24, 2026

Copy link
Copy Markdown

What

make_2611_stages.py runs the engine tiers in the user's exact pass order (2026-10-24) and N defaults to 5: n=1 T1, n=2 T1, n=3 T1, n=1 T2, n=2 T2, n=3 T2, n=4 T1, n=4 T2, n=5 T1, n=5 T2 (schedule_order). Passes 1-3 are Tier 1 then Tier 2, then passes 4 and 5 alternate T1, T2. --reps 3 stops after the first six groups. One tier alone is plain rep-major.

--check prints the schedule and proves it: the plan collapsed into (tier, pass) groups equals schedule_order, no tier skips, repeats or splits a pass, and each group holds every cell of its pass once, so every prefix of the schedule is a complete set of equal replicates for the tiers covered so far. --check --drafts DIR lints written scripts for the same order.

--no-tiers and --tiers --reps 3 write byte-identical scripts to origin/main (diffed). Past 99 stages the ids take three digits: N=5 with tiers is 110 stages and qS100 sorts before qS11, which the chain's id order forbids.

ArcadeDB gate

DBBench keeps one ArcadeDB version per host, so the four ArcadeDB arms of Tier 1 must not start before the benchmark pin is 26.11.1; the comparators do not depend on the version. The ArcadeDB arms were already their own chain (--arcadedb-release, Tier 1 only, passes 1..N in order), generated with --after the last comparator stage, so the comparators run the schedule first. New: <prefix>00 waits after its --after for the marker file $HOME/conf/go-arcadedb-2611 on the host, before the ALL-DONE trap (a killed waiter hands nothing on; it never creates the file), and every ArcadeDB lane stage aborts without it. The lint demands the gate in the pair install and in every ArcadeDB stage and in no comparator stage. The user or the agent touches the file when CURRENT_PIN moves to 26.11.1.

Tests

test_stage_generator_2611.py (51 passed with test_stage_generator_repmajor.py): the exact order for N=5, 3, 4, one tier and three tiers; the prefix proof; the check catches a swapped boundary, the old all-of-Tier-1-first order, a flipped pass 4, a skipped pass, a lost cell, a mixed stage and a repeated id; the gate in the pair install and the lane stages and its lint; the ArcadeDB chain runs passes 1..5 in order. stage_drafts_2611.sha256 regenerated (default reps is now 5). CAMPAIGN.md describes the order and the gate.

🤖 Generated with Claude Code

… go marker

make_2611_stages.py: N defaults to 5 and --tiers follows schedule_order: T1 pass 1-3, T2 pass 1-3, then T1 pass 4, T2 pass 4, T1 pass 5, T2 pass 5 (2026-10-24). --check proves the order and that every prefix is a complete set of equal replicates (the plan collapsed into (tier, pass) groups equals the schedule, each group holds the whole pass); --check --drafts lints written scripts for the same. Stage ids take three digits past 99 stages (qS100 sorted before qS11). --no-tiers and --reps 3 write byte-identical scripts to before.

ArcadeDB gate (DBBench: one ArcadeDB version per host): the ArcadeDB chain's pair install waits after its --after for $HOME/conf/go-arcadedb-2611, before the ALL-DONE trap, and every ArcadeDB lane stage aborts without it; the lint demands the gate exactly where an ArcadeDB arm runs.

Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
@tae898
tae898 merged commit ba31b0f into main Oct 24, 2026
4 checks passed
@tae898
tae898 deleted the subset-schedule branch October 24, 2026 14:25
tae898 added a commit that referenced this pull request Oct 24, 2026
…ule, #487 pins)

Conflicts in COMPARATORS.md, QUANTIZATION.md, build_images.sh and runner.py are
resolved to main's text everywhere except the pins this PR carries: the SurrealDB
server stays at v3.3.4 (digest 0d25256b...) and the client image keeps
python-arango==8.3.6 (main's build_images.sh still read 8.3.5 beside its own
8.3.6 in COMPARATORS.md and the smoke script).

stage_drafts_2611.sha256 is regenerated with
`python3 test_stage_generator_2611.py --rewrite`, as #503 did, so it reflects
the new schedule and these pins.

Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
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.

1 participant