Official Codebase of 3DCodeVerse
Python
Commits on main
render: refuse to publish a round the working tree is not at
fa5a1aa gaoypeng committed 2 months ago
runlock: collapse the nested with-statements the CI lint flagged (SIM117)
e7a5144 gaoypeng committed 2 months ago
budget: enforce max_usd on money that is billed, not on list price
9045d8a gaoypeng committed 2 months ago
health: count console-script harness processes, so the pool budget is not always 64
1513780 gaoypeng committed 2 months ago
bench: flag harness cells that waited on the provider instead of iterating
934f88e ziyao committed 2 months ago
bench: pin one plan across both A/B arms — the variance the A/A found is the planner's
baae4b5 gaoypeng committed 2 months ago
skills: one library, one delivery policy — and the routed set as a TOOL for backends without a loader
a15c3b9 gaoypeng committed 2 months ago
plan-loop: the wave's writeup, and the rule that decides when two A/B arms may share one plan
42bf95c gaoypeng committed 2 months ago
python floor 3.10 -> 3.12: delete the compatibility layer and the duplicate CI job
b51d4bf gaoypeng committed 2 months ago
docs: the A/B's final state as of writing, and the second cell that confirms api-agent reads none of them
3f46a73 gaoypeng committed 2 months ago
docs: the A/B result — 1 scored pair at +0.002, and the reason: api-agent read 0 of 5 listed bundles while making 52 read_file calls
fe6fcd7 gaoypeng committed 2 months ago
bench: --report-only rebuilt summary.md from the bare flags, relabelling a real A/B as an A/A in the one file that outlives the run
dbf25d5 gaoypeng committed 2 months ago
docs: SKILLS.md section 6 — the rig, the noise floor it has to clear, and why the first attempt measured nothing
27073b0 gaoypeng committed 2 months ago
bench: the worktree-import guard is tested by making the bad thing happen, not by grepping for its message
d969953 gaoypeng committed 2 months ago
docs: V-9 in the release notes — the A/B rig was comparing a tree with itself
befda84 gaoypeng committed 2 months ago
bench: an A/B launched from a worktree ran the MAIN tree's harness in both arms, and looked healthy doing it
de28e73 gaoypeng committed 2 months ago
bench: ab_gate_rates.py — the same A/B run's deterministic readouts, paired per prompt, because the judged score is not the thing a contact skill targets
528269a gaoypeng committed 2 months ago
docs: section 7 said the surfaced-vs-deep split holds on the strength of one manual run; the four-CLI negative control says it does not
e4a5d20 gaoypeng committed 2 months ago
docs: the read-rate targets count only sessions the control says were measurable
d26adab gaoypeng committed 2 months ago
docs: RELEASE_NOTES gains the skills wave — what landed, the eight defects verification found, and the read-probe correction
6e4eb1b gaoypeng committed 2 months ago
skills: the read probe was measuring git, not the agent — so it now carries a control that says when it is blind
f44f8af gaoypeng committed 2 months ago