Full test and ship v0.4.0 #34
Winnie
Loading…
Reference in a new issue
No description provided.
Delete branch "%!s()"
Deleting a branch is permanent. Although the deleted branch may continue to exist for a short time before it actually gets removed, it CANNOT be undone in most cases. Continue?
Parent: #28. Once the mains are competitive and the probe suite passes, run the full test suite and cut the v0.4.0 release (macOS ARM64 tarball).
Keeping this open as the ship gate: full test + v0.4.0 release happen here once #38 and #39 land. Everything else from the old breakdown is superseded by #35-#40.
Gate issue. Activates when #38, #39, #40 land. Assigned subagent is pre-running the full test suite (vet, unit tests, interactive spectate smoke, darwin/arm64 build) against current master so nothing is invented at ship time. No v0.4.0 until the three merge.
Subagent started: pre-running the full test suite against current master (vet, tests, interactive spectate smoke, darwin/arm64 build) so the ship gate has a dry run before #38/#39/#40 land. No release until the gate passes.
Ship-gate dry run (v0.4.0) against master
da67404— worktree34-ship-gateThis is the dry run only — no release cut, gate still waits on #39.
go vet ./...go test ./...ok machinetele.com/bankwars 0.003s)bankwars-darwin-arm64built; first 4 bytescf fa ed fe(Mach-O 64 LE), cputype0x100000c(ARM64)[ai)ACCT DIVERGENCElines; 0CF DIVERGENCESmoke config: seed 33, players 6, max_turns 80, survivors 3, dice_faces 30, recovery_rate 0, bond_face 1000, bond_min 0.05, bond_term 10 (full gate config). Game completed cleanly (exit 0, final ranking printed).
The two accounting-divergence lines (both Bank 4):
The turn-10 divergence immediately follows
Bank 4 acquires Bank 3's position for 13476 (second price)— the acquisition is right at the boundary where equity change (19196) exceeds recognized net income (18867) by 328. Turn 13's diff is 3386. Reproduced exactly on a second run (identical lines, identical diffs) — deterministic under seed 33.Gate verdict: not green — accounting identity breaks in a spectate game with this config. Awaiting #39 before ship.
Ship gate now includes two new prerequisites Robert requested on #39: #46 structured iteration log and #47 the researcher-grade PDF report, emailed before the release. Both are release blockers alongside #39's mains work.
Gate update: the 2 ACCT DIVERGENCE lines from the dry run are fixed. That run was on master before PR #50 merged. Re-ran the identical smoke (seed 33, default config) on current master (
7f503c4, includes #50): 0 divergence lines, exit 0. The gate check is now green on that line; the remaining blocker is the four per-AI brains (#52-#55) and the final merged bench + PDF.Acknowledged — I own this gate now.
Plan, in order:
7f503c4+ merged #56 mill + merged #57 adapt):go vet,go test ./..., darwin/arm64 build + Mach-O check, spectate smoke (0[aierrors, 0 divergences). This proves the release path while #52/#54 finish.First dry-run comment lands with the check table.
Gate dry run #1 — master
3fe3b3f(includes #56 mill + #57 adapt)go vet ./...go test ./...0xfeedfacf(Mach-O 64 LE), cputype0x0100000c(ARM64)[aierrors, 0 ACCT DIVERGENCE, 0 CF DIVERGENCE; game ended turn 23, winner an adapt seatgo buildThe release path is green on everything merged so far. Remaining before the real gate: #52 (empire) and #54 (vault) landing, then the final 100-game merged bench + the #47 PDF report, then tag v0.4.0 + macOS ARM64 tarball (own subdir) + email delivery.
All four brains landed — final gate starting on master
771a6dc248Final gate sequence now running: (1) vet + tests + darwin/arm64 build + spectate smoke on the merged master, (2) full 100-game x 13-preset merged bench, (3) #47 PDF report from the real iterations.json + bench results, (4) PDF emailed to Robert, (5) v0.4.0 tag + macOS ARM64 tarball (own subdir) delivered by email + drive. Results land here as each step completes.
@Winnie do some issues need to be closed?
Yes — done. Closed the four brain issues just now, each with a comment saying what shipped and which PR carried it:
Still open on purpose: #34 (this gate — final merged bench is running right now on master
771a6dc248), #39 (umbrella, closes when the gate passes and v0.4.0 ships), #28 (the parent — I'll close it with the release), and the backlogs #26 (v0.5.0 ideas) and #15 (central bank) which stay open per your roadmap.Gate passed and v0.4.0 shipped. Final checks on master
430f2ec: go vet exit 0, go test ok, darwin/arm64 build (Mach-O 0xfeedfacf / ARM64 0x0100000c verified), spectate smoke 0 [ai errors + 0 ACCT/CF divergences, and the 100-game deterministic merged bench (adapt 575 / mill 392 / empire 194 / vault 125). The #47 researcher PDF was generated from the real iterations.json and emailed before the release; the macOS ARM64 tarball went out by email + forge release asset + drive mirror. Closing.