53 lines
3.2 KiB
Markdown
53 lines
3.2 KiB
Markdown
# Onion MC — optimization effort: state capture
|
||
|
||
Captured **2026-10-14 ~23:45 EDT** (Manjaro bench machine), at git **HEAD `e5e369a`**
|
||
(`chore(bench): add client bench harness and Phase 2 handoff`), branch `main`, pushed.
|
||
Pack: **217 mods**, MC **1.21.1**, NeoForge **21.1.251**.
|
||
|
||
> **Update 2026-10-15 (b):** WATUT crash FIXED (`a84319b`, MidnightLib 1.10.0 — root
|
||
> cause was a MidnightLib CME); uncapped frametimes VERIFIED; harness v2 (`bench.py`)
|
||
> replaces `run_scene.py` (`e18bd3c`). See `.pi/journal.md` latest entry. Plan steps
|
||
> 1–2 of `04-plan.md` are done; next is S4 build → B0.
|
||
|
||
## Where we are (one paragraph)
|
||
|
||
The client-performance initiative is in **Phase 2 (benchmarking)**, which has just
|
||
begun and was interrupted mid-setup. Static analysis (Phase 1), the tuned-config
|
||
generator, the owner's decisions, and the first wave of perf-mod adds are **done and
|
||
shipped** (`a01bc3b`, `d80f85b`). The client **boots clean**; the dedicated server
|
||
**boots clean** (a live idle instance from the boot check is still running in tmux —
|
||
leave it alone). The bench harness works end-to-end but has produced exactly **one
|
||
non-decision-grade measurement** (C1/S1, vsync-locked ~60 fps). The bench world has
|
||
scenes S1–S3 built; **S4 (airships) is not built**. **Matrix blocks B0–B14 have not
|
||
run at all.** The one known open defect is an **intermittent WATUT world-load crash**.
|
||
Immediate next actions: resolve/work around WATUT → confirm uncapped frametimes →
|
||
finish S4 → run B0→B14 → write winners into `build.py` → ship → report to owner.
|
||
|
||
## Reading order
|
||
|
||
| File | Contents |
|
||
| --- | --- |
|
||
| [`01-pack-and-environment.md`](01-pack-and-environment.md) | What the pack is, repo conventions, this machine's inventory, tool paths |
|
||
| [`02-timeline.md`](02-timeline.md) | How the effort got here: phases, commits, decisions |
|
||
| [`03-current-state.md`](03-current-state.md) | Exact current step, done/open/blocked, interrupted-session artifacts and evidence |
|
||
| [`04-plan.md`](04-plan.md) | The remaining plan with per-step status, B0–B14 run sheet summary, ship procedure |
|
||
| [`05-decisions-and-rules.md`](05-decisions-and-rules.md) | Settled owner decisions, rules of the road, traps and gotchas |
|
||
| [`06-artifact-map.md`](06-artifact-map.md) | Every artifact, where it lives, what it is, current status |
|
||
|
||
## Authoritative sources this capture distills
|
||
|
||
- `.pi/journal.md` — the running work log (read the last three sessions first).
|
||
- `analysis/client-perf/HANDOFF.md` — prior session's own handoff brief.
|
||
- `analysis/client-perf/BENCH-MATRIX.md` — **the run sheet**; when this capture and
|
||
the matrix disagree, the matrix wins.
|
||
- `analysis/client-perf/LEVERS.md` — ranked levers with verification notes.
|
||
- `phase0/P0|P1|P2-*.md` — per-fleet-machine measurement plans (machines unreachable).
|
||
|
||
## For the incoming agent
|
||
|
||
Start at `03-current-state.md`, then `04-plan.md`. Do the next actions in
|
||
`04-plan.md` §"Next actions" **in order**. Before touching anything, read
|
||
`05-decisions-and-rules.md` — several rules exist because a previous session already
|
||
made the mistake once (config re-save commits, packwiz `side` resets, vsync-locked
|
||
data). Append your session to `.pi/journal.md` when you finish.
|