docs(tested): RX_VK_WB_MAX_KB=1024 measured on an RX 5600 XT (no mesh shaders)

waryxxful commented

2026-11-14 19:15:03 +00:00

First-time contributor

Type: docs · Risk: low · Tested on: Windows 11, RX 5600 XT · Needs: none

Why

RX_VK_WB_MAX_KB is off by default until it shows a gain on more than one machine, and the README asks for results. On an RX 5600 XT, a 6 GB card without mesh shaders, it takes 2.5 ms off the GPU's frame: 54-56 fps against 46-48 in Kokiri Forest. Runs on other such cards are asked for in #548.

What changed

Five run files under docs/tested/ and one paragraph of the README, so the next reader of the switch finds a card where it gains. No code changes.

docs: docs/tested/2026-11-14-wb-max-waryxxful-run01.md to run05.md, one per run: the four interleaved runs, and a 56-minute play session with the switch on.
README: the paragraph on RX_VK_WB_MAX_KB names the card and what it gained, and asks for other cards without mesh shaders.

Measurements

Verdict: it helps on this card. The GPU's frame is 2.2-2.9 ms shorter in Kokiri Forest and at the Desert Colossus (Kokiri Forest 54.2 / 55.5 fps against 45.9 / 48.0), 0.5-2.3 ms shorter in Hyrule Field, where the replay's submit holds the frame and the fps does not rise; 16 MB written back a frame instead of 61. No device lost and nothing wrong in the screenshots.

Hardware: Windows 11 Pro 26200, RX 5600 XT 6 GB (RDNA1, no mesh shaders), Adrenalin 26.10.1, Ryzen 5 5600X, 24 GB
Scene: six areas as child Link; Output resolution 1920x1080, Render resolution 50% (960x540), Quality preset Custom, VRAM budget Auto, frame generation off, frame rate cap 120, vsync off; RX_VK_NO_PREDICT=1, RX_VK_NO_WARM=1, RX_VK_MIRROR_SPARSE=0
Method: four runs interleaved (unset, 1024, two runs of another renderer, unset, 1024). In each area 10 s to settle, then 25 s while Link walks by scripted input; fps is the median of ten samples from the control bridge, GPU and replay submit are the log's per-frame medians. Both runs of each side are shown, not their median.
How to reproduce: add RX_VK_WB_MAX_KB=1024 to longshot_env.txt next to the game, load a save in the area, walk the same stretch with and without the line, and read the FPS counter's Breakdown and the log's a frame: lines.

Baseline: one renderer, main @ 9c9d69ed plus the mirror retry of #288 (ea6dc5f5 cherry-picked alone), with the switch unset. Main against this PR was not the comparison because the PR adds no code: the two sides are that renderer with the switch unset and set. The retry is there because that is what this PC plays with; plain main was not measured with the switch.

Measured on main 9c9d69ed, and the PR is on main 7ff5256e: the numbers still hold as what they are, rows about the build they name. The PR adds no code, so they make no claim about a later main.

The walk ends in the same place in Kokiri Forest, Hyrule Field and at the Desert Colossus (the draws agree). In Kakariko, the Market and Hyrule Castle it ends in a different place run to run (604 to 2,046 draws in Kakariko), so those three are in the docs/tested/ files but not compared here.

	Before, `RX_VK_WB_MAX_KB` unset: main @ `9c9d69ed` + `ea6dc5f5` (run 1 / run 3)	After, `RX_VK_WB_MAX_KB=1024`: main @ `9c9d69ed` + `ea6dc5f5` (run 2 / run 4)
Written back a frame (whole run)	60.8 / 60.7 MB	16.4 / 16.2 MB
Kokiri Forest: fps	45.9 / 48.0	54.2 / 55.5
Kokiri Forest: GPU a frame	14.8 / 14.5 ms	12.0 / 12.1 ms
Kokiri Forest: draws	727 / 741	737 / 746
Desert Colossus: fps	39.0 / 42.2	43.8 / 47.0
Desert Colossus: GPU a frame	18.7 / 18.0 ms	15.8 / 15.8 ms
Desert Colossus: draws	744 / 695	680 / 708
Hyrule Field: fps	43.5 / 45.6	46.6 / 41.0
Hyrule Field: GPU a frame	16.3 / 15.6 ms	14.0 / 15.1 ms
Hyrule Field: draws	1,337 / 1,324	1,266 / 1,364

Hyrule Field's GPU frame is shorter but its fps is not higher in run 4: there the replay's submit (13.4 ms) holds the frame.

vk: a frame: mirrors 563 (1020 MB): 7.51 MB synced in 1108.8 copies, 59.07 MB written back        (unset)
vk: a frame: mirrors 620 (1060 MB): 7.64 MB synced in 1108.3 copies, 16.20 MB written back        (1024)
vk: write-backs left out (600 submits): 817, 3682.0 MB; 26 bindings followed, 37 ranges the CPU read, 0 resets; CPU reads or writes of one left out: 1686 so far
vk: the CPU wrote 0x117ec0e000 +0x1000, in a binding whose write-back was left out: written back from now on (RX_VK_WB_MAX_KB=0: every one)

One RX_VK_PASS_TIMING=2 block of the same renderer with the switch unset, for where the write-back sits in the GPU's frame:

vk:   21.53 ms a frame in all (recordings' command buffers + upload command buffers; ...)
vk:     m pass        5.777 ms  33.2%    123.4 spans    1113.1 items     0.00 MB     46.8 us a span
vk:     m dispatch    3.817 ms  22.0%    146.3 spans     146.3 items     0.00 MB     26.1 us a span
vk:     m wb          2.101 ms  12.1%      7.6 spans     163.1 items    40.40 MB    278.2 us a span

Flags and settings

None changed. Results for one existing opt-in switch:

Flag	What it is	Before this PR	After this PR	Effect	Helps / doesn't help / untested	Recommendation
`RX_VK_WB_MAX_KB=1024`	Leaves out the write-back of the parts of large GPU-written bindings the CPU does not touch	off	off (unchanged)	61 → 16 MB written back a frame, the GPU's frame about 2.5 ms shorter on this card	Helps: RX 5600 XT 6 GB, Windows (the compute fallback). Doesn't help fps: RTX 3060 Ti, RX 580 (the README's earlier runs). Untested: other cards without mesh shaders (RX 5500 XT, RX 5700, GTX 10 series, Intel), laptops, Linux on this card	Stay off until other cards are measured; this PR adds one card where it gains

Testing

Tested: the four runs above in the game on the RX 5600 XT (Windows 11, Adrenalin 26.10.1). With 1024: no device lost, the twelve screenshots (six areas, two runs) looked at with nothing wrong seen, 20 the CPU wrote ... in a binding whose write-back was left out lines a run and none for a read. Then a 56-minute play session by hand with 1024 on the same renderer (run 5): Goron City 66.8 fps at the median, Death Mountain Trail 48-51, Dodongo's Cavern 56.8 over 32 minutes, Kakariko 36.5; no device lost, readback waits of 0.12-0.30 ms at the median, 20 the CPU wrote ... lines in the first area and none for a read. python -I -B tests/test_tested_configurations.py: 24 of 25 pass; the one failure is the Windows line-ending one of #482. python -I -B tests/run_all.py (Windows, llvm-mingw, with the branch on main 7ff5256e): 533 passed, 4 failed, 34 skipped. The four fail the same way on main 7ff5256e on this PC, each run alone: test_cpm_discovery, test_oot_state_tap, test_tested_configurations (the line-ending one) and test_walk_key (ld.lld: error: undefined symbol: hos_oot_autoload_input).

Not tested: plain main without the mirror retry; the areas and dungeons not named here; Linux on this card; the tests again on a main later than 7ff5256e.

Must be tested: None (docs only). Runs of the switch on other cards without mesh shaders and on laptops are asked for in #548, with the steps for a release build.

Risks

What could break: nothing in the game; five documentation files and one README paragraph. #753 leaves that paragraph alone because this PR rewrites it.

Turning it off: revert the commit.

Compatibility: Nothing changes.

Issues

See #548
See #482

Checklist

Everyone

One topic.
The title is a Conventional Commit (type(scope): description) of the status line's type.
Why gives the reason plainly and links the issues that fit (the tracker searched).
Measurements has a verdict, or one line saying why there is nothing to measure.
Testing names what was tested, what wasn't, and what must still be tested, where and how.
Rebased on current main: on 7ff5256e, not on today's main; documentation only, and it merges into aa95d35f without conflicts.
python -I -B tests/run_all.py passes (any failure that also fails on main is named): 4 failures, the same on main 7ff5256e on this PC (Testing).
No game files, dumps, build output or generated files.
third_party/ untouched (a new folder with a PROVENANCE.md instead).
The license is OK: GPL-3.0-or-later, nothing incompatible and nothing decompiled.

When it applies

AI assistance recorded with Assisted-by: / Co-Authored-By: trailers.
Numbers from the same PC, scene and settings, at least 2 interleaved runs each, the median reported: both runs of each side are shown.
A file in docs/tested/ for each new measured run (format in docs/tested-configurations.md).
Docs updated.

🤖 Generated with Claude Code

**Type:** docs · **Risk:** low · **Tested on:** Windows 11, RX 5600 XT · **Needs:** none ## Why `RX_VK_WB_MAX_KB` is off by default until it shows a gain on more than one machine, and the README asks for results. On an RX 5600 XT, a 6 GB card without mesh shaders, it takes 2.5 ms off the GPU's frame: 54-56 fps against 46-48 in Kokiri Forest. Runs on other such cards are asked for in https://git.magicbean.dev/MagicBeanDev/Longshot/issues/548. ## What changed Five run files under `docs/tested/` and one paragraph of the README, so the next reader of the switch finds a card where it gains. No code changes. - docs: `docs/tested/2026-11-14-wb-max-waryxxful-run01.md` to `run05.md`, one per run: the four interleaved runs, and a 56-minute play session with the switch on. - README: the paragraph on `RX_VK_WB_MAX_KB` names the card and what it gained, and asks for other cards without mesh shaders. ## Measurements **Verdict:** it helps on this card. The GPU's frame is 2.2-2.9 ms shorter in Kokiri Forest and at the Desert Colossus (Kokiri Forest 54.2 / 55.5 fps against 45.9 / 48.0), 0.5-2.3 ms shorter in Hyrule Field, where the replay's submit holds the frame and the fps does not rise; 16 MB written back a frame instead of 61. No device lost and nothing wrong in the screenshots. Hardware: Windows 11 Pro 26200, RX 5600 XT 6 GB (RDNA1, no mesh shaders), Adrenalin 26.10.1, Ryzen 5 5600X, 24 GB Scene: six areas as child Link; Output resolution 1920x1080, Render resolution 50% (960x540), Quality preset Custom, VRAM budget Auto, frame generation off, frame rate cap 120, vsync off; `RX_VK_NO_PREDICT=1`, `RX_VK_NO_WARM=1`, `RX_VK_MIRROR_SPARSE=0` Method: four runs interleaved (unset, 1024, two runs of another renderer, unset, 1024). In each area 10 s to settle, then 25 s while Link walks by scripted input; fps is the median of ten samples from the control bridge, GPU and replay submit are the log's per-frame medians. Both runs of each side are shown, not their median. How to reproduce: add `RX_VK_WB_MAX_KB=1024` to `longshot_env.txt` next to the game, load a save in the area, walk the same stretch with and without the line, and read the FPS counter's Breakdown and the log's `a frame:` lines. Baseline: one renderer, main @ 9c9d69ed plus the mirror retry of https://git.magicbean.dev/MagicBeanDev/Longshot/pulls/288 (ea6dc5f5 cherry-picked alone), with the switch unset. Main against this PR was not the comparison because the PR adds no code: the two sides are that renderer with the switch unset and set. The retry is there because that is what this PC plays with; plain main was not measured with the switch. Measured on main 9c9d69ed, and the PR is on main 7ff5256e: the numbers still hold as what they are, rows about the build they name. The PR adds no code, so they make no claim about a later main. The walk ends in the same place in Kokiri Forest, Hyrule Field and at the Desert Colossus (the draws agree). In Kakariko, the Market and Hyrule Castle it ends in a different place run to run (604 to 2,046 draws in Kakariko), so those three are in the `docs/tested/` files but not compared here. | | Before, `RX_VK_WB_MAX_KB` unset: main @ 9c9d69ed + ea6dc5f5 (run 1 / run 3) | After, `RX_VK_WB_MAX_KB=1024`: main @ 9c9d69ed + ea6dc5f5 (run 2 / run 4) | |---|---|---| | Written back a frame (whole run) | 60.8 / 60.7 MB | 16.4 / 16.2 MB | | Kokiri Forest: fps | 45.9 / 48.0 | 54.2 / 55.5 | | Kokiri Forest: GPU a frame | 14.8 / 14.5 ms | 12.0 / 12.1 ms | | Kokiri Forest: draws | 727 / 741 | 737 / 746 | | Desert Colossus: fps | 39.0 / 42.2 | 43.8 / 47.0 | | Desert Colossus: GPU a frame | 18.7 / 18.0 ms | 15.8 / 15.8 ms | | Desert Colossus: draws | 744 / 695 | 680 / 708 | | Hyrule Field: fps | 43.5 / 45.6 | 46.6 / 41.0 | | Hyrule Field: GPU a frame | 16.3 / 15.6 ms | 14.0 / 15.1 ms | | Hyrule Field: draws | 1,337 / 1,324 | 1,266 / 1,364 | Hyrule Field's GPU frame is shorter but its fps is not higher in run 4: there the replay's submit (13.4 ms) holds the frame. ``` vk: a frame: mirrors 563 (1020 MB): 7.51 MB synced in 1108.8 copies, 59.07 MB written back (unset) vk: a frame: mirrors 620 (1060 MB): 7.64 MB synced in 1108.3 copies, 16.20 MB written back (1024) vk: write-backs left out (600 submits): 817, 3682.0 MB; 26 bindings followed, 37 ranges the CPU read, 0 resets; CPU reads or writes of one left out: 1686 so far vk: the CPU wrote 0x117ec0e000 +0x1000, in a binding whose write-back was left out: written back from now on (RX_VK_WB_MAX_KB=0: every one) ``` One `RX_VK_PASS_TIMING=2` block of the same renderer with the switch unset, for where the write-back sits in the GPU's frame: ``` vk: 21.53 ms a frame in all (recordings' command buffers + upload command buffers; ...) vk: m pass 5.777 ms 33.2% 123.4 spans 1113.1 items 0.00 MB 46.8 us a span vk: m dispatch 3.817 ms 22.0% 146.3 spans 146.3 items 0.00 MB 26.1 us a span vk: m wb 2.101 ms 12.1% 7.6 spans 163.1 items 40.40 MB 278.2 us a span ``` ### Flags and settings None changed. Results for one existing opt-in switch: | Flag | What it is | Before this PR | After this PR | Effect | Helps / doesn't help / untested | Recommendation | |---|---|---|---|---|---|---| | `RX_VK_WB_MAX_KB=1024` | Leaves out the write-back of the parts of large GPU-written bindings the CPU does not touch | off | off (unchanged) | 61 → 16 MB written back a frame, the GPU's frame about 2.5 ms shorter on this card | Helps: RX 5600 XT 6 GB, Windows (the compute fallback). Doesn't help fps: RTX 3060 Ti, RX 580 (the README's earlier runs). Untested: other cards without mesh shaders (RX 5500 XT, RX 5700, GTX 10 series, Intel), laptops, Linux on this card | Stay off until other cards are measured; this PR adds one card where it gains | ## Testing **Tested:** the four runs above in the game on the RX 5600 XT (Windows 11, Adrenalin 26.10.1). With 1024: no device lost, the twelve screenshots (six areas, two runs) looked at with nothing wrong seen, 20 `the CPU wrote ... in a binding whose write-back was left out` lines a run and none for a read. Then a 56-minute play session by hand with 1024 on the same renderer (run 5): Goron City 66.8 fps at the median, Death Mountain Trail 48-51, Dodongo's Cavern 56.8 over 32 minutes, Kakariko 36.5; no device lost, readback waits of 0.12-0.30 ms at the median, 20 `the CPU wrote ...` lines in the first area and none for a read. `python -I -B tests/test_tested_configurations.py`: 24 of 25 pass; the one failure is the Windows line-ending one of https://git.magicbean.dev/MagicBeanDev/Longshot/issues/482. `python -I -B tests/run_all.py` (Windows, llvm-mingw, with the branch on main 7ff5256e): 533 passed, 4 failed, 34 skipped. The four fail the same way on main 7ff5256e on this PC, each run alone: `test_cpm_discovery`, `test_oot_state_tap`, `test_tested_configurations` (the line-ending one) and `test_walk_key` (`ld.lld: error: undefined symbol: hos_oot_autoload_input`). **Not tested:** plain main without the mirror retry; the areas and dungeons not named here; Linux on this card; the tests again on a main later than 7ff5256e. **Must be tested:** None (docs only). Runs of the switch on other cards without mesh shaders and on laptops are asked for in https://git.magicbean.dev/MagicBeanDev/Longshot/issues/548, with the steps for a release build. ## Risks **What could break:** nothing in the game; five documentation files and one README paragraph. https://git.magicbean.dev/MagicBeanDev/Longshot/pulls/753 leaves that paragraph alone because this PR rewrites it. **Turning it off:** revert the commit. **Compatibility:** Nothing changes. ## Issues See https://git.magicbean.dev/MagicBeanDev/Longshot/issues/548 See https://git.magicbean.dev/MagicBeanDev/Longshot/issues/482 ## Checklist ### Everyone - [x] One topic. - [x] The title is a Conventional Commit (`type(scope): description`) of the status line's type. - [x] Why gives the reason plainly and links the issues that fit (the tracker searched). - [x] Measurements has a verdict, or one line saying why there is nothing to measure. - [x] Testing names what was tested, what wasn't, and what must still be tested, where and how. - [x] Rebased on current `main`: on 7ff5256e, not on today's main; documentation only, and it merges into aa95d35f without conflicts. - [x] `python -I -B tests/run_all.py` passes (any failure that also fails on `main` is named): 4 failures, the same on main 7ff5256e on this PC (Testing). - [x] No game files, dumps, build output or generated files. - [x] `third_party/` untouched (a new folder with a `PROVENANCE.md` instead). - [x] The license is OK: GPL-3.0-or-later, nothing incompatible and nothing decompiled. ### When it applies - [x] AI assistance recorded with `Assisted-by:` / `Co-Authored-By:` trailers. - [x] Numbers from the same PC, scene and settings, at least 2 interleaved runs each, the median reported: both runs of each side are shown. - [x] A file in `docs/tested/` for each new measured run (format in `docs/tested-configurations.md`). - [x] Docs updated. 🤖 Generated with [Claude Code](https://claude.com/claude-code)

waryxxful added 1 commit

2026-11-14 19:15:03 +00:00

docs(tested): RX_VK_WB_MAX_KB=1024 on an RX 5600 XT (no mesh shaders): 2.5 ms of GPU less a frame

beanstalk/approval awaiting maintainer approval

Details

PR description / Description follows the template (pull_request_target) Successful in 4s

Details

CI / Conventional Commit title (pull_request) Has been cancelled

Details

CI / Python source checks (ubuntu-24.04, part 1/2) (pull_request) Has been cancelled

Details

CI / Python source checks (ubuntu-24.04, part 2/2) (pull_request) Has been cancelled

Details

CI / SPIR-V fixtures (global narrow load/store, direct surface atomics, F2I immediate) (pull_request) Has been cancelled

Details

CI / Workflow lint (actionlint) (pull_request) Has been cancelled

Details

Third-party parts / As their notes say (pull_request) Has been cancelled

Details

Third-party parts / The same as upstream (pull_request) Has been cancelled

Details

Website / Website on Cloudflare (pull_request) Has been cancelled

Details

02bc3bf9c3

Four interleaved runs on main 9c9d69ed plus the mirror retry (the same
renderer in all four): the cap unset twice, 1024 twice. Written back a
frame 60.8 -> 16.3 MB; GPU a frame 14.5-14.8 -> 12.0-12.1 ms in Kokiri
Forest and 18.0-18.7 -> 15.8 ms at the Desert Colossus at the same draws.
The README's paragraph on the switch names the card.

Co-Authored-By: Claude Fable 5.2 <[email protected]>

waryxxful referenced this pull request

2026-11-14 19:15:17 +00:00

Idea: runs of RX_VK_WB_MAX_KB=1024 wanted on cards without mesh shaders (2.5 ms of GPU less on an RX 5600 XT) #548

waryxxful referenced this pull request

2026-11-14 19:15:26 +00:00

Idea: cut the per-draw cost of the compute fallback (the Market at 2,300 draws: GPU 22.7 ms plus 19 ms of replay submit on an RX 5600 XT) #549

waryxxful changed title from ~~WIP: docs(tested): RX_VK_WB_MAX_KB=1024 measured on an RX 5600 XT (no mesh shaders)~~ to docs(tested): RX_VK_WB_MAX_KB=1024 measured on an RX 5600 XT (no mesh shaders)

2026-11-14 19:16:50 +00:00

waryxxful added 1 commit

2026-11-14 20:42:18 +00:00

docs(tested): a 56-minute play session with RX_VK_WB_MAX_KB=1024 on the RX 5600 XT

CI / Conventional Commit title (pull_request) Blocked by required conditions

Details

CI / Python source checks (ubuntu-24.04, part 1/2) (pull_request) Blocked by required conditions

Details

CI / Python source checks (ubuntu-24.04, part 2/2) (pull_request) Blocked by required conditions

Details

CI / SPIR-V fixtures (global narrow load/store, direct surface atomics, F2I immediate) (pull_request) Blocked by required conditions

Details

CI / Workflow lint (actionlint) (pull_request) Blocked by required conditions

Details

Third-party parts / As their notes say (pull_request) Blocked by required conditions

Details

Third-party parts / The same as upstream (pull_request) Blocked by required conditions

Details

Website / Website on Cloudflare (pull_request) Blocked by required conditions

Details

beanstalk/approval Test farm not requested

PR description / Description follows the template (pull_request_target) Successful in 4s

Details

Website / Website on Cloudflare (pull_request_target) Has been skipped

Details

e20c316c3a

Goron City, Death Mountain Trail, Dodongo's Cavern (32 minutes) and
Kakariko, played by hand on the same renderer as runs 1 to 4: no device
lost, no read of a range whose write-back was left out.

Co-Authored-By: Claude Fable 5.2 <[email protected]>

waryxxful commented

2026-11-14 20:42:45 +00:00

Author

First-time contributor

Added e20c316c: a fifth docs/tested/ file, a 56-minute play session by hand with RX_VK_WB_MAX_KB=1024 on the same renderer as runs 1 to 4. It covers what the description listed as not tested: a dungeon and a session longer than the six-minute runs.

Area	Time	fps, median	fps, 10th percentile	Draws, median	GPU a frame	Written back a frame
Goron City	6.8 min	66.8	56.6	780	9.1 ms	2.7 MB
Death Mountain Trail	3.0 min	48.3	37.9	919	14.0 ms	16.7 MB
Dodongo's Cavern	32.5 min	56.8	49.6	608	11.1 ms	3.4 MB
Death Mountain Trail	10.3 min	50.8	40.2	778	14.1 ms	17.7 MB
Kakariko	3.5 min	36.5	21.8	1,245	17.6 ms	18.3 MB

No device lost. Readback waits 0.12-0.30 ms at the median. 20 the CPU wrote ... in a binding whose write-back was left out lines, all in the first area, and none for a read. Kakariko was under 30 fps for 1.5 of its 3.5 minutes, at up to 2,527 draws: that is the per-draw cost of #549, not the switch.

There is no unset side to this run: it is a check that the switch holds up in play, not a second measurement of its gain. The description's Testing section has it now.

Added `e20c316c`: a fifth `docs/tested/` file, a 56-minute play session by hand with `RX_VK_WB_MAX_KB=1024` on the same renderer as runs 1 to 4. It covers what the description listed as not tested: a dungeon and a session longer than the six-minute runs. | Area | Time | fps, median | fps, 10th percentile | Draws, median | GPU a frame | Written back a frame | |---|---|---|---|---|---|---| | Goron City | 6.8 min | 66.8 | 56.6 | 780 | 9.1 ms | 2.7 MB | | Death Mountain Trail | 3.0 min | 48.3 | 37.9 | 919 | 14.0 ms | 16.7 MB | | Dodongo's Cavern | 32.5 min | 56.8 | 49.6 | 608 | 11.1 ms | 3.4 MB | | Death Mountain Trail | 10.3 min | 50.8 | 40.2 | 778 | 14.1 ms | 17.7 MB | | Kakariko | 3.5 min | 36.5 | 21.8 | 1,245 | 17.6 ms | 18.3 MB | No device lost. Readback waits 0.12-0.30 ms at the median. 20 `the CPU wrote ... in a binding whose write-back was left out` lines, all in the first area, and none for a read. Kakariko was under 30 fps for 1.5 of its 3.5 minutes, at up to 2,527 draws: that is the per-draw cost of https://git.magicbean.dev/MagicBeanDev/Longshot/issues/549, not the switch. There is no unset side to this run: it is a check that the switch holds up in play, not a second measurement of its gain. The description's Testing section has it now.