[frahlg]

What

go/internal/sitesim drops its port of the old gG curve and imports github.com/srcfl/device-simulator/fuse at 736051f (srcfl/device-simulator#9, merged 23 Oct). Each plant second steps every phase through fuse.Lookup("iec_gg").Step at the package's default 25 °C. Fuse.Preheat starts a phase warm through the package's Preheat. No tuning of the model, and no change to Core.

The site rig judges Core by the same package: on the device-simulator stack (#8 → #10 → #11, now on main), cmd/simulator steps fuse/, and the stack head 3bf737d carries the same fuse/fuse.go as main. So go/test/site (#1797) and sitesim share one fuse model once #1797 moves devsim.version to that stack; it still pins cba7142, which has the old curve.

Test-only. Core's binaries do not link device-simulator (go list -deps ./cmd/... for linux and darwin).

Preheat

Only where a scenario starts with the house near its limit:

• the field matrix's three probe/25A/auto/mixed/*/car3p/2-20-2/1p-to-3p cases (L2 at 20 A of 25 A), through a new fieldCase.warm that is not part of the case name;
• the new property test, at its first random load.

Every other case starts cold, as before.

Old curve and new model

Seconds to blow at a held current, from cold (the old curve was the same at every rating) and, for the model, warm after hours at the fuse less 2 A:

× In	old	16 A cold	16 A warm	20 A cold	20 A warm	25 A cold	25 A warm
1.05	3,600	never	never	never	never	never	never
1.25	3,600	never	never	never	never	never	never
1.4	2,619	never	never	5,772	5,098	3,819	3,079
1.5	2,158	2,420	1,749	1,726	1,162	1,630	1,013
1.6	1,801	1,340	766	993	512	963	438
1.8	1,008	443	31	345	30	342	24
2	600	32	10	42	12	41	11
2.5	248	7	4	9	5	9	4
3	121	4	2	5	2	5	2
5	20	1	1	1	1	1	1

They agree around 1.5 ×. At 2 × and above the old curve was 14 to 35 times slower (600 s at 2 ×, against 32 to 42 s). Just above the rating it blew in an hour, where a gG fuse never does. Cooling differs too: the old damage fell to zero over 300 s, while the model keeps heat in the body and holder. After 20 s at 2 × 25 A and 15 min at 0 A, a second 2 × overload melts in 38 s, against 41 s cold.

Verdicts

None changed.

• TestSiteSimScenarios: every row as on master.
• Field matrix: 340 cases, 29 fail an invariant (idle 23, own-raise 6), the same rows. FTW_FIELD_KNOWN=write leaves sitesim_field_known.txt as it was.

No case reaches the range where the two differ. The matrix peaks at 1.25 × (20 A on 16 A, the kettle); the scenarios peak at 1.56 × for 3 s (39 A on 25 A, "meter every 30 s after a cut"). A preloaded case that fails, or a 2 × case that passes, needs an overload of 1.4 to 2 × held for tens of seconds to minutes, and the matrix has none. The warm 2-20-2 cases peak at 20 A.

Property test

TestSiteSimRandomHouse, in go/internal/control/sitesim_property_test.go, runs seeds 1 to 5 for an hour each. Each seed picks a fuse (16, 20 or 25 A) and a charger delay (2, 5 or 40 s). The house steps every 15 to 180 s to random currents up to the fuse less 2 A on each phase. The fuse starts warm from the first load, and the car charges on Charge now. simSite.check judges every plant second:

• the fuse never blows;
• no phase stays over the fuse less 1 A (heat-read, as invariant (a)) longer than the charger's response plus 3 s, or 3 s after a raise of Core's own.

The run goes on after a failure, so a later blow still shows. Idle time with room is logged, not judged; the matrix judges it with the fuse's own memory. All five seeds take 0.5 s wall.

seed	fuse	charger	house steps	peak phase	fuse damage, max	car idle with room
1	20 A	2 s	34	25.1 A	0.41	14 min 18 s
2	25 A	40 s	36	37.8 A	0.60	17 min 7 s
3	16 A	40 s	32	19.9 A	0.45	11 min 16 s
4	16 A	40 s	44	20.2 A	0.47	9 min 18 s
5	20 A	5 s	35	26.9 A	0.44	16 min 18 s

The fuse holds in every seed. Seed 2 breaks (a), and the test lists it as known.

Finding: a landed cut still covers the next overage (seed 2)

25 A fuse, 40 s charger with no declared response time, so Core counts a cut as possibly unshown for 180 s.

1. 12:31:44: L2 rises to 27.2 A. At 12:32:51 its heat budget runs out and Core cuts 8,970 → 5,520 W. The cut lands at 12:33:31.
2. L1 then reads 24.8 A: over 24 A by reading, under it by heat. The run over the limit never ends.
3. 12:35:04: L2 steps to 30.8 A, over 1.2 × 25 A. Core's log reads overage 4,712 W, unshown cut 3,450 W (the 12:32:51 cut, landed 94 s before), offer 5,520 W, cap 4,258 W. Core sends 4,140 W (6 A), not 0.
4. L2 reads 30.8 A until the 6 A order lands at 12:35:45, then 28.8 A. (a) fails at 12:35:52, 47 s after the step, past the 46 s it allows.

The fuse is not at risk: 1.15 × never melts, and damage peaks at 0.60. It belongs to the matrix's netting family, where a cut nets the overage against the run's earlier cuts (ev_fuse_budget.go:436). randomHouseKnown lists it; the test fails once it passes, as the matrix's known rows do.

Cost

The model costs CPU. go test ./internal/sitesim/ ./internal/control/ -run 'SiteSim|Fuse' took 72 s user and 15 s wall on master, and 100 s and 21 s here (M5, 10 cores). Each Step finds the melting time twice: from cold for its Status, and from now. At a current that never melts, both searches scan to 10⁷ s.

Dependency

go get github.com/srcfl/device-simulator@736051f. The fuse package imports only the standard library. Minimal version selection raises github.com/go-ole/go-ole 1.2.6 → 1.3.0 (indirect, through gopsutil's WMI), which only Windows builds link. Dependabot's #1749 also edits go/go.mod.

Open PRs on these files

#1782 and #1789 (drafts) change the meter reading in sitesim.go and the checks in control/sitesim_test.go. This PR leaves both files alone: the fuse keeps step and AnyBlown, and Damage and Blown mirror the package's state. #1807 edits other hunks of sitesim_field_test.go.

Test evidence

This branch, M5 MacBook:

$ cd go && go test ./internal/sitesim/... ./internal/control/... -count=1
ok  	github.com/srcfl/ftw/go/internal/sitesim	0.156s
ok  	github.com/srcfl/ftw/go/internal/control	23.395s

$ FTW_FIELD_KNOWN=write go test ./internal/sitesim/ ./internal/control/ -run 'SiteSim|Fuse' -count=1 -v
field matrix: 340 cases, 29 fail an invariant; by kind map[idle:23 own-raise:6]
(sitesim_field_known.txt unchanged)

$ make verify
ok  	github.com/srcfl/ftw/go/internal/control	28.426s
ok  	github.com/srcfl/ftw/go/internal/sitesim	0.443s
verify: vet + test + build clean

🤖 Generated with Claude Code

[chatgpt-codex-connector]

Codex Review Summary

This comment shows the latest Codex review activity on this pull request.

Review	Status	Commit	Review trigger
📝 Code Review	✅ Completed 2026-10-23T05:41:29.164288Z	4c124bd	PR opened

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review" or "@codex security review".

Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings.