Fix machine boundary validation and implementation ownership

[SandroMaglione]

Summary

• Keep Machine.can behavior and fix its types to accept internal events. MachineRef.send still accepts only public events; the JSDoc example is compiled in CI.
• Preserve interruption during snapshot serialization, revalidate reused event and emission values, and capture machine definitions so later caller mutations cannot change their behavior.
• Make MachineTest.verify lazy and compare decoded values without losing information through JSON. Extract target builders, child bookkeeping, invocation definitions, and shared requirement types into focused internal modules.
• Serialize devtools refreshes, stop watcher work with its server scope, and prevent lint diagnostics for shadowed machine bindings. Add regression tests and strengthen generic-versus-optimized comparisons.

Changeset

• Added or updated for a library or package-metadata change
• Not required because this PR does not change a publishable package

Compatible fixes and implementation improvements use a patch changeset.

Validation

• pnpm check
• Automated type-performance measurement passed or was not required
• Automated runtime- and memory-performance measurement passed or was not required

Local pnpm perf:types and pnpm perf:runtime passed. Hosted CI, type performance, and runtime performance passed, and both comparison artifacts were inspected. Type scenarios added at most four instantiations. Across five independent processes per version, throughput changes ranged from -0.8% to +1.9%, heap usage remained essentially unchanged, and the regression guard passed.

[github-actions]

Type performance

Measured with TypeScript 6.0.3 and skipLibCheck=true.

Scenario	Base	PR	Difference
Effect only	55	55	0 (0.0%)
Import effect-machine	55	55	0 (0.0%)
Machine.states (3 states)	3,039	3,039	0 (0.0%)
Machine.make (3 states, 2 events)	10,787	10,787	0 (0.0%)
machine.handle (3 states, 2 transitions)	28,551	28,551	0 (0.0%)
fluent transition (10 named branches)	122,775	122,777	+2 (+0.0%)
fluent invocation (state-dependent Effect)	108,429	108,429	0 (0.0%)
machine.handle (depth 24)	207,970	207,972	+2 (+0.0%)
machine.handle (wide depth 16)	236,332	236,334	+2 (+0.0%)
machine.handle (parallel/history/choice)	144,812	144,814	+2 (+0.0%)
machine definition (3 independent implementations)	143,617	143,619	+2 (+0.0%)
machine exact input/output/error/services	129,873	129,877	+4 (+0.0%)
execution adapter readiness	136,469	136,468	-1 (-0.0%)

Marginal instantiations are measured against the matching setup without that API call:

Scenario	Base	PR	Difference
Import effect-machine	0	0	0
Machine.states (3 states)	2,984	2,984	0 (0.0%)
Machine.make (3 states, 2 events)	7,740	7,740	0 (0.0%)
machine.handle (3 states, 2 transitions)	17,764	17,764	0 (0.0%)
fluent transition (10 named branches)	112,498	112,500	+2 (+0.0%)
fluent invocation (state-dependent Effect)	99,018	99,018	0 (0.0%)
machine.handle (depth 24)	187,474	187,476	+2 (+0.0%)
machine.handle (wide depth 16)	217,043	217,045	+2 (+0.0%)
machine.handle (parallel/history/choice)	124,167	124,169	+2 (+0.0%)
machine definition (3 independent implementations)	127,576	127,578	+2 (+0.0%)
machine exact input/output/error/services	115,217	115,221	+4 (+0.0%)
execution adapter readiness	106,299	106,298	-1 (-0.0%)

Check times (informational)

Scenario	Base	PR
Effect only	0.02s	0.02s
Import effect-machine	0.02s	0.02s
Machine.states (3 states)	0.07s	0.07s
Machine.make (3 states, 2 events)	0.12s	0.12s
machine.handle (3 states, 2 transitions)	0.19s	0.19s
fluent transition (10 named branches)	0.41s	0.42s
fluent invocation (state-dependent Effect)	0.41s	0.42s
machine.handle (depth 24)	0.61s	0.62s
machine.handle (wide depth 16)	0.63s	0.63s
machine.handle (parallel/history/choice)	0.50s	0.50s
machine definition (3 independent implementations)	0.52s	0.52s
machine exact input/output/error/services	0.47s	0.48s
execution adapter readiness	0.48s	0.47s

Type instantiations are the comparison metric. Check time varies with runner load and is informational only.

[github-actions]

Runtime performance

Median of 5 independent benchmark processes on AMD EPYC 7763 64-Core Processor with Node v24.20.0.

Pull request baseline

Scenario	Effect Machine
Plan counter transitions	121,970 transitions/s
Drain burst with terminal fence	391,577 increments/s
Drain burst with a change observer	364,206 increments/s
Lookup and send to one child	319,279 increments/s
Start and stop a machine	143,390 machines/s
Start and stop a parent with one child	28,965 families/s
Plan transitions through a compound state	105,098 transitions/s
Plan transitions through parallel regions	86,785 transitions/s
Drain burst through a compound state	309,812 events/s
Drain burst through two parallel regions	306,028 events/s
Drain a compound-state burst with a change observer	296,104 events/s

Process runtime reference points

Scenario	Effect Machine
Start and stop a raw generic process	14,969 processes/s
Start and stop a raw compiled process	62,073 processes/s

Memory profile	Effect Machine
Idle machine	1.7 KiB
Raw generic managed process	13.9 KiB
Raw compiled process	3.2 KiB
Two independent idle machines	3.4 KiB
Idle parent with one child	5.8 KiB
Parent with observed child registry	9.8 KiB
Parent with observed invoked child snapshots	6.3 KiB

Effect Machine change from base

Metric	Base	Base variability	PR	PR variability	Difference
Plan counter transitions	121,257 transitions/s	0.8% MAD	121,970 transitions/s	2.4% MAD	+0.6%
Drain burst with terminal fence	388,619 increments/s	1.0% MAD	391,577 increments/s	0.5% MAD	+0.8%
Drain burst with a change observer	360,143 increments/s	0.2% MAD	364,206 increments/s	0.4% MAD	+1.1%
Lookup and send to one child	316,495 increments/s	0.5% MAD	319,279 increments/s	0.5% MAD	+0.9%
Start and stop a machine	143,823 machines/s	1.1% MAD	143,390 machines/s	1.0% MAD	-0.3%
Start and stop a parent with one child	28,873 families/s	2.0% MAD	28,965 families/s	0.7% MAD	+0.3%
Plan transitions through a compound state	105,826 transitions/s	1.5% MAD	105,098 transitions/s	2.1% MAD	-0.7%
Plan transitions through parallel regions	87,457 transitions/s	0.7% MAD	86,785 transitions/s	0.2% MAD	-0.8%
Drain burst through a compound state	304,448 events/s	2.6% MAD	309,812 events/s	2.1% MAD	+1.8%
Drain burst through two parallel regions	301,802 events/s	1.6% MAD	306,028 events/s	2.0% MAD	+1.4%
Drain a compound-state burst with a change observer	290,559 events/s	1.6% MAD	296,104 events/s	2.0% MAD	+1.9%
Idle machine heap per unit	1.8 KiB	0.2% MAD	1.7 KiB	0.0% MAD	-0.3%
Raw generic managed process heap per unit	13.9 KiB	0.0% MAD	13.9 KiB	0.0% MAD	+0.0%
Raw compiled process heap per unit	3.2 KiB	0.3% MAD	3.2 KiB	0.1% MAD	+0.6%
Two independent idle machines heap per unit	3.4 KiB	0.1% MAD	3.4 KiB	0.0% MAD	+0.0%
Idle parent with one child heap per unit	5.8 KiB	0.1% MAD	5.8 KiB	0.0% MAD	+0.0%
Parent with observed child registry heap per unit	9.8 KiB	0.0% MAD	9.8 KiB	0.0% MAD	+0.0%
Parent with observed invoked child snapshots heap per unit	6.3 KiB	0.0% MAD	6.3 KiB	0.1% MAD	-0.4%

Process runtime reference change from base

Metric	Base	Base variability	PR	PR variability	Difference
Start and stop a raw generic process	14,994 processes/s	1.4% MAD	14,969 processes/s	1.1% MAD	-0.2%
Start and stop a raw compiled process	61,763 processes/s	2.2% MAD	62,073 processes/s	2.4% MAD	+0.5%

Regression guard

No large, noise-adjusted throughput or heap regressions detected.

Versions and interpretation

• Effect Machine: 0.31.2

Higher throughput is better; lower heap is better. Variability is the median absolute deviation across independent processes, relative to their median. Small differences on shared GitHub-hosted hardware remain informational; the required guard rejects only large changes beyond the measured noise allowance.