[SandroMaglione]

Summary

• Work invoked by an entered state now starts in the sender's turn and runs until it first suspends. Before, every invocation was hosted by a process whose worker was scheduled for a later turn, so even the first synchronous step ran after send returned. A browser call that requires user activation (window.open, navigator.clipboard.writeText, HTMLMediaElement.play) therefore ran outside the click that sent the event; Chrome tolerates that through its activation window, stricter browsers block it.
• The invocation's outcome is delivered one turn later, even when the work completes without suspending. The entered state is always published before onDone/onFailure leaves it, a synchronous defect still fails the machine after send returns, and a loop of instantly completing invocations still yields between steps. The outcome of a state exited in between is still discarded.
• Implementation: runtimeGeneric forks the hosted worker with startImmediately (which also makes its trailing yieldNow unnecessary), and invocation.ts delivers outcomes behind one yieldNow.
• No public API or type changes. Cluster entities are unaffected: they reject invocations and only use the pure planner. Stream sources are unchanged: Stream.callback registers in its own fiber.
• test/machine/InvocationTurn.test.ts covers both runtime strategies: invoked work and an invoked child's initial work start before send returns (fail on main), and the entered state is published first, outcomes of exited states are discarded, synchronous defects fail later, and instant-completion loops do not run inside send (pass on main and here).
• Docs: MachineRef JSDoc, the agent guide ("Let states own running work") and the Effect Atom guide ("Send without subscribing").

Changeset

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

Minor: the timing of invoked work is observable (it now runs on the sender until it first suspends).

Validation

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

Also checked against the consuming apps on a patched 0.45.1 build: creator-self-serve 357/357 tests, and in Chrome the Google sign-in popup opens within the click task (false on 0.45.1, true with this change).

🤖 Generated with Claude Code

[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.state (3 children)	2,664	2,664	0 (0.0%)
Machine.make (3 states, 2 events)	8,932	8,932	0 (0.0%)
machine.handle (3 states, 2 transitions)	39,449	39,449	0 (0.0%)
declared transition (10 named branches)	60,402	60,402	0 (0.0%)
atomic construction, guards, and reentry	85,710	85,710	0 (0.0%)
registered invocation (input-taking Effect)	39,819	39,819	0 (0.0%)
machine.handle (depth 24)	194,531	194,531	0 (0.0%)
machine.handle (wide depth 16)	218,238	218,238	0 (0.0%)
machine.handle (parallel/history/choice)	91,642	91,642	0 (0.0%)
machine definition (3 independent implementations)	99,173	99,173	0 (0.0%)
machine exact input/output/error/services	82,941	82,941	0 (0.0%)
execution adapter readiness	71,428	71,428	0 (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.state (3 children)	2,609	2,609	0 (0.0%)
Machine.make (3 states, 2 events)	6,260	6,260	0 (0.0%)
machine.handle (3 states, 2 transitions)	30,517	30,517	0 (0.0%)
declared transition (10 named branches)	51,849	51,849	0 (0.0%)
atomic construction, guards, and reentry	70,370	70,370	0 (0.0%)
registered invocation (input-taking Effect)	32,682	32,682	0 (0.0%)
machine.handle (depth 24)	177,222	177,222	0 (0.0%)
machine.handle (wide depth 16)	201,707	201,707	0 (0.0%)
machine.handle (parallel/history/choice)	79,144	79,144	0 (0.0%)
machine definition (3 independent implementations)	88,492	88,492	0 (0.0%)
machine exact input/output/error/services	70,732	70,732	0 (0.0%)
execution adapter readiness	31,922	31,922	0 (0.0%)

Check times (informational)

Scenario	Base	PR
Effect only	0.02s	0.02s
Import effect-machine	0.02s	0.02s
Machine.state (3 children)	0.05s	0.05s
Machine.make (3 states, 2 events)	0.08s	0.08s
machine.handle (3 states, 2 transitions)	0.19s	0.21s
declared transition (10 named branches)	0.28s	0.22s
atomic construction, guards, and reentry	0.32s	0.31s
registered invocation (input-taking Effect)	0.20s	0.19s
machine.handle (depth 24)	0.36s	0.38s
machine.handle (wide depth 16)	0.43s	0.39s
machine.handle (parallel/history/choice)	0.33s	0.32s
machine definition (3 independent implementations)	0.32s	0.34s
machine exact input/output/error/services	0.29s	0.32s
execution adapter readiness	0.28s	0.28s

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 9V74 80-Core Processor with Node v24.21.0.

Pull request baseline

Scenario	Effect Machine
Plan counter transitions	146,065 transitions/s
Drain burst with terminal fence	487,518 increments/s
Drain burst with a change observer	457,111 increments/s
Lookup and send to one child	381,461 increments/s
Start and stop a machine	155,521 machines/s
Start and stop a parent with one child	46,529 families/s
Complete an Effect invocation with tracing enabled	12,117 invocations/s
Complete an Effect invocation with tracing disabled	15,051 invocations/s
Plan transitions through a compound state	132,385 transitions/s
Plan transitions through parallel regions	106,885 transitions/s
Drain burst through a compound state	412,868 events/s
Drain burst through two parallel regions	388,691 events/s
Drain a compound-state burst with a change observer	386,207 events/s

Process runtime reference points

Scenario	Effect Machine
Start and stop a raw generic process	43,814 processes/s
Start and stop a raw compiled process	92,799 processes/s

Memory profile	Effect Machine
Idle machine	1.9 KiB
Raw generic managed process	11.2 KiB
Raw compiled process	3.1 KiB
Two independent idle machines	3.5 KiB
Idle parent with one child	6.0 KiB
Parent with observed child registry	9.5 KiB
Parent with observed invoked child snapshots	6.9 KiB

Effect Machine change from base

Metric	Base	Base variability	PR	PR variability	Difference
Plan counter transitions	149,183 transitions/s	0.5% MAD	146,065 transitions/s	0.2% MAD	-2.1%
Drain burst with terminal fence	492,675 increments/s	0.7% MAD	487,518 increments/s	0.4% MAD	-1.0%
Drain burst with a change observer	460,544 increments/s	0.5% MAD	457,111 increments/s	0.3% MAD	-0.7%
Lookup and send to one child	395,576 increments/s	0.4% MAD	381,461 increments/s	0.3% MAD	-3.6%
Start and stop a machine	155,039 machines/s	0.6% MAD	155,521 machines/s	0.4% MAD	+0.3%
Start and stop a parent with one child	46,551 families/s	0.4% MAD	46,529 families/s	1.1% MAD	-0.0%
Complete an Effect invocation with tracing enabled	13,626 invocations/s	0.8% MAD	12,117 invocations/s	0.3% MAD	-11.1%
Complete an Effect invocation with tracing disabled	17,533 invocations/s	1.2% MAD	15,051 invocations/s	0.8% MAD	-14.2%
Plan transitions through a compound state	132,517 transitions/s	0.4% MAD	132,385 transitions/s	0.2% MAD	-0.1%
Plan transitions through parallel regions	107,654 transitions/s	0.3% MAD	106,885 transitions/s	0.8% MAD	-0.7%
Drain burst through a compound state	409,838 events/s	0.7% MAD	412,868 events/s	0.2% MAD	+0.7%
Drain burst through two parallel regions	382,913 events/s	0.3% MAD	388,691 events/s	0.6% MAD	+1.5%
Drain a compound-state burst with a change observer	386,147 events/s	0.0% MAD	386,207 events/s	0.1% MAD	+0.0%
Idle machine heap per unit	1.9 KiB	0.3% MAD	1.9 KiB	0.1% MAD	-0.2%
Raw generic managed process heap per unit	11.2 KiB	0.1% MAD	11.2 KiB	0.0% MAD	-0.2%
Raw compiled process heap per unit	3.1 KiB	0.1% MAD	3.1 KiB	0.0% MAD	+0.0%
Two independent idle machines heap per unit	3.5 KiB	0.0% MAD	3.5 KiB	0.0% MAD	+0.0%
Idle parent with one child heap per unit	6.0 KiB	0.0% MAD	6.0 KiB	0.0% MAD	-0.0%
Parent with observed child registry heap per unit	9.5 KiB	0.0% MAD	9.5 KiB	0.1% MAD	+0.1%
Parent with observed invoked child snapshots heap per unit	6.9 KiB	0.0% MAD	6.9 KiB	0.1% MAD	+0.0%

Process runtime reference change from base

Metric	Base	Base variability	PR	PR variability	Difference
Start and stop a raw generic process	38,537 processes/s	1.9% MAD	43,814 processes/s	0.5% MAD	+13.7%
Start and stop a raw compiled process	92,276 processes/s	0.7% MAD	92,799 processes/s	0.3% MAD	+0.6%

Regression guard

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

Versions and interpretation

• Effect Machine: 0.45.1

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.