Observation
In one 281-run fleet, the reason on a child's materialized receipt separates survivors from deaths almost perfectly:
| receipt reason at materialization |
children settled |
settled down |
mortality |
executor-receipt-pending |
41 |
41 |
100% |
| (no reason) |
111 |
30 |
27% |
n=152 settled children with a materialization event. No child that materialized with a pending executor receipt ever survived.
Why it costs more than it looks
15 of the 41 died cold (0 tokens). The other 26 died after doing real work: 10,337,224 input tokens and 441,045 output tokens spent on children that were already, by this signal, doomed at materialization.
Their recorded down reasons:
| count |
reason |
| 14 |
execution aborted (generic — see #820, now fixed, which should split this bucket) |
| 3 |
root budget deadline exceeded |
| 3 |
child deadline exceeded |
| 2 |
pi exit 78: Pi profile execution controls failed: requested tool "agent_runtime_coordination_kb_record" is unavailable |
| 2 |
bridge 503: cli-bridge admission timed out after 30000ms (active: 12, maxActive: 12) |
| 1 |
bridge 404: no backend matches model "claude-code/tangle-router/deepseek-v4-flash" |
The last three are diagnosable at or before materialization: the tool is either mounted or not, the admission queue is either full or not, and the model wire id either matches a registered backend or does not.
The ask
Is executor-receipt-pending expected to persist on a healthy child's materialization record, or does it indicate the executor never confirmed the binding? If the latter, the supervisor has a usable fail-fast signal at materialization — before the child accumulates spend that will be discarded — rather than waiting for a deadline.
Happy to supply the journals or run a targeted probe if that would help.
Measured on runtime 0.133.3/0.133.4, bridge executor, pi and claude-code harnesses (both showed the pattern).
🤖 Generated with Claude Code
Observation
In one 281-run fleet, the
reasonon a child'smaterializedreceipt separates survivors from deaths almost perfectly:downexecutor-receipt-pendingn=152 settled children with a materialization event. No child that materialized with a pending executor receipt ever survived.
Why it costs more than it looks
15 of the 41 died cold (0 tokens). The other 26 died after doing real work: 10,337,224 input tokens and 441,045 output tokens spent on children that were already, by this signal, doomed at materialization.
Their recorded down reasons:
execution aborted(generic — see #820, now fixed, which should split this bucket)pi exit 78: Pi profile execution controls failed: requested tool "agent_runtime_coordination_kb_record" is unavailablebridge 503: cli-bridge admission timed out after 30000ms(active: 12, maxActive: 12)bridge 404: no backend matches model "claude-code/tangle-router/deepseek-v4-flash"The last three are diagnosable at or before materialization: the tool is either mounted or not, the admission queue is either full or not, and the model wire id either matches a registered backend or does not.
The ask
Is
executor-receipt-pendingexpected to persist on a healthy child's materialization record, or does it indicate the executor never confirmed the binding? If the latter, the supervisor has a usable fail-fast signal at materialization — before the child accumulates spend that will be discarded — rather than waiting for a deadline.Happy to supply the journals or run a targeted probe if that would help.
Measured on runtime 0.133.3/0.133.4, bridge executor, pi and claude-code harnesses (both showed the pattern).
🤖 Generated with Claude Code