FA-99861 / Agent workflow termination / Open access

Workflow: tool-call allowance counts issued calls only · case 01

The loop issues too many tools when some requests are still in flight.

Verified by executionVariant 1 · 8 checks per implementationDownload source bundle ↓JSON ↗

ROOT CAUSE

The budget check subtracts completed calls but not pending calls.

THE FAILURE

The budget check subtracts completed calls but not pending calls.

Unsuccessful approach: Counting pending calls twice rejects work that still fits the budget.

Case contract

Return whether one more call can be issued when completed + pending is strictly below limit.

Why this case matters

This toy decision rule isolates one bounded agent-loop completion contract.

1 / The failure

Exit 1

"""Failure Map reference implementation. Python standard library only."""
import json

N = 1
observations = []
def solve(completed, pending, limit):
    return completed < limit
def check(label, actual, expected):
    observations.append({"check": label, "actual": actual, "expected": expected, "passed": actual == expected})
check('all budget free', solve(0,0,N),True)
check('pending consumes last slot', solve(N-1,1,N),False)
check('one slot remains', solve(0,1,N+1),True)
check('single pending with room', solve(0,1,2),True)
check('limit exhausted', solve(N,0,N),False)
check('zero limit', solve(0,0,0),False)
check('completed calls', solve(1,0,1),False)
check('empty counters', solve(0,0,1),True)
print(json.dumps({"observations": observations, "passed": all(x["passed"] for x in observations)}, ensure_ascii=False))
raise SystemExit(0 if all(x["passed"] for x in observations) else 1)

Boundary fixture	Actual	Expected	Outcome
all budget free	`True`	`True`	Passed
pending consumes last slot	`True`	`False`	Failed
one slot remains	`True`	`True`	Passed
single pending with room	`True`	`True`	Passed
limit exhausted	`False`	`False`	Passed
zero limit	`False`	`False`	Passed
completed calls	`False`	`False`	Passed
empty counters	`True`	`True`	Passed

SHA-256 / 08b22ee96a8a435266b33b2f64ed7889e653d20e30d5f82fb3679a64db3df456

2 / The unsuccessful fix

Exit 1

"""Failure Map reference implementation. Python standard library only."""
import json

N = 1
observations = []
def solve(completed, pending, limit):
    return completed + 2*pending < limit
def check(label, actual, expected):
    observations.append({"check": label, "actual": actual, "expected": expected, "passed": actual == expected})
check('all budget free', solve(0,0,N),True)
check('pending consumes last slot', solve(N-1,1,N),False)
check('one slot remains', solve(0,1,N+1),True)
check('single pending with room', solve(0,1,2),True)
check('limit exhausted', solve(N,0,N),False)
check('zero limit', solve(0,0,0),False)
check('completed calls', solve(1,0,1),False)
check('empty counters', solve(0,0,1),True)
print(json.dumps({"observations": observations, "passed": all(x["passed"] for x in observations)}, ensure_ascii=False))
raise SystemExit(0 if all(x["passed"] for x in observations) else 1)

Boundary fixture	Actual	Expected	Outcome
all budget free	`True`	`True`	Passed
pending consumes last slot	`False`	`False`	Passed
one slot remains	`False`	`True`	Failed
single pending with room	`False`	`True`	Failed
limit exhausted	`False`	`False`	Passed
zero limit	`False`	`False`	Passed
completed calls	`False`	`False`	Passed
empty counters	`True`	`True`	Passed

SHA-256 / 3101865f9bba708004fd0fd69851e96497c6ac771eded7306970eecd1b2b2b41

HELD IN THE MEMBER ARCHIVE

The verified repair and its recorded checks are member-only.

This mechanism has 8 recorded checks per implementation. The open-access tier publishes the failure and the unsuccessful fix; the repaired source that passes every check, and the observations that prove it, are available to members.

Every case sharing this mechanism uses the same contract and the same repair, so this one record is held back for all of them.

Member access is invitation-based. Sign in with your invited account to inspect the repair.

Verification & scope

Counters are nonnegative integers and the limit is total calls per run. This reproducer isolates one failure mechanism. Results cover the supplied fixtures. Variants within a family share a test contract and should remain grouped when constructing evaluation splits. Related mechanisms with a shared evaluation_group must also remain together; these controlled models are not independent production incidents.

Observations recorded using Python 3.12.14 at 2026-09-30T16:48:28.969387+00:00.

Case digest / 5fa1b638c2700086454c9ce4898328c5d49c1844e15883944731ac54dbb1dc64