FA-100796 / Model output stopping rules / Open access

Output stopping: A stop sequence split across chunks is missed · case 01

A stop sequence split across chunks is missed.

Verified by executionVariant 1 · 6 checks per implementationDownload source bundle ↓JSON ↗

ROOT CAUSE

The matcher checks chunks independently, so markers split between chunks are missed.

THE FAILURE

The matcher checks chunks independently, so markers split between chunks are missed.

Unsuccessful approach: Checking individual chunks can trim a local match but cannot recognize a marker crossing a boundary.

Case contract

Given chunks and a nonempty stop string, concatenate logically and return emitted text before first occurrence and stop flag. Chunk boundaries have no semantic effect.

Why this case matters

A deterministic output-control model isolates one finish or streaming boundary from any particular runtime.

1 / The failure

Exit 1

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

N = 1
observations = []
def solve(text_chunks, stop):
    joined=''.join(text_chunks)
    found=any(stop in chunk for chunk in text_chunks)
    return joined,found
def check(label, actual, expected):
    observations.append({"check": label, "actual": actual, "expected": expected, "passed": actual == expected})
check('split two chunks',solve(['ab','cd','z'],'bc'),('a',True))
check('single chunk',solve(['xSTOPy'],'STOP'),('x',True))
check('no marker',solve(['abc','def'],'ZZ'),('abcdef',False))
check('empty chunks',solve([], 'x'),('',False))
check('marker across empty chunk',solve(['a','','xy'],'xy'),('a',True))
check('marker at boundary',solve(['a','b'],'ab'),('',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
split two chunks	`['abcdz', False]`	`['a', True]`	Failed
single chunk	`['xSTOPy', True]`	`['x', True]`	Failed
no marker	`['abcdef', False]`	`['abcdef', False]`	Passed
empty chunks	`['', False]`	`['', False]`	Passed
marker across empty chunk	`['axy', True]`	`['a', True]`	Failed
marker at boundary	`['ab', False]`	`['', True]`	Failed

SHA-256 / b4513d6cda969a1235599a48300826f18e68f8d6dc964202f2f51ab40deafb1b

2 / The unsuccessful fix

Exit 1

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

N = 1
observations = []
def solve(text_chunks, stop):
    joined=''.join(text_chunks)
    for chunk in text_chunks:
        if stop in chunk:return joined[:joined.find(stop)],True
    return joined,False
def check(label, actual, expected):
    observations.append({"check": label, "actual": actual, "expected": expected, "passed": actual == expected})
check('split two chunks',solve(['ab','cd','z'],'bc'),('a',True))
check('single chunk',solve(['xSTOPy'],'STOP'),('x',True))
check('no marker',solve(['abc','def'],'ZZ'),('abcdef',False))
check('empty chunks',solve([], 'x'),('',False))
check('marker across empty chunk',solve(['a','','xy'],'xy'),('a',True))
check('marker at boundary',solve(['a','b'],'ab'),('',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
split two chunks	`['abcdz', False]`	`['a', True]`	Failed
single chunk	`['x', True]`	`['x', True]`	Passed
no marker	`['abcdef', False]`	`['abcdef', False]`	Passed
empty chunks	`['', False]`	`['', False]`	Passed
marker across empty chunk	`['a', True]`	`['a', True]`	Passed
marker at boundary	`['ab', False]`	`['', True]`	Failed

SHA-256 / 499a55d0f58db236eb8f13a9d00463b085a722e76330e703dda52e1d83c8aa7c

HELD IN THE MEMBER ARCHIVE

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

This mechanism has 6 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

This toy contract covers the supplied token records only; it does not claim compatibility with a particular model runtime or tokenizer. 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 2026-09-30T16:48:39.593504+00:00.

Case digest / 5d9e42c8e75ba3a9b9fb8b19754d2b19a69ec012e2cd402bb3681429bcdf8ec7