FAILURE MAP
← Case archive

FA-100751 / Model output stopping rules / Open access

Output stopping: Token ID zero is treated as a disabled stop token · case 01

Token ID zero is treated as a disabled stop token.

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

ROOT CAUSE

A truthiness check conflates token ID zero with the absence of a token.

THE FAILURE

A truthiness check conflates token ID zero with the absence of a token.

Unsuccessful approach: Recognizing the ID while retaining it leaves the stop marker visible in returned output.

Case contract

Given generated token IDs and a set of stop IDs, truncate before the first stop ID. Zero is an ordinary valid ID. Return (prefix, found_bool).

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(tokens, stop_ids):
    for i,t in enumerate(tokens):
        if t and t in stop_ids:return tokens[:i],True
    return tokens,False
def check(label, actual, expected):
    observations.append({"check": label, "actual": actual, "expected": expected, "passed": actual == expected})
check('zero stop',solve([4,0,7],[0]),([4],True))
check('nonzero stop',solve([2,3,5],[5]),([2,3],True))
check('zero not configured',solve([0,1],[8]),([0,1],False))
check('stop first',solve([0,1],[0]),([],True))
check('empty tokens',solve([],[0]),([],False))
check('repeat before stop',solve([3,3,0],[0]),([3,3],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 fixtureActualExpectedOutcome
zero stop[[4, 0, 7], False][[4], True]Failed
nonzero stop[[2, 3], True][[2, 3], True]Passed
zero not configured[[0, 1], False][[0, 1], False]Passed
stop first[[0, 1], False][[], True]Failed
empty tokens[[], False][[], False]Passed
repeat before stop[[3, 3, 0], False][[3, 3], True]Failed

SHA-256 / e4b664703bd8b3e52baa5d10c42a8105eee103d2c953074f86472836f8b8ab8d

2 / The unsuccessful fix

Exit 1
"""Failure Map reference implementation. Python standard library only."""
import json

N = 1
observations = []
def solve(tokens, stop_ids):
    for i,t in enumerate(tokens):
        if t in stop_ids:return tokens[:i+1],True
    return tokens,False
def check(label, actual, expected):
    observations.append({"check": label, "actual": actual, "expected": expected, "passed": actual == expected})
check('zero stop',solve([4,0,7],[0]),([4],True))
check('nonzero stop',solve([2,3,5],[5]),([2,3],True))
check('zero not configured',solve([0,1],[8]),([0,1],False))
check('stop first',solve([0,1],[0]),([],True))
check('empty tokens',solve([],[0]),([],False))
check('repeat before stop',solve([3,3,0],[0]),([3,3],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 fixtureActualExpectedOutcome
zero stop[[4, 0], True][[4], True]Failed
nonzero stop[[2, 3, 5], True][[2, 3], True]Failed
zero not configured[[0, 1], False][[0, 1], False]Passed
stop first[[0], True][[], True]Failed
empty tokens[[], False][[], False]Passed
repeat before stop[[3, 3, 0], True][[3, 3], True]Failed

SHA-256 / 7e82458f88955dbce38509c58983a6fc4faf113bd95cf4c09d41623944a3efb0

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.

Sign in to the archive ↗

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 2026-09-30T16:48:39.096211+00:00.

Case digest / 1a039ef18af642477ef3bcb935e2b775300ac943fceb670a7e4eb35a1e3d1bb6