FAILURE MAP
← Case archive

FA-98481 / Beam search bookkeeping / Open access

Beam search: length penalty counts prompt tokens as generated output · case 01

length penalty counts prompt tokens as generated output.

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

ROOT CAUSE

The length field includes a fixed prompt prefix that is shared by all candidates.

THE FAILURE

The length field includes a fixed prompt prefix that is shared by all candidates.

Unsuccessful approach: Subtracting one token regardless of prompt size leaves the penalty distorted.

Case contract

Rows [id,score,total_length]; normalize by max(1,total_length-prompt_length) and return scores by ID.

Why this case matters

Beam decoders rely on consistent hypothesis state while expanding and ranking candidates.

1 / The failure

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

N = 1
observations = []
def solve(rows, prompt_length):
    return {r[0]:r[1]/r[2] for r in rows}
def check(label, actual, expected):
    observations.append({"check": label, "actual": actual, "expected": expected, "passed": actual == expected})
check('shared prompt removed',solve([['a',N,N+2]],N),{'a':N/2})
check('prompt only floor',solve([['a',N,N]],N),{'a':float(N)})
check('two generated',solve([['b',2*N,N+2]],N),{'b':float(N)})
check('empty input',solve([],N),{})
check('zero prompt',solve([['c',N,2]],0),{'c':N/2})
check('long prefix',solve([['d',2*N,3*N]],2*N),{'d':2.0})
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
shared prompt removed{'a': 0.3333333333333333}{'a': 0.5}Failed
prompt only floor{'a': 1.0}{'a': 1.0}Passed
two generated{'b': 0.6666666666666666}{'b': 1.0}Failed
empty input{}{}Passed
zero prompt{'c': 0.5}{'c': 0.5}Passed
long prefix{'d': 0.6666666666666666}{'d': 2.0}Failed

SHA-256 / f78ce5141068858b2603d384bbd1f535f8cb34cf759dea0015905b74b4076800

2 / The unsuccessful fix

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

N = 1
observations = []
def solve(rows, prompt_length):
    return {r[0]:r[1]/max(1,r[2]-1) for r in rows}
def check(label, actual, expected):
    observations.append({"check": label, "actual": actual, "expected": expected, "passed": actual == expected})
check('shared prompt removed',solve([['a',N,N+2]],N),{'a':N/2})
check('prompt only floor',solve([['a',N,N]],N),{'a':float(N)})
check('two generated',solve([['b',2*N,N+2]],N),{'b':float(N)})
check('empty input',solve([],N),{})
check('zero prompt',solve([['c',N,2]],0),{'c':N/2})
check('long prefix',solve([['d',2*N,3*N]],2*N),{'d':2.0})
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
shared prompt removed{'a': 0.5}{'a': 0.5}Passed
prompt only floor{'a': 1.0}{'a': 1.0}Passed
two generated{'b': 1.0}{'b': 1.0}Passed
empty input{}{}Passed
zero prompt{'c': 1.0}{'c': 0.5}Failed
long prefix{'d': 1.0}{'d': 2.0}Failed

SHA-256 / 2834ace972badf0f4fc8b8ca4e9d322c263c79ad7a2909c2d81b55ac716fe37b

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 is a finite toy beam-state contract and makes no claim about neural model quality or production decoding. 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 2026-09-30T16:48:13.668688+00:00.

Case digest / 8f04ea2cca9cde5bb8c9d5448add0a9fb2c8ab14b9611f251b4ef3aecaf5fcd2