FA-98441 / Beam search bookkeeping / Open access

Beam search: child score drops its parent history · case 01

child score drops its parent history.

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

ROOT CAUSE

Expansion replaces a hypothesis score with only the newest token contribution.

THE FAILURE

Expansion replaces a hypothesis score with only the newest token contribution.

Unsuccessful approach: Taking the larger of parent and child still loses negative log-probability contributions.

Case contract

Given parent scores and [parent_index, token_delta] expansions, return cumulative child scores in expansion order.

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(parents, expansions):
    return [delta for parent,delta in expansions]
def check(label, actual, expected):
    observations.append({"check": label, "actual": actual, "expected": expected, "passed": actual == expected})
check('two parent histories',solve([-N,-2*N],[[0,-1],[1,-2]]),[-N-1,-2*N-2])
check('negative child contribution',solve([0],[[0,-N]]),[-N])
check('positive toy contribution',solve([N],[[0,2]]),[N+2])
check('no expansions',solve([N],[]),[])
check('repeated parent',solve([-N],[[0,-1],[0,-3]]),[-N-1,-N-3])
check('later parent selected',solve([0,-N],[[1,1]]),[1-N])
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
two parent histories	`[-1, -2]`	`[-2, -4]`	Failed
negative child contribution	`[-1]`	`[-1]`	Passed
positive toy contribution	`[2]`	`[3]`	Failed
no expansions	`[]`	`[]`	Passed
repeated parent	`[-1, -3]`	`[-2, -4]`	Failed
later parent selected	`[1]`	`[0]`	Failed

SHA-256 / 290237311442e3ba25ea23b1269b1c5c9e5c858f579368dbd43d903605eb598d

2 / The unsuccessful fix

Exit 1

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

N = 1
observations = []
def solve(parents, expansions):
    return [max(parents[parent],delta) for parent,delta in expansions]
def check(label, actual, expected):
    observations.append({"check": label, "actual": actual, "expected": expected, "passed": actual == expected})
check('two parent histories',solve([-N,-2*N],[[0,-1],[1,-2]]),[-N-1,-2*N-2])
check('negative child contribution',solve([0],[[0,-N]]),[-N])
check('positive toy contribution',solve([N],[[0,2]]),[N+2])
check('no expansions',solve([N],[]),[])
check('repeated parent',solve([-N],[[0,-1],[0,-3]]),[-N-1,-N-3])
check('later parent selected',solve([0,-N],[[1,1]]),[1-N])
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
two parent histories	`[-1, -2]`	`[-2, -4]`	Failed
negative child contribution	`[0]`	`[-1]`	Failed
positive toy contribution	`[2]`	`[3]`	Failed
no expansions	`[]`	`[]`	Passed
repeated parent	`[-1, -1]`	`[-2, -4]`	Failed
later parent selected	`[1]`	`[0]`	Failed

SHA-256 / 46d8c8475bd0dfe7b3156b12e745cd01229ea4956d537c889b8eb8bd39ced694

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 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 2026-09-30T16:48:13.239133+00:00.

Case digest / 03471caaa696f30ffa7e6a85e7b6c158b3e6b1af5a860c7c9967ea2aabb91d08