FA-98496 / Beam search bookkeeping / Open access

Beam search: search stops before a live path can beat the finished set · case 01

search stops before a live path can beat the finished set.

Verified by executionVariant 1 · 7 checks per implementationDownload source bundle ↓JSON ↗

ROOT CAUSE

Stopping compares current live scores directly with finished scores despite nonpositive future deltas.

THE FAILURE

Stopping compares current live scores directly with finished scores despite nonpositive future deltas.

Unsuccessful approach: Comparing against the worst finished score stops while the kth result is still improvable.

Case contract

Stop only if at least k finished scores exist and the best live score is no greater than the kth-best finished score; future deltas are nonpositive.

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(live_scores, finished_scores, k):
    return bool(finished_scores) and max(live_scores,default=float("-inf"))<=max(finished_scores)
def check(label, actual, expected):
    observations.append({"check": label, "actual": actual, "expected": expected, "passed": actual == expected})
check('one result can stop',solve([-N],[0],1),True)
check('live still better',solve([1],[0],1),False)
check('need k results',solve([-N],[0],2),False)
check('kth threshold',solve([-N],[3,1],2),True)
check('below kth threshold',solve([2],[3,1],2),False)
check('first result still improvable',solve([2],[3,1],1),True)
check('no live paths',solve([],[3],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
one result can stop	`True`	`True`	Passed
live still better	`False`	`False`	Passed
need k results	`True`	`False`	Failed
kth threshold	`True`	`True`	Passed
below kth threshold	`True`	`False`	Failed
first result still improvable	`True`	`True`	Passed
no live paths	`True`	`True`	Passed

SHA-256 / d52f28465d9503389472b215a2e7180248abfb9b6f90df50911a318e07b7a0c7

2 / The unsuccessful fix

Exit 1

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

N = 1
observations = []
def solve(live_scores, finished_scores, k):
    return len(finished_scores)>=k and max(live_scores,default=float("-inf"))<=min(finished_scores)
def check(label, actual, expected):
    observations.append({"check": label, "actual": actual, "expected": expected, "passed": actual == expected})
check('one result can stop',solve([-N],[0],1),True)
check('live still better',solve([1],[0],1),False)
check('need k results',solve([-N],[0],2),False)
check('kth threshold',solve([-N],[3,1],2),True)
check('below kth threshold',solve([2],[3,1],2),False)
check('first result still improvable',solve([2],[3,1],1),True)
check('no live paths',solve([],[3],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
one result can stop	`True`	`True`	Passed
live still better	`False`	`False`	Passed
need k results	`False`	`False`	Passed
kth threshold	`True`	`True`	Passed
below kth threshold	`False`	`False`	Passed
first result still improvable	`False`	`True`	Failed
no live paths	`True`	`True`	Passed

SHA-256 / b012378c59571446a76af47a647814913d60e9b6c4c07a5dd17c14bc8ec62f20

HELD IN THE MEMBER ARCHIVE

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

This mechanism has 7 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 2026-09-30T16:48:13.668747+00:00.

Case digest / 2f52c5fadff7d5c4c69354ee8b7fc64cfd3583d3ebefc283e1b38cdf1dbb8b9c