FA-98521 / Beam search bookkeeping / Open access

Beam search: child ancestry points to sorted candidate position rather than source parent · case 01

child ancestry points to sorted candidate position rather than source parent.

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

ROOT CAUSE

After ranking a flattened candidate list, the child stores its position in that list as parent ID.

THE FAILURE

After ranking a flattened candidate list, the child stores its position in that list as parent ID.

Unsuccessful approach: Recovering parent IDs from the first candidate per parent fails when one parent has several candidates.

Case contract

Candidates [parent,score,token]; return selected rows [parent,token] after global top width.

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(candidates, width):
    return [[i,r[2]] for i,r in enumerate(sorted(candidates,key=lambda r:-r[1])[:width])]
def check(label, actual, expected):
    observations.append({"check": label, "actual": actual, "expected": expected, "passed": actual == expected})
c=[[7,N,'a'],[3,N+2,'b'],[7,N+1,'c']]
check('sorted child keeps source',solve(c,2),[[3,'b'],[7,'c']])
check('same source repeated',solve(c,3),[[3,'b'],[7,'c'],[7,'a']])
check('empty set',solve([],N),[])
check('zero width',solve(c,0),[])
check('single child',solve([[9,1,'x']],1),[[9,'x']])
check('tie by parent',solve([[8,N,'z'],[2,N,'a']],2),[[2,'a'],[8,'z']])
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
sorted child keeps source	`[[0, 'b'], [1, 'c']]`	`[[3, 'b'], [7, 'c']]`	Failed
same source repeated	`[[0, 'b'], [1, 'c'], [2, 'a']]`	`[[3, 'b'], [7, 'c'], [7, 'a']]`	Failed
empty set	`[]`	`[]`	Passed
zero width	`[]`	`[]`	Passed
single child	`[[0, 'x']]`	`[[9, 'x']]`	Failed
tie by parent	`[[0, 'z'], [1, 'a']]`	`[[2, 'a'], [8, 'z']]`	Failed

SHA-256 / 5d2996212b517dd3b3e9517f0e436829886cfc0ac4808b0f1f567243ca0fc426

2 / The unsuccessful fix

Exit 1

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

N = 1
observations = []
def solve(candidates, width):
    return [[next(i for i,r in enumerate(candidates) if r[0]==x[0]),x[2]] for x in sorted(candidates,key=lambda r:-r[1])[:width]]
def check(label, actual, expected):
    observations.append({"check": label, "actual": actual, "expected": expected, "passed": actual == expected})
c=[[7,N,'a'],[3,N+2,'b'],[7,N+1,'c']]
check('sorted child keeps source',solve(c,2),[[3,'b'],[7,'c']])
check('same source repeated',solve(c,3),[[3,'b'],[7,'c'],[7,'a']])
check('empty set',solve([],N),[])
check('zero width',solve(c,0),[])
check('single child',solve([[9,1,'x']],1),[[9,'x']])
check('tie by parent',solve([[8,N,'z'],[2,N,'a']],2),[[2,'a'],[8,'z']])
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
sorted child keeps source	`[[1, 'b'], [0, 'c']]`	`[[3, 'b'], [7, 'c']]`	Failed
same source repeated	`[[1, 'b'], [0, 'c'], [0, 'a']]`	`[[3, 'b'], [7, 'c'], [7, 'a']]`	Failed
empty set	`[]`	`[]`	Passed
zero width	`[]`	`[]`	Passed
single child	`[[0, 'x']]`	`[[9, 'x']]`	Failed
tie by parent	`[[0, 'z'], [1, 'a']]`	`[[2, 'a'], [8, 'z']]`	Failed

SHA-256 / e2e028a785eb8d7c7fee3339eab03728d70550053c5a04603d389ee125afc082

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 2026-09-30T16:48:14.105858+00:00.

Case digest / 34ca65497ecec7e6dc4f594355a716e8eed93dd7bd3d8be3644cefb78e402238