FA-98501 / Beam search bookkeeping / Open access

Beam search: completed pool is capped by discovery order · case 01

completed pool is capped by discovery order.

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

ROOT CAUSE

Finished hypotheses are truncated before they are ranked by final score.

THE FAILURE

Finished hypotheses are truncated before they are ranked by final score.

Unsuccessful approach: Sorting ascending retains the weakest completions first.

Case contract

Rows [id,score]; keep up to capacity highest scores, lexical ID for ties.

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, capacity):
    return rows[:capacity]
def check(label, actual, expected):
    observations.append({"check": label, "actual": actual, "expected": expected, "passed": actual == expected})
rows=[['early',N],['best',N+3],['mid',N+1]]
check('capacity winner',solve(rows,1),[['best',N+3]])
check('top two',solve(rows,2),[['best',N+3],['mid',N+1]])
check('zero capacity',solve(rows,0),[])
check('empty pool',solve([],N),[])
check('tie lexical',solve([['z',N],['a',N]],1),[['a',N]])
check('all retained sorted',solve(rows,5),[['best',N+3],['mid',N+1],['early',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
capacity winner	`[['early', 1]]`	`[['best', 4]]`	Failed
top two	`[['early', 1], ['best', 4]]`	`[['best', 4], ['mid', 2]]`	Failed
zero capacity	`[]`	`[]`	Passed
empty pool	`[]`	`[]`	Passed
tie lexical	`[['z', 1]]`	`[['a', 1]]`	Failed
all retained sorted	`[['early', 1], ['best', 4], ['mid', 2]]`	`[['best', 4], ['mid', 2], ['early', 1]]`	Failed

SHA-256 / 6193e35316321e9f3ff237966a69cbb9fbda046b4d502980e8ca159d5127060c

2 / The unsuccessful fix

Exit 1

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

N = 1
observations = []
def solve(rows, capacity):
    return sorted(rows,key=lambda r:(r[1],r[0]))[:capacity]
def check(label, actual, expected):
    observations.append({"check": label, "actual": actual, "expected": expected, "passed": actual == expected})
rows=[['early',N],['best',N+3],['mid',N+1]]
check('capacity winner',solve(rows,1),[['best',N+3]])
check('top two',solve(rows,2),[['best',N+3],['mid',N+1]])
check('zero capacity',solve(rows,0),[])
check('empty pool',solve([],N),[])
check('tie lexical',solve([['z',N],['a',N]],1),[['a',N]])
check('all retained sorted',solve(rows,5),[['best',N+3],['mid',N+1],['early',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
capacity winner	`[['early', 1]]`	`[['best', 4]]`	Failed
top two	`[['early', 1], ['mid', 2]]`	`[['best', 4], ['mid', 2]]`	Failed
zero capacity	`[]`	`[]`	Passed
empty pool	`[]`	`[]`	Passed
tie lexical	`[['a', 1]]`	`[['a', 1]]`	Passed
all retained sorted	`[['early', 1], ['mid', 2], ['best', 4]]`	`[['best', 4], ['mid', 2], ['early', 1]]`	Failed

SHA-256 / 7dff26ce3367b35a887a16e30dfdfe901950588d9392f74f99afc7cc0431e057

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 2026-09-30T16:48:13.669345+00:00.

Case digest / 3e58ec7a8a97918927ffb7f2ea4f6ad25b1ca211e4475a9b3adb5041c40a9f39