FA-98356 / Token sampling filters / Open access

Token sampling: Minimum-p cutoff uses an absolute probability floor · case 01

Minimum-p cutoff uses an absolute probability floor.

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

ROOT CAUSE

The alpha threshold is compared directly with each probability instead of scaling by the peak.

THE FAILURE

The alpha threshold is compared directly with each probability instead of scaling by the peak.

Unsuccessful approach: A strict relative comparison loses a candidate exactly on the configured floor.

Case contract

Keep tokens with probability at least alpha times the maximum probability; empty input returns empty.

Why this case matters

A finite, deterministic token-selection contract isolates one decoding operation from model inference and runtime behavior.

1 / The failure

Exit 1

"""Failure Map reference implementation. Python standard library only."""
import json
import math
N = 1
observations = []
def solve(probabilities, alpha):
    return sorted(t for t,p in probabilities.items() if p>=alpha)
def check(label, actual, expected):
    observations.append({"check": label, "actual": actual, "expected": expected, "passed": actual == expected})
check('relative scale',solve({'a':.8,'b':.4,'c':.1},.5),['a','b'])
check('equality retained',solve({'a':.6,'b':.3},.5),['a','b'])
check('all equal',solve({'b':N,'a':N},1),['a','b'])
check('empty',solve({},.2),[])
check('one item',solve({'x':.25},.9),['x'])
check('zero alpha',solve({'a':.5,'b':.1},0),['a','b'])
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
relative scale	`['a']`	`['a', 'b']`	Failed
equality retained	`['a']`	`['a', 'b']`	Failed
all equal	`['a', 'b']`	`['a', 'b']`	Passed
empty	`[]`	`[]`	Passed
one item	`[]`	`['x']`	Failed
zero alpha	`['a', 'b']`	`['a', 'b']`	Passed

SHA-256 / 7b12cb8a7fa3cd893a70731f1a410c07baaccd89a96b7a7f2bcb56910d49fd0f

2 / The unsuccessful fix

Exit 1

"""Failure Map reference implementation. Python standard library only."""
import json
import math
N = 1
observations = []
def solve(probabilities, alpha):
    peak=max(probabilities.values(),default=0)
    return sorted(t for t,p in probabilities.items() if p>alpha*peak)
def check(label, actual, expected):
    observations.append({"check": label, "actual": actual, "expected": expected, "passed": actual == expected})
check('relative scale',solve({'a':.8,'b':.4,'c':.1},.5),['a','b'])
check('equality retained',solve({'a':.6,'b':.3},.5),['a','b'])
check('all equal',solve({'b':N,'a':N},1),['a','b'])
check('empty',solve({},.2),[])
check('one item',solve({'x':.25},.9),['x'])
check('zero alpha',solve({'a':.5,'b':.1},0),['a','b'])
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
relative scale	`['a']`	`['a', 'b']`	Failed
equality retained	`['a']`	`['a', 'b']`	Failed
all equal	`[]`	`['a', 'b']`	Failed
empty	`[]`	`[]`	Passed
one item	`['x']`	`['x']`	Passed
zero alpha	`['a', 'b']`	`['a', 'b']`	Passed

SHA-256 / 0fe06dd242d7ea562e17f46c851eb8ee85947d56f6fe8ce5804f41e654340afd

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

These are toy finite-vocabulary contracts with supplied scores, histories, and draws; they do not claim production model quality or compatibility. 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 2026-09-30T16:48:11.933164+00:00.

Case digest / a2aa0ac06c22b6de8a8a9305b7f1e91f76b26fcdd42a68ed9fa1fb5a3692d123