FA-100681 / Speculative decoding bookkeeping / Open access

Decoding: Proposal contains one token beyond the verifier window · case 01

Proposal contains one token beyond the verifier window.

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

ROOT CAUSE

The configured window is applied as an inclusive last index.

THE FAILURE

The configured window is applied as an inclusive last index.

Unsuccessful approach: A fixed small cap ignores larger configured windows.

Case contract

Clamp window to at least zero, then return at most that many draft tokens from the beginning.

Why this case matters

A self-contained bookkeeping model makes prefix and token-count semantics observable without a model runtime.

1 / The failure

Exit 1

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

N = 1
observations = []
def solve(draft, window):
    return draft[:window+1]
def check(label, actual, expected):
    observations.append({"check": label, "actual": actual, "expected": expected, "passed": actual == expected})
check('truncate', solve([0,1,2,3,4],N),list(range(N)))
check('zero window', solve([1],0),[])
check('window exceeds', solve([1,2],5),[1,2])
check('empty proposal', solve([],N),[])
check('single slot', solve([N,8],1),[N])
check('preserve order', solve([4,3,2],2),[4,3])
check('negative window', solve([1,2],-1),[])
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
truncate	`[0, 1]`	`[0]`	Failed
zero window	`[1]`	`[]`	Failed
window exceeds	`[1, 2]`	`[1, 2]`	Passed
empty proposal	`[]`	`[]`	Passed
single slot	`[1, 8]`	`[1]`	Failed
preserve order	`[4, 3, 2]`	`[4, 3]`	Failed
negative window	`[]`	`[]`	Passed

SHA-256 / 2dd55b81b8c6356e7b7c8b97fc5511148ef556745db57e668f4813ec98a04f58

2 / The unsuccessful fix

Exit 1

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

N = 1
observations = []
def solve(draft, window):
    return draft[:min(window,2)]
def check(label, actual, expected):
    observations.append({"check": label, "actual": actual, "expected": expected, "passed": actual == expected})
check('truncate', solve([0,1,2,3,4],N),list(range(N)))
check('zero window', solve([1],0),[])
check('window exceeds', solve([1,2],5),[1,2])
check('empty proposal', solve([],N),[])
check('single slot', solve([N,8],1),[N])
check('preserve order', solve([4,3,2],2),[4,3])
check('negative window', solve([1,2],-1),[])
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
truncate	`[0]`	`[0]`	Passed
zero window	`[]`	`[]`	Passed
window exceeds	`[1, 2]`	`[1, 2]`	Passed
empty proposal	`[]`	`[]`	Passed
single slot	`[1]`	`[1]`	Passed
preserve order	`[4, 3]`	`[4, 3]`	Passed
negative window	`[1]`	`[]`	Failed

SHA-256 / 263eb3e27ea9600e8f67dd139461d5eea24ac2bd8a01a121644b42a50d4c5fd7

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

These are finite toy sequences with explicit contracts; they make no claim about production decoder behavior or model quality. 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 2026-09-30T16:48:38.567763+00:00.

Case digest / 9576dcc48a64c6a97187806c3848a9fa66ef8d6c07932a4e0d7651d89291a8d1