FA-100736 / Speculative decoding bookkeeping / Open access

Decoding: Per-step acceptance record marks the rejected position as accepted · case 01

Per-step acceptance record marks the rejected position as accepted.

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

ROOT CAUSE

The boundary comparison includes the rejected position.

THE FAILURE

The boundary comparison includes the rejected position.

Unsuccessful approach: Treating rejection index zero as a special falsey value mistakenly marks the first token accepted.

Case contract

For length m and rejection index k, return m booleans true exactly at indices less than k.

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(length, rejected_at):
    return [i <= rejected_at for i in range(length)]
def check(label, actual, expected):
    observations.append({"check": label, "actual": actual, "expected": expected, "passed": actual == expected})
check('middle rejection', solve(4,2),[True,True,False,False])
check('first rejection', solve(3,0),[False,False,False])
check('tail rejection', solve(3,2),[True,True,False])
check('empty sequence', solve(0,0),[])
check('all accepted boundary', solve(N,N),[True]*N)
check('singleton reject', solve(1,0),[False])
check('one accepted', solve(3,1),[True,False,False])
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
middle rejection	`[True, True, True, False]`	`[True, True, False, False]`	Failed
first rejection	`[True, False, False]`	`[False, False, False]`	Failed
tail rejection	`[True, True, True]`	`[True, True, False]`	Failed
empty sequence	`[]`	`[]`	Passed
all accepted boundary	`[True]`	`[True]`	Passed
singleton reject	`[True]`	`[False]`	Failed
one accepted	`[True, True, False]`	`[True, False, False]`	Failed

SHA-256 / 53a3cab6dff3f55068d6fa795faf896a11296e3674913e3b410fd84004053421

2 / The unsuccessful fix

Exit 1

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

N = 1
observations = []
def solve(length, rejected_at):
    return [i < max(1,rejected_at) for i in range(length)]
def check(label, actual, expected):
    observations.append({"check": label, "actual": actual, "expected": expected, "passed": actual == expected})
check('middle rejection', solve(4,2),[True,True,False,False])
check('first rejection', solve(3,0),[False,False,False])
check('tail rejection', solve(3,2),[True,True,False])
check('empty sequence', solve(0,0),[])
check('all accepted boundary', solve(N,N),[True]*N)
check('singleton reject', solve(1,0),[False])
check('one accepted', solve(3,1),[True,False,False])
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
middle rejection	`[True, True, False, False]`	`[True, True, False, False]`	Passed
first rejection	`[True, False, False]`	`[False, False, False]`	Failed
tail rejection	`[True, True, False]`	`[True, True, False]`	Passed
empty sequence	`[]`	`[]`	Passed
all accepted boundary	`[True]`	`[True]`	Passed
singleton reject	`[True]`	`[False]`	Failed
one accepted	`[True, False, False]`	`[True, False, False]`	Passed

SHA-256 / 2d06ad3723d9697fee55fa12e7012b3f11d7b9ace9cc2b127d75d6b0ed97880c

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 2026-09-30T16:48:39.096512+00:00.

Case digest / 2b4f8e3970a8a0bb9f8830800535a218c60dbe7ba3f6ab8f4993f431370d4096