FA-100646 / Speculative decoding bookkeeping / Open access
Decoding: Rejection location is reported as a one-based position · case 01
Rejection location is reported as a one-based position.
ROOT CAUSE
The first differing index is converted to a one-based position.
THE FAILURE
The first differing index is converted to a one-based position.
Unsuccessful approach: Returning None for complete agreement loses the useful append position.
Case contract
Return the zero-based index of the first differing token, or the number of compared tokens when all compared pairs match.
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, verified):
return next((i+1 for i,(a,b) in enumerate(zip(draft,verified)) if a!=b), min(len(draft),len(verified)))
def check(label, actual, expected):
observations.append({"check": label, "actual": actual, "expected": expected, "passed": actual == expected})
check('first index', solve([8],[7]),0)
check('later index', solve([1,2,3],[1,2,9]),2)
check('all equal', solve([1,2],[1,2]),2)
check('empty lists', solve([],[]),0)
check('verified ends', solve([1,2],[1]),1)
check('draft ends', solve([1],[1,2]),1)
check('singleton equal', solve([N],[N]),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 |
|---|---|---|---|
| first index | 1 | 0 | Failed |
| later index | 3 | 2 | Failed |
| all equal | 2 | 2 | Passed |
| empty lists | 0 | 0 | Passed |
| verified ends | 1 | 1 | Passed |
| draft ends | 1 | 1 | Passed |
| singleton equal | 1 | 1 | Passed |
SHA-256 / 45ac75ed0da5e207326cafc664508518e4bf4dea7bb47b1b06ab953f18a9764b
2 / The unsuccessful fix
Exit 1"""Failure Map reference implementation. Python standard library only."""
import json
N = 1
observations = []
def solve(draft, verified):
return next((i for i,(a,b) in enumerate(zip(draft,verified)) if a!=b), None)
def check(label, actual, expected):
observations.append({"check": label, "actual": actual, "expected": expected, "passed": actual == expected})
check('first index', solve([8],[7]),0)
check('later index', solve([1,2,3],[1,2,9]),2)
check('all equal', solve([1,2],[1,2]),2)
check('empty lists', solve([],[]),0)
check('verified ends', solve([1,2],[1]),1)
check('draft ends', solve([1],[1,2]),1)
check('singleton equal', solve([N],[N]),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 |
|---|---|---|---|
| first index | 0 | 0 | Passed |
| later index | 2 | 2 | Passed |
| all equal | None | 2 | Failed |
| empty lists | None | 0 | Failed |
| verified ends | None | 1 | Failed |
| draft ends | None | 1 | Failed |
| singleton equal | None | 1 | Failed |
SHA-256 / 44484e93e4f378cd326f692f3038fb97c307329df28c4b2e5cb2d21f13ff9a24
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.
Sign in to the archive ↗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 2026-09-30T16:48:38.099146+00:00.
Case digest / 09c49f0b85214f51bc2f012b25ead0d4a6b43e89b840b0fcfbf32e7ea01891d0