FA-100666 / Speculative decoding bookkeeping / Open access
Decoding: Draft positions without verifier results are counted as covered · case 01
Draft positions without verifier results are counted as covered.
ROOT CAUSE
The entire draft length is counted even when the verifier returned fewer positions.
THE FAILURE
The entire draft length is counted even when the verifier returned fewer positions.
Unsuccessful approach: Subtracting one from verifier length undercounts when the verifier returned every draft position.
Case contract
Return the number of positions present in both the draft and verifier result lists; token values do not affect coverage.
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 len(draft)
def check(label, actual, expected):
observations.append({"check": label, "actual": actual, "expected": expected, "passed": actual == expected})
check('same coverage', solve([1,2],[1,9]),2)
check('short verifier', solve([1,2,3],[1,2]),2)
check('short draft', solve([1],[1,2]),1)
check('empty verifier', solve([1,2],[]),0)
check('both empty', solve([],[]),0)
check('single coverage', solve([N],[N+1]),1)
check('different tokens still covered', solve([1,2],[8,7]),2)
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 |
|---|---|---|---|
| same coverage | 2 | 2 | Passed |
| short verifier | 3 | 2 | Failed |
| short draft | 1 | 1 | Passed |
| empty verifier | 2 | 0 | Failed |
| both empty | 0 | 0 | Passed |
| single coverage | 1 | 1 | Passed |
| different tokens still covered | 2 | 2 | Passed |
SHA-256 / 38f4420ba3ee5b877e8fd2713c866b6abf4c9a3d748f2e3b86c9b58d7d027e2b
2 / The unsuccessful fix
Exit 1"""Failure Map reference implementation. Python standard library only."""
import json
N = 1
observations = []
def solve(draft, verified):
return max(0,len(verified)-1)
def check(label, actual, expected):
observations.append({"check": label, "actual": actual, "expected": expected, "passed": actual == expected})
check('same coverage', solve([1,2],[1,9]),2)
check('short verifier', solve([1,2,3],[1,2]),2)
check('short draft', solve([1],[1,2]),1)
check('empty verifier', solve([1,2],[]),0)
check('both empty', solve([],[]),0)
check('single coverage', solve([N],[N+1]),1)
check('different tokens still covered', solve([1,2],[8,7]),2)
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 |
|---|---|---|---|
| same coverage | 1 | 2 | Failed |
| short verifier | 1 | 2 | Failed |
| short draft | 1 | 1 | Passed |
| empty verifier | 0 | 0 | Passed |
| both empty | 0 | 0 | Passed |
| single coverage | 0 | 1 | Failed |
| different tokens still covered | 1 | 2 | Failed |
SHA-256 / 682ecf82a367cd22cb55237148df922dbba380998d5e05823cb3239339f00c8f
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 2026-09-30T16:48:38.099320+00:00.
Case digest / 16c19c3c178c748dd076732cebb85c5cc6835fe94f18c1d519cf73409839b875