FA-99026 / Optimizer update state / Open access

Every parameter group is forced to use the slow learning rate · case 01

Every parameter group is forced to use the slow learning rate.

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

ROOT CAUSE

The update substitutes the slow group rate for every record regardless of its group name.

THE FAILURE

The update substitutes the slow group rate for every record regardless of its group name.

Unsuccessful approach: Replacing the global rate with the fast rate fixes fast records but corrupts slow records in a mixed group list.

Case contract

Records [group,p,g] use rates {'slow':1,'fast':3}; return [group,p-rate*g] in input order.

Why this case matters

Optimizer implementations must associate each parameter with its own mathematically specified update state.

1 / The failure

Exit 1

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

N = 1
observations = []
def solve(records):
    return [[r[0],r[1]-r[2]] for r in records]
def check(label, actual, expected):
    observations.append({"check": label, "actual": actual, "expected": expected, "passed": actual == expected})
check('both groups',solve([['slow',10,2],['fast',10,2]]),[['slow',8],['fast',4]])
check('reversed groups',solve([['fast',10,2],['slow',10,2]]),[['fast',4],['slow',8]])
check('single slow',solve([['slow',N,1]]),[['slow',N-1]])
check('single fast',solve([['fast',N,1]]),[['fast',N-3]])
check('empty group list',solve([]),[])
check('negative gradient',solve([['fast',4,-2]]),[['fast',10]])
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
both groups	`[['slow', 8], ['fast', 8]]`	`[['slow', 8], ['fast', 4]]`	Failed
reversed groups	`[['fast', 8], ['slow', 8]]`	`[['fast', 4], ['slow', 8]]`	Failed
single slow	`[['slow', 0]]`	`[['slow', 0]]`	Passed
single fast	`[['fast', 0]]`	`[['fast', -2]]`	Failed
empty group list	`[]`	`[]`	Passed
negative gradient	`[['fast', 6]]`	`[['fast', 10]]`	Failed

SHA-256 / 2e7dd1a2a829010549c4723ede010e7913dc427df9b8c1d132e4bb097f3eb6f8

2 / The unsuccessful fix

Exit 1

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

N = 1
observations = []
def solve(records):
    return [[r[0],r[1]-3*r[2]] for r in records]
def check(label, actual, expected):
    observations.append({"check": label, "actual": actual, "expected": expected, "passed": actual == expected})
check('both groups',solve([['slow',10,2],['fast',10,2]]),[['slow',8],['fast',4]])
check('reversed groups',solve([['fast',10,2],['slow',10,2]]),[['fast',4],['slow',8]])
check('single slow',solve([['slow',N,1]]),[['slow',N-1]])
check('single fast',solve([['fast',N,1]]),[['fast',N-3]])
check('empty group list',solve([]),[])
check('negative gradient',solve([['fast',4,-2]]),[['fast',10]])
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
both groups	`[['slow', 4], ['fast', 4]]`	`[['slow', 8], ['fast', 4]]`	Failed
reversed groups	`[['fast', 4], ['slow', 4]]`	`[['fast', 4], ['slow', 8]]`	Failed
single slow	`[['slow', -2]]`	`[['slow', 0]]`	Failed
single fast	`[['fast', -2]]`	`[['fast', -2]]`	Passed
empty group list	`[]`	`[]`	Passed
negative gradient	`[['fast', 10]]`	`[['fast', 10]]`	Passed

SHA-256 / 0a65e3ee80abc00f6f1c92c509ab20921b07e0faed6e56a2d5e05731189dfb1f

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 deterministic scalar or short-vector toy contracts, not claims about production optimizer APIs or training 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 2026-09-30T16:48:19.455552+00:00.

Case digest / 6617544e5b48b50274b2db1f0b8a4eccf2b7b6e76575e569a61969aba310c7ee