← Kingy Local AIv22 stopped before the native comparison began
document-comparison-v22-20261018.1 · candidate 1.5.6
Incomplete and not graded. Candidate 1.5.6 remains held; the active measured workflow remains 1.4.0.No new answer-quality result, head-to-head advantage, latency ranking or overall winner is established.
What happened
Kingy's frozen study coordinator mishandled its empty initial state. It treated two absent records as a match, then tried to read a path from the missing pending record. This stopped B1 before any native model startup, prompt input, Duplicate or Send. No native original was attempted. This is a Kingy test-coordinator defect, not evidence of an LM Studio product failure.
Retained operational accounting
Operational metadata reviewed by a separate AI agent| Planned originals | 240 |
|---|
| Completed Kingy originals (A1) | 40 |
|---|
| Completed native LM Studio originals | 0 |
|---|
| Completed originals, total | 40 |
|---|
| Unexecuted planned originals | 200 |
|---|
| A1 execution errors / timeouts | 0 / 0 |
|---|
| A1 answer-generation requests | 104 |
|---|
| A1 document-free warmup requests | 1 |
|---|
| B1 model loads / warmups | 0 / 0 |
|---|
| B1 prompt inputs / Send calls | 0 / 0 |
|---|
| B1 native acceptance attempts | 0 |
|---|
The 40 Kingy originals completed without recorded execution errors or timeouts. Those counts do not establish answer correctness. The 104 answer-generation requests belong to these originals and must not be counted as 104 completed originals. Scientific source inputs remained unchanged.
What remains unrun
- The full 240-original comparison and its answer-quality grading were not completed. Reference values remained closed.
- Numeric quality qualification, scorer evaluation, fresh package selftests, package authentication and denied-egress checks were not run for this study. They are unrun checks, not failed answer-quality or security tests.
- No general accuracy, speed, product reliability, privacy, cost, hardware-upgrade or competitor-superiority claim follows.
Evidence and preservation
Actual root closure, owned generation-model/logger cleanup and the operational review are bound by receipt hashes. A separate AI agent reviewed operational metadata and the coordinator defect on this same Mac. Human validation, reproduction on another machine and semantic answer-quality validation were not performed. Later coordinator repairs are prospective; they do not change this retained outcome or establish a new comparison result.
Original evidence remains preserved locally. These four public files contain a bounded operational projection and hash lineage; raw model answers, reference values, grading keys, native backing files, private UI, source code and private paths are excluded.
Safe operational summary · Receipt and preserved-study hash lineage · Published file hashes
Preserved v19 study · Preserved v20 study · Preserved v21 study. Their evidence and limitations remain separate.