Shareable benchmark table
Local verify arms side by side (never pooled)
Adjacent descriptive results across arms that share fixtures or lineage. Denominators are per-arm; protocols, harness generations, and endpoints differ.
Table 19 · Adjacent arms
Local verify arms side by side (never pooled)
| Arm (agent system) | Harness | Runs | Verified | Median wall | Primary caveat |
|---|---|---|---|---|---|
| GLM-5.3 (ZCode CLI) | v2 fairer harness | 9 | 9/9 | 0m 09s | Operator-role contamination disclosed; n = 1/cell; no token metering |
| Qwen3.8-27B · Q4_K_M (local Ollama) | v1 strict harness | 9 | 0/9 | 6m 48s | 8K-context Q4_K_M local serving, thinking disabled; strict v1 prompt |
| Qwen3.8-27B · Q4_K_M (local Ollama) (v2) | v2 fairer harness | 1 of 54 scheduled | pending | pending | Collection incomplete (2 partial run dirs); quality withheld, never zero |
These are not comparable measurements and must not be read as a leaderboard. The Qwen v2 arm (same harness generation as the GLM arm) is still collecting: its quality cells read "pending" and are never shown as zero. The controlled Grok/GPT campaign measures a different endpoint (reported findings on audits) and is excluded from this table by design.
