Shareable benchmark table
GLM-5.3 (ZCode) verified-proof rate by condition
Verified-proof rate over terminal runs per condition (1 run × 3 tasks) for the GLM-5.3 agent system, with the parent pilot's hierarchical bootstrap 95% CI.
Table 15 · GLM-5.3 verified-proof rate
GLM-5.3 (ZCode) verified-proof rate by condition
| Condition | Agent access | Terminal | Verified | Rate | 95% CI |
|---|---|---|---|---|---|
| Tool-assisted | Bounded search/read + forge build/test feedback, then one submitted proof | 3 | 3/3 | 100% | 100–100% |
| Static (no tools) | Full source context (target + provided test harness); no tool feedback | 3 | 3/3 | 100% | 100–100% |
| Budgeted triage | A strict budget of four inspect operations, then one submission | 3 | 3/3 | 100% | 100–100% |
The interval is degenerate at 9/9 and n is 1 per cell: this is a descriptive result only, with no inferential claim. Condition-level contrasts are confounded by within-session carryover (tools ran before static and triage).
