2026-10-13 full qualification evidence¶
This is the closed, public-safe artifact set for the 2026-10-13 full qualification campaign. It preserves native evidence shapes and negative results; it is not a deployment record or a promotion authorization.
| Item | Purpose |
|---|---|
| Source registry, core directory, and community addendum | 100 public research and implementation sources, dated and classified as priors rather than local results. |
| Source guide | Curated reading path through the 100-source directory for expert model, runtime, and hardware research. |
| Fixed-five results | All thirteen fixed-denominator quality outcomes with native links. |
| Summary | Derived bounded outcome and decision labels. |
| Workload manifest | Fixed-five grading and capacity-cell contract. |
| Configuration and identity | Public identity boundary and measured hardware scope. |
| Coverage and gaps | Requested contracts, retained coverage, and open gaps. |
| Friction log | Failure dispositions and durable fix-forward work. |
| Restoration and R19 summary | Independently accepted restoration evidence, including separate thinking-off failure and high-thinking client-pass boundaries. |
| Cleanup summary | Eleven exact model-cache snapshots and 106 image IDs removed; byte accounting and retained references are explicit. |
| Reflection | Campaign learning without broadening any result. |
| Baseline chart | Derived matrix; graph data names its native inputs. |
| Native projections | Approved stage-preserving public projections only: R1, R3, R5, R6, R10, R11, R14, R15, R16-r2, R17, R18, R19. |
Excluded staging revisions R12, R13, and R16 are not retained. R18 terminal Ignis/lineage and R19 restoration materials were independently reviewed before inclusion.