research-package RP-2026-07-30-NHE-LINEAGE
Non-Human Experimental Lineage and Dependency Map
Non-Human Experimental Lineage and Dependency Map
flowchart LR
E1["EX-E001 taxonomy mapping<br/>design evidence"] -->|selected pilots| E2["EX-E002 ET-004 designer pilot"]
E1 -->|selected pilots| E3["EX-E003 ET-014 designer pilot"]
E2 -->|same four probes| E4["EX-E004 deterministic calibration"]
E3 -->|same three probes| E4
E2 -->|same fixture| E7["EX-E007 export preflight"]
E3 -->|same fixture| E7
E2 -. independent review missing .-> E5["EX-E005 repeated agent baseline<br/>blocked"]
E3 -. independent review missing .-> E5
E7 -->|isolation prerequisite| E5
E6["EX-E006 contamination comparison<br/>planned"] -. may revise isolation .-> E5
Relationship classifications
| Parent | Descendant | Relationship | Shared dependencies | Independence judgment |
|---|---|---|---|---|
| EX-E001 | EX-E002 | Follow-up, task-family instantiation | Same repository interpretation and researcher | Not an independent replication |
| EX-E001 | EX-E003 | Follow-up, task-family instantiation | Same repository interpretation and researcher | Not an independent replication |
| EX-E002 | EX-E004 | Shared-data and shared-evaluator dependency | Four probes, expected labels, task, researcher | Calibration reuses evidence; no new independent outcome sample |
| EX-E003 | EX-E004 | Shared-data and shared-evaluator dependency | Three probes, expected labels, task, researcher | Calibration reuses evidence; no new independent outcome sample |
| EX-E002 | EX-E007 | Extension/preflight | Same ET-004 source fixture and exporter | New property measured, but not independent support for task validity |
| EX-E003 | EX-E007 | Extension/preflight | Same ET-014 source fixture and exporter | New property measured, but not independent support for task validity |
| EX-E002/003/007 | EX-E005 | Prerequisites | Tasks and exporter will be reused | Future runs can add independent stochastic attempts, not independent task sampling |
Replication matrix
| Original experiment | Replication | Independence | Result | Replication strength |
|---|---|---|---|---|
| EX-E002 | None | — | — | None |
| EX-E003 | None | — | — | — |
| EX-E004 | 2026-07-30 local rerun of evalctl.py all |
Same code, same probes, same labels, later environment | 7/7 again | Regression reproduction only; weak scientific replication |
| EX-E007 | No retained independent repeat | — | — | None |
There are zero direct, conceptual, or independent replications of a scientific finding. Different probe filenames are not independent experiments. The effective independent sample for grader generalization is zero because all seven cases were authored alongside the evaluator.
Inflated-count warning
The defensible counts are:
- two distinct task contracts;
- seven purposive probe artifacts;
- one deterministic classification mechanism;
- one recorded calibration execution, plus a same-artifact reproducibility run;
- two export instances through one exporter.
“Seven successful experiments,” “four replications,” or “two validated tasks” would each be false.