research-document
ECR-000003 Hypothesis Impact Matrix
ECR-000003 Hypothesis Impact Matrix
| Hypothesis | Tested By | Expected Evidence | Kill / Weakening Signal | Notes |
|---|---|---|---|---|
| H016 Structural Recognition | EXP-001 graph perturbation variants | Recognition persists across topology-preserving perturbations | Graph-only topology variants consistently reduce recognition while preserving structure | Primary metric: Recognition Persistence across graph perturbations |
| H017 Procedural Compression | EXP-002 narrative, bullets, graph, and transition table | Different representations compress into stable recovered structure | Representations produce materially unrelated recovered structures | Use Recognition Persistence alongside structural stability |
| H018 Procedural Isomorphism | EXP-003 matched control-flow across different domains | Different domains yield structurally similar extractions | Isomorphic procedures produce unrelated structural outputs | Status remains proposed |
| H013 Recognition Bias | EXP-001 and EXP-002 recognition behavior | Recognition varies with representation while structure may remain stable | Recognition remains entangled with extraction across every variant | Recognition Persistence helps distinguish lexical recognition from topology-based recognition |
| H003 Multi-Model Convergence | Cross-provider comparison in all experiments | Provider agreement remains meaningful across variants | Cross-model agreement collapses under representation shifts | Reliability observation |
| H015 Measurement Instrument Reliability | Shared schema usage across all packets | Valid JSON and comparable multi-layer outputs across the series | Schema or packet format produces routine failure | Instrument observation |
| H002 Representation Independence | EXP-002 and EXP-003 layer comparison | Structural and constraint layers remain more stable than surface form | Surface form or domain semantics dominates all recovered layers | Representation-level comparison |