Adds compatibility, execution and quality constraints to Origin: prune mappings, select diverse elites, execute supported AST terms and route proof strategies by conjecture type.
Philosophy
More combinations do not equal better discovery. Remove placeholder properties, identity mappings and non-executable combinations before numerical and proof-route checks.
Architecture & control flow
- 01
Resolver 3.0 fixes zero-hit and substring errors; a grounded, anti-template-filtered D-Tree proposer interface is added.
- 02
Compatibility pruning precedes mapping construction, and a quality referee checks nontriviality and executable/theorem artifacts.
- 03
MAP-Elites and bucket caps limit repeated shifts, AST signatures and domain operators; prior deltas use layered path traces.
- 04
Term dispatch executes supported RHS contributions, and CoMath routes nine conjecture types to distinct strategies.
- 05
Seven Aurora-V57 gates consume candidate evidence, with cross-run path memory and a persisted blind-review queue.
Architecture outline derived from this version’s control flow.
Inputs & outputs
- Inputs
Origin paths/mappings, direction, compatibility rules, supported term evaluators, conjecture strategies and prior traces.
- Outputs
Quality reports, elites, AST/simulation reports, type-specific proof graphs and kill tests, expert-review cards/queue, Aurora results and cross-run memory.
Implemented components
- Compatibility pruning, spurious-mapping detection, diversity selection, supported AST RHS execution, conjecture-sensitive proof routes, seven gates and a persisted review queue.
Implementation & evidence scope
The LLM-named proposer is still a deterministic stub, and review cards/queues do not imply human endorsement. The 5.8 audit finds reused v5.6 ideas, misaligned evidence links and possible no-op replay of dictionary ASTs; candidate chains are not fully closed. Finite term solvers and proof routes are exploration prototypes, not general proof or open-world discovery systems.
Code & bundled material
The introduction draws on bundled notes, changelogs and central code. Software tests, synthetic diagnostics and scientific effectiveness use different evidence standards.
Source references
novelty-idea-generator/V5_7_GENESIS_CHANGELOG.md· 62–128novelty-idea-generator/V5_7_GENESIS_CHANGELOG.md· 161–246novelty-idea-generator/harness/discovery_v57/d_tree_llm/llm_dtree_proposer.py· 1–44novelty-idea-generator/harness/discovery_v57/equation_execution/term_dispatch.py· 23–150novelty-idea-generator/harness/discovery_v57/pipeline_v57.py· 367–434novelty-idea-generator/V5_8_LOGOS_CHANGELOG.md· 6–17