01
Problem and philosophy
Eureka moves upstream from filtering weak originality to constructing inspectable candidates. Scoring ideas alone leaves crucial work implicit: where questions come from, how principles become equations, and how equations motivate mathematical objects and definitions.
Questions, mathematics and physics are intended to influence one another. Counterexamples can expose defects in definitions, while principles help generate equation terms rather than merely constrain a final proposal. The output is a linked candidate package.
02
From input to candidate
The pipeline constructs a contract, question bank and insight ledger, then represents shared assumptions and unexplored axes in a frontier tension graph. Typed operators act on equations, mathematical objects and physics abstractions; scheduling uses Pareto-UCB and MAP-Elites structures.
The Equation Lab generates genomes within an admissible-term grammar, then constructs objects, definitions and conjectures. A domain–math–physics–equation–theorem–idea–refutation bridge tracks transitions. Only harvestable clusters become ideas, which the CLI hands to Aurora.
03
Modules and state
Research questions, physical principles and mathematical objects participate in generation together. Counterexamples should feed definition repair, with the process represented as persistent artifacts.
The D/M/P/E/T/I/R bridge is a provenance map across domain, math, physics, equation, conjecture, idea and refutation. It reveals dependencies when upstream objects change, without guaranteeing external research or a real solver behind every transition.
- Questions and tensions
Socratic research contract and question bank → frontier tension graph → innovation-operator scheduling.
- Equation Lab
The Equation Lab generates and mutates genomes within admissible/forbidden-term grammar, using surrogate traces for diagnostic signals.
- CoMath workspace
A CoMath workspace constructs objects and conjectures, recording proof strategies, counterexample attempts and definition patches.
- Bridge and harvest
A domain–math–physics–equation–theorem–idea–refutation bridge yields harvestable clusters; the CLI subsequently invokes Aurora.
04
Inputs, outputs and contracts
Inputs include the user direction, field and subfield, mode, operator registry and equation grammar. Built-in seed priors and tensions drive the workflow; they are not an externally retrieved, comprehensive map of the field.
EquationGenome records state variables, terms, principles, derivation routes, limits and mutation history. The discovery package retains object, definition, conjecture and bridge IDs, while the CoMath workspace produces a working paper and JSONL records.
- Inputs
User direction, field/subfield, mode configuration, operator and term grammars, and seeded priors/tensions.
- Outputs
discovery_package.json, questions and insights, equation/object/definition/conjecture/bridge IDs, and a CoMath working paper with JSONL records.
05
A conceptual reading scenario
For a reading example about history dependence in dissipative dynamics, start with the memoryless assumption, inspect a memory-bearing genome, then follow its state-space definition and conjecture. This illustrates how a question is organized; surrogate anomalies provide leads for later checks.
06
Changes from the preceding stage
Compared with Aurora’s post-search verification, Eureka adds an artifact-producing discovery layer. Contracts and tensions become workflow inputs, and equations, objects, definitions and counterexamples become explicit nodes and transitions.
At this stage artifact organization precedes complete execution semantics. V5.3 later repairs the discovery-to-Aurora adapters and verdict consistency, while Spark checks whether verification uses the harvested candidate’s actual payloads.
07
What the source establishes
The package contains a 74-operator registry, genome mutation, object workspaces and bridge logic. Its trace function is explicitly a deterministic simulation surrogate: term roles determine diagnostic shapes. These values do not solve arbitrary generated equations.
A registry of 74 typed operators, Pareto-UCB/MAP-Elites scheduling structures, genome mutation, workspace/schema persistence, bridge cycles and idea harvesting.
08
Historical scope and next steps
The equation model is explicitly a deterministic simulation surrogate, not physical PDE/ODE validation. Default tensions and priors are template seeds; the default conjecture path forces none_found and does not establish domain counterexample coverage. The subsequent 5.3 audit identified incomplete discovery-to-Aurora payload and verdict closure.
Reading example · Explaining the workflow
Reading example: trace one candidate
For a reading example about history dependence in dissipative dynamics, start with the memoryless assumption, inspect a memory-bearing genome, then follow its state-space definition and conjecture. This illustrates how a question is organized; surrogate anomalies provide leads for later checks.
- 01
Read the tension and the scope of a memoryless assumption.
- 02
Trace the genome’s memory term to its operator and principle.
- 03
Inspect the definition and conjecture to identify the representational problem.
- 04
Read bridge and harvest reasons, then identify missing external evidence.
Locate the saved candidate, diagnostics and status, and identify evidence that remains absent.
This is a conceptual reading scenario derived from archive structures, not a new run, discovery result or performance experiment.
Changes in this version
Generated artifacts
- Earlier approach
Ideas as primary units
- This version’s approach
Persist equations, objects, definitions and conjectures
Make upstream reasoning traceable.
Physics principles
- Earlier approach
Primarily post-hoc checks
- This version’s approach
Participate in generation and grammar
Make constraints part of construction.
Use of counterexamples
- Earlier approach
Primarily reject candidates
- This version’s approach
Connect to definition repair and bridges
Preserve questions exposed by refutation.