01
Problem and philosophy
Spark targets a concrete failure: a verifier may receive a candidate’s text while checking a newly built seed equation, or treat an unknown residual as passing evidence. V5.4 follows candidate IDs to the actual equation, conjecture and abstraction payloads.
The design favors small, explicit witnesses. It identifies which simple system was integrated instead of presenting surrogate traces as general solutions, and compares canonical term signatures rather than equation names.
02
From input to candidate
The entry remains an Eureka discovery package. The CLI indexes equations and conjectures by the idea’s IDs and follows physics-model links. Canonical signatures and Jaccard distances against 22 known equation families expose structural similarity and prior equivalence.
Term roles select one of four probes: linear decay, quadratic gradient flow, projected circle flow or memory decay. PhysicsTester reports energy, constraint, limit and dimension findings for mutation. Unknown residuals remain unresolved and require repair.
03
Modules and state
Resolve the identity of the validated candidate and the presence of numerical evidence before novelty claims. Unknown units or residuals require repair rather than a silent pass.
Findings have a scope: projection residuals are meaningful in the relevant probe, while a probe without constraints cannot establish arbitrary constraint claims. Reading preset systems, selection priority and returned fields separates small witnesses from broader research commitments.
- Actual payloads
The CLI indexes actual equation/conjecture payloads by idea IDs and resolves physical abstractions through physics_model_id.
- Structural priors
Canonical term signatures and Jaccard distance compare against 22 known equation families.
- Small integration probes
Term roles select one of four probes—linear decay, quadratic gradient flow, circle projection or memory decay—to produce residual, energy and stability traces.
- Findings and status
PhysicsTester emits mutation signals from energy, constraint, limit and unit checks; Aurora sends unresolved numerical evidence to repair.
04
Inputs, outputs and contracts
The checked object must be locatable: genomes carry terms and model links, conjectures carry definitions and proof routes, and probe reports identify their method and traces. A missing conjecture fallback is marked with a synthetic missing ID.
Outputs add PhysicsTester findings and mutation signals. They identify failures and suggest structural repairs, such as dissipation signs, constraint projections or memory limits. Each finding must be read within the selected probe’s scope.
- Inputs
Actual genomes, physical abstractions, conjecture/object IDs, the known-equation library and small probe configurations.
- Outputs
Full payload arrays and per-kind JSONL, canonical-prior metrics, residual samples, PhysicsTester reports and mutation signals, with separate research and Aurora contracts.
05
A conceptual reading scenario
For a genome with a projection term, check that the actual payload was retrieved, inspect selection of the circle-flow probe, then read its residuals and repair suggestion. This explains a requested projection repair without establishing arbitrary high-dimensional equation solving.
06
Changes from the preceding stage
Compared with V5.3, Spark connects payload indexing, canonical prior comparison and small numerical witnesses, while tightening abstract-unit and unknown-status handling. Verification increasingly refers to the candidate’s own artifacts.
Forge’s later audit still finds coefficient-unit and harvest-viability gaps. Spark advances diagnostics; preventing failed equations from becoming ideas requires connecting those diagnostics to evolution and harvesting decisions.
07
What the source establishes
The four probes perform actual small integrations, unlike V5.2’s shape surrogate. They select preset dynamics by role, however, rather than interpreting arbitrary genomes. The 22-family index is a built-in comparison library, and prior and counterexample fields retain synthetic components.
Abstract-unit parsing, rejection of unknown residuals, a 22-family prior index, four genuinely numerical small probes and interface audits.
08
Historical scope and next steps
Probes select fixed small systems by term category rather than solve arbitrary genomes. The 5.5 audit identified coefficient-unit and harvest-viability gaps; failed equations could still be packaged as ideas. Synthetic priors and default counterexample fields cannot establish original mathematical theorems.
Reading example · Explaining the workflow
Reading example: trace one candidate
For a genome with a projection term, check that the actual payload was retrieved, inspect selection of the circle-flow probe, then read its residuals and repair suggestion. This explains a requested projection repair without establishing arbitrary high-dimensional equation solving.
- 01
Locate the candidate’s genome and physics-model link.
- 02
Read term roles and confirm the projected-circle probe.
- 03
Inspect residuals, energy and the suggested operator.
- 04
Interpret repair within the probe’s scope and retain unresolved claims.
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
Payload fidelity
- Earlier approach
CLI rebuilds a seed equation
- This version’s approach
Index actual candidate payloads
Reduce separation between candidate and check.
Numerical witnesses
- Earlier approach
Shape-based surrogate traces
- This version’s approach
Four small integration probes
Make actual computation explicit.
Failure feedback
- Earlier approach
Pass/fail-oriented diagnosis
- This version’s approach
Findings include suggested repairs
Prepare inputs for later repair routing.