AutoResearch/Innovation Generator/V6/Archimedes / DiscoForge

VERSION / V6.0

Archimedes / DiscoForge

Archimedes / DiscoForge

Extends idea generation into mathematics, algorithms and architectures, organizing candidates through evidence graphs, strongest-prior checks and route-specific gates.

Historical version · Design & implementation

Current: Daedalus v7.4.0

V6.0 / ARCHITECTURE

How does an exploration move forward?

01方向与领域Direction and domain02三类候选Three candidate kinds03证据与先例Evidence and priors04类型门控Route-specific gates05审计后存档Audit, then archive01方向与领域Direction and domain02三类候选Three candidate kinds03证据与先例Evidence and priors04类型门控Route-specific gates05审计后存档Audit, then archive
Architecture illustration derived from this version’s actual control flow.
  1. 01

    Direction and domain

    Stop on unresolved domains

  2. 02

    Three candidate kinds

    Equation / algorithm / architecture

  3. 03

    Evidence and priors

    Retrieve and bind per candidate

  4. 04

    Route-specific gates

    Aurora evaluates each route

  5. 05

    Audit, then archive

    Align verdict and persisted record

01

From an idea to a research object

Archimedes / DiscoForge addresses a mismatch between prose-based ideation and research that needs equations, algorithms or computational architectures. V6.0 places these routes in one package and makes the candidate, domain contract, strongest prior and evidence graph addressable objects. An unresolved domain stops the run rather than being silently replaced.

02

Structure before scoring

Its organizing principle is to specify the object before assessing it. Equations carry a state space, operators, units, invariants and limits; algorithms carry code and baselines; architectures carry component graphs and recurrence settings. They share provenance and prior checks, but require different grounds for success. The DiscoForge meta-rule demonstration does not substitute for candidate-specific evidence.

03

Three routes, one package

The pipeline resolves the domain and requirements, constructs route-specific candidates, then plans retrieval and binds evidence and priors for each candidate. Architectures can enter a recurrence-depth sweep. Aurora aggregates route-specific gates. The final package is audited before persistence so the displayed verdict and archived verdict agree.

04

Candidate and evidence contracts

The Typed Equation DSL distinguishes TypeExpr, UnitSignature, OperatorExpr, Invariant and LimitCase instead of storing an equation as an opaque string. EvidenceGraph connects candidate identities, claims and source nodes; strongest-prior and equivalence records expose possible repackaging. These contracts make checks inspectable, although populated fields are not evidence of external experimental validation.

05

Inputs and deliverables

Inputs are the research direction, mode budget and a replaceable AnySearch client; the default client uses offline fixtures. The discovery package includes domain and requirements, candidate genomes, evidence graphs, prior relations, Aurora verdicts, configuration and audit. An unresolved domain still receives a minimal record explaining why no candidate was produced.

06

Conceptual walkthrough: graph dynamics

For graph dynamics, the system can organize a mathematical candidate with a time derivative, graph-related operators and limits, then attach the existing dynamics it should be compared against. Candidate IDs make the evidence and gates traceable. This illustrates the workflow: the default builder's success=True and energy-decay values are synthetic fields, not a record of a new numerical discovery.

07

The transition from V5

Building on V5.8's verification approach, V6.0 broadens the research object into three genome types and adds a shared domain contract, evidence graph and prior relations. It supplies interfaces for later execution hardening while exposing the gap between a well-formed object and actually executing that object. V6.1 turns this gap into concrete module checks.

08

What is implemented

Inspected code includes domain blocking, per-candidate evidence construction, route-specific gates, package auditing and persistence. Test definitions check unresolved-domain blocking, package existence and disk/CLI verdict consistency. Architecture smoke reports and depth scores are still supplied by the pipeline. This account inspected code and tests without rerunning the historical software.

09

Evidence scope

This generation demonstrates research-software organization: what was proposed, which sources it uses and which checks it requires. Offline evidence, fixed candidates and synthetic execution fields do not establish new theorems, algorithmic improvements or architectural gains. Its historical value is making the next engineering gaps explicit, rather than guaranteeing novelty for every gated idea.

Reading example · Explaining the workflow

Reading an Archimedes package

Inputs

An identifiable graph-dynamics research direction.

  1. 01

    The domain contract fixes the recognized scope.

  2. 02

    Requirement classification selects the mathematical route.

  3. 03

    The candidate preserves operators, invariants and limits.

  4. 04

    Its identity connects evidence, priors and gates.

  5. 05

    Audit precedes persistence of the authoritative verdict.

Output

A structured record of the candidate, sources and checks.

Conceptual example; not executed here, and synthetic execution fields do not count as discovery.

Changes in this version

Research object

Earlier approach

Primarily verified ideas

This version’s approach

Equation, algorithm and architecture genomes

Specify what each route inspects

Evidence organization

Earlier approach

Verification concepts distributed across the flow

This version’s approach

Candidate-bound evidence graph and prior relations

Enable claim-level traceability

Persistence

Earlier approach

Displayed and archived results can diverge

This version’s approach

Audit before writing one authoritative package

Persist the actual verdict