01
Stage one: specify the object being checked
V6.0 establishes three genomes, domain contracts and candidate evidence graphs. V6.1 hardens execution modules without wiring the archive's main entry. V6.2 connects execution to the candidate through compilers, routing, benchmarks and relation binding. The original, adapted and Full-stack packages are one EulerForge generation.
02
Stage two: change problems and agendas
V6.3 adds axioms, concept operators, quality-diversity and failure quarantine to a generation layer. V6.4 connects provenance ceilings, executable criticism, paired duels and anomaly-derived questions into revisable programmes. Rich designs coexist with local templates, heuristics and synthetic smoke; they must not be presented as open-science effectiveness.
03
Stage three: credit survives rechecking
V6.5 advances world challenges and event credit, but V6.6 fixes the three-gate imitation, exposed answer digest and fixture promotion. Noether requires true V6.2 gates, prior predictions, controlled settlement and independent prosecution. Demonstrations become PASS_DEMO with zero credit. The correction is part of the historical trajectory, not an error erased from the archive.
04
Stage four: lineage enters an observation loop
V6.7 connects ResearchDelta, causal events, external anchors and lineage, although missing adapters keep many candidates unsettled. V6.8 adds provider protocols, fixed-task execution, diagnosis, counterevidence permissions, difference memory and one-shot confirmation after freezing. Observations enter the actual dataflow, while external-model research gains still require real-task evaluation.
05
Contracts carried across versions
Shared scaffolding includes candidate identity, strongest priors, discriminating consequences, failures and inspectable artifacts. Final authority shifts from package-field gates toward real executors, independent recomputation and host-controlled confirmation. Each minor page has its own flow, conceptual walkthrough, interface fields, before/after changes and source anchors, keeping design distinct from established effects.
06
How to read historical validation
Test definitions, bundled author logs and independent research evaluations are different evidence layers. V6.8's reported 2,026 repository tests, 113 dedicated tests and 17 re-executions validate software and synthetic flows, not real-LLM scientific breakthroughs. Missing release dates remain unstated. This review inspected source, contracts and records only.
How this series evolves
- V6.0
V6.0 · Archimedes
Three candidate types and evidence graphs establish the package.
- V6.1
V6.1 · Dedalus
Hardened modules; the archive's default entry still runs 6.0.
- V6.2
V6.2 · EulerForge
Compilation and real routing; three packages are delivery variants.
- V6.3
V6.3 · Grothendieck
Axioms, concept transformations and quarantine organize generations.
- V6.4
V6.4 · Lakatos
Executable criticism, duels and anomalies generate agendas.
- V6.5
V6.5 · Feynman
World challenges and accounting; later fixes correct demo labels.
- V6.6
V6.6 · Noether
True gates, sealed challenges and recomputation repair credit.
- V6.7
V6.7 · Poincaré
ResearchDelta, events and lineage with limited world coverage.
- V6.8
V6.8 · Poincaré
Provider protocols, observation loops and one-shot confirmation.
Versions in this series
Extends idea generation into mathematics, algorithms and architectures, organizing candidates through evidence graphs, strongest-prior checks and route-specific gates.
Read the full introductionDedalus
Hardens Archimedes with subprocess timeouts, numeric benchmarks, finite-graph execution, candidate-ID binding and required-route checks.
Read the full introductionEulerForge
Turns candidates into compilable objects: equations use registered operator semantics, algorithms face subprocess tests and fixed benchmarks, and architectures produce executable forward paths, all bound to evidence and prior deltas.
Read the full introductionGrothendieck
Adds a generation-side concept-space layer: explicit axiom cards, twelve T operators, diversity archives, quarantined failures, taste forecasts, reality bets and strategy genomes.
Read the full introductionLakatos
Turns criticism into executable counterexamples and discriminating experiments, and organizes candidates as research-programme sequences through provenance, re-instantiation, null models, paired bets, anomalies and question genomes.
Read the full introductionFeynman
Organizes research around challenge responses, preregistered predictions and explanatory obligations, adding world-evidence accounting, learning-progress curiosity, prequential meters, explanation checks and world witnesses.
Read the full introductionNoether
Treats reports as evidence indexes: true audits, answer-free sealed challenges, adjudicator-minted settlements, stable hash identities and prosecutor replay must agree, with demonstrations isolated as PASS_DEMO.
Read the full introductionPoincaré
Replaces one-shot prose with ResearchDelta objects that record questions, representations, assumptions, mechanisms, falsifiers and lineage, organizing multi-generation search through isolated populations, multi-niche archives and independent event replay.
Read the full introductionPoincaré
Upgrades the default path to an observation-driven controller: HTTP model providers propose constrained candidates, the operator locks tasks and baselines, execution yields concrete counterevidence and revision obligations, and evidence authorizes representation changes.
Read the full introductionWhat carries forward
Candidate and evidence identities must align; missing adapters and failures remain visible.
Background, priors, support and refutation serve different roles.
Report demonstrations, synthetic checks, task confirmation and discovery separately.
Sources, budgets, protocols, raw artifacts and verification paths remain reusable engineering assets.
Implementation & evidence across the series
V6 is a historical research-software and design archive. Unverified release dates and performance remain unstated; no historical experiments were run here. The current Innovation Generator is Aletheia V7.3.0.