Organizes research around challenge responses, preregistered predictions and explanatory obligations, adding world-evidence accounting, learning-progress curiosity, prequential meters, explanation checks and world witnesses.
Philosophy
Evidence should come from testable responses to world tasks, and explanations should help reconstruct the phenomenon they describe.
Architecture & control flow
- 01
Construct candidates and challenge protocols
- 02
Record predictions and world responses with bounded evidence credit
- 03
Check candidate audits and transformational witnesses
- 04
Schedule duels, learning progress and explanation tests
- 05
Separate honest-zero results from evidence-bearing candidates
Architecture outline derived from this version’s control flow.
What this version changes
Begins replacing provenance labels and prose scores with challenge protocols and evidence accounting.
Inputs & outputs
- Inputs
A topic, challenge budget, candidates and world tasks.
- Outputs
Challenge certificates, evidence ledgers, explanation records, per-candidate audits and tiered results.
Implementation & evidence scope
Default lite conservatively yields zero credit. Its full-stack sorting demonstration once used PASS_DISCOVERY, which did not mean scientific discovery; audit-interface imitation is repaired in v6.6.
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
pyproject.toml· 1–5V6_5_FEYNMAN_ARCHITECTURE_R2.md· 1–12harness/discovery_v65/pipeline_v65.py· 17–33harness/discovery_v65/pipeline_v65.py· 60–115harness/discovery_v65/hard_audit/v62_adapter.py· 99–129V6_5_FEYNMAN_CHANGELOG.md· 42–45