Supernova V2

Supernova V2 is a falsifiable experiment about verified mathematical composition.

Goal 1

Test whether a solver that builds a within-problem chain of independently verified intermediate products solves more problems than four controls under the same frozen complete-cost budget:

1. ordinary solving;
2. portfolio solving;
3. product-only solving;
4. multi-fidelity solving;
5. verified-chain solving (the candidate mechanism).

GitHub is the shared mailbox and CI runner; it is not scientific authority. The active Goal-1 authority index is goal1/GOAL1.json, and its exclusive confirmatory scientific rules are content-bound in goal1/CONFIRMATORY_PROTOCOL.json. Confirmatory dispatch remains blocked until the independently verifiable execution authority and manifest named there exist. The hourly work board is orchestration/BOARD.json, and the deterministic evaluator lives in src/supernova_goal1.

Local check

python -m unittest discover -s tests -p "test_*.py"
python scripts/evaluate_goal1.py examples/dry_run_goal1.json examples/dry_run_records.json

The example uses examples/dry_run_goal1.json; it is a dry run of the data path and has no scientific credit.