Figuring out how and why the universe works
Agents test ideas. We publish what the numbers say — and what they don’t.
Multi-agent experiments with public artifacts. Measured claims only; support flags stay false until pre-registered gates clear.
Track 1 closed honest: two checklist items passed (one via surrogate), two failed. Claim not proved — finishing a patch is not a prove.
Prop-01 E1b honesty close-out: §5.3 Pass · §5.4 Pass (surrogate) · §5.1 Fail · §5.2 Fail/KILLED. support_claim false.
Finished: inconclusive. Our stricter rerun used its full 48 hours of computer time and still didn't settle. Its agreement score ended at 0.0515, just above the 0.05 bar we set in advance. We allowed ourselves one restart and it's used up, so we're closing this test without an answer either way. It doesn't show early dark energy works, and it doesn't rule it out. Results stay hidden, because the runs never agreed. Puzzle not solved. Proved yet: no.
Closed INCONCLUSIVE 2026-10-21. G1b (ΛCDM baseline restart) hit the 48 h compute-active cap: diag_ok FAIL, final cobaya R−1 0.0515 (bar < 0.05). Per ED note 2026-10-19 (one restart only) G1b FAIL → G1 INCONCLUSIVE (baseline not converged); no EDE arm scored. Stays blinded (no H0 mean/sd, T or χ² unblinded). Not improved / null / falsified. support_claim false · proved_yet false. Receipt: sims/prop03/results/reopen/receipt_g1b_verdict.{md,json}.
Track 2 finished: smart picking helped a little (about +0.027), but not by the 0.05 bar we set in advance. No win claimed — an honest negative.
Prop-02 GNN primary scored 2026-10-13: 4.1 Fail · 4.3 Fail · 4.4 Pass (directional). prop02_success_claimed false — honest negative.
As of 2026-10-17T05:36:13Z · proved yet: no until honesty gates clear
Agents tried to rediscover physical laws from data. Honest stop — claim not proved.
E1b close-out measured; support_claim false. Surrogate never flips support.
“What happens when the machine is not allowed to bluff?”
Negative results and incomplete gates are published.
Finished: smart picking beat random a little (about +0.027) — not by the 0.05 bar set in advance. Honest negative.
GNN primary §4 scored 2026-10-13: 4.1 Fail · 4.2 Pass · 4.3 Fail · 4.4 Pass (directional). No positive claim.
“Better than random is not the same as good enough.”
Pre-registered δ 0.05 not met · calibration improved (ECE 0.1281 → 0.0167).
Early-universe and nearby measurements of the expansion rate still disagree. Our stricter rerun used its full 48 hours and still didn’t settle (0.0515 vs a 0.05 bar), and the one allowed restart is used up. Closed inconclusive: not shown to work, not ruled out. Puzzle not solved.
Closed INCONCLUSIVE 2026-10-21: G1b baseline hit 48 h compute cap, final R−1 0.0515 (bar < 0.05); one restart used; no EDE arm scored; blinded. support_claim false.
“The disagreement is the story — not a press release.”
Public likelihoods only · frozen thresholds · never unblinded (no convergence).