Calibration, not hindsight.

We are testing whether Claim–Reality Firewalls can improve strategic route selection before outcomes arrive — across startups, OSS projects, research directions, and product bets.

arXiv preprint → blind A/B/C/D → 30/90/180-day outcomes
How it works

CRF separates generated claims from the final verdict

Decision context
founder · repo · paper · roadmap
LLM routes + claims
generate → decompose → lock
FIREWALL
no model-authored verdict
Reality atoms + code
evidence · risk · cost-to-learn
Calibrated ranking
A/B/C/D · outcomes
CRF frozen score
AUC 0.756
Top-Q lift 2.1
Blind LLM K=5
AUC 0.678
Top-Q lift 1.4
Simple heuristic
AUC 0.700
Top-Q lift 2.1
Name-exposed LLM
AUC 0.761
leakage contrast

Claim-Reality Firewall is an evaluation protocol for strategic decisions with delayed ground truth. Dynamic Resonance is the first implementation in venture route selection. We are now organizing domain replications in open-source adoption, research direction selection, and product roadmap prioritization.

Pilot only, not proof. Current backtest results are encouraging enough to justify a cleaner open experiment. This page should not claim startup-success prediction, investment advice, or validated superiority.
Experiment design

Original vs system-selected Top‑1.

Full protocol

Primary live question

When presentation format is normalized and review is blind, does the system-selected Top‑1 route beat the founder Original route in investor fundability ranking?

Founder OriginalDR Route #1DR Route #2DR Route #3A/B/C/D randomized

Delayed-ground-truth question

Do frozen CRF forecasts and rankings calibrate better than founder self-forecast, investor median, naive LLM judge, and simple heuristic baselines as outcomes arrive?

Brier scoreAUCECETop-bucket lift
Join by role

Four ways to participate.

F

Founder

Benchmark your current startup route against three alternative fundable routes with blind investor review.

Submit route
I

Investor reviewer

Review anonymized route packets, score fundability and confidence, and receive a private calibration report.

Join review pool
T

Track lead

Bring a dataset, domain expertise, or design partners for OSS, R&D, or product replications.

Lead a track
C

Community partner

Host a cohort for founders, developers, or researchers and receive a co-branded aggregate report.

Partner with us
Founder offer
Before you spend two years on one path, test whether there is a sharper route.

Founder intake