Find where the assumption breaks.
An 8,760-hour dry-lab replay you can rerun offline.
- Fixtures run
- 154
- Runtime mismatches
- 0
- Public cohorts pinned (S-PUB)
- 0
- Field tests (S-EXT)
- 0
- Engineer sign-off (S-SIGN)
- null
- Composite score
- none
What one run prints
Every run returns the same three things, whatever the inputs.
Break conditions
Set your ownThe smallest stress at which the result crosses your ceiling. For the default plan on SYN-A with a ceiling of 130 USD/MWh:
Stress Base case Breaks at Forced curtailment 5.0% 57.3% Grid price multiplier 1.00 times 2.05 times Interconnection delay 12 months 55 months Gas price 4.00 USD/MMBtu Not applicable: no gas capacity An hourly ledger
8,760 rows of load, renewable supply, battery state, grid import, gas, curtailment and unserved energy, with one hash over the whole year.
How it is hashedA replay kit
Your inputs, the hourly file, the Python kernel and the cohort, zipped in your browser. Run it on a machine with no network and compare the hash.
Offline commands
A dry lab on purpose
The aeolipile of antiquity was never an engine anyone sold. It showed, on a table, that heat becomes work. This group keeps that role. It checks an energy assumption before plant, battery or substation money is committed, and it owns none of those things.
Computed here
- Hourly dispatch, storage and cost in three runtimes
- Closed-form calculators with stated limits
- Receipts that replay on your own machine
- Readings of official decision texts on fixed axes
Not done here
- Operating a power plant or a grid asset
- Wet-lab battery calorimetry
- Hardware-in-the-loop tests on RTDS or OPAL-RT
- Live SCADA or substation control
- Licensed engineering sign-off
| Group | Checks in the dry lab | Leaves out | Catches an error before |
|---|---|---|---|
| Corvel Silicon | Low-bit inference RTL, workload and timing simulation | Foundry tape-out, measured silicon | A tape-out |
| GALENOS | Public cohort classification, physiological models | Wet-lab and animal work, clinical trials | A wet-lab campaign |
| Aeolipile | Hourly energy dispatch, storage, cost and rule reading | Plants, battery labs, substation rigs | A capex, interconnection or PPA signature |
Three runtimes, one answer
The kernel is written three times, in Python, JavaScript and C11. A release ships only when all three return the same fault codes, the same hash over every hourly state and the same raw result bits.
154 fixtures ran: 35 boundary workloads and 119 fault injections, with 0 mismatches. As a control, the C build with fused multiply-add switched on disagrees on 34 of 35 workloads, and the gate reports each one. Read the full ledger.
One hub now, five centres later
Phase 1 keeps everything on this address. A centre moves to its own when its ledger is large enough to need one.
PYRTHEON
Generation and thermodynamics. Cycle limits, output variability and cost recalculation.
Seven calculatorsNODALIS
Grid and power flow. Inertia, transfer limits, contingency and voltage.
Six calculatorsLATENS
Battery, thermal and AI power. The 8,760-hour data-centre model lives here.
The ForgeMENSURA
Rules and source texts. Energy decisions on eight axes, against the official text.
Six seed recordsMETRIED
The evidence gate. Byte anchoring, a Wilson quorum and signed receipts.
Try the gate
Run it without us
The kernel and the verifier need Python 3 and nothing else. These commands work with the network cable out.
curl -O https://aeolipile.risconlabs.com/downloads/aeolipile-kit-0.1.0.tar.gz
tar -xzf aeolipile-kit-0.1.0.tar.gz && cd aeolipile-kit
python3 -I aeo_cli.py fixtures
python3 -I aeo_cli.py run --cohort cohorts/SYN-A.csv
python3 -I metried_verify.py demoThe kit is 271,880 bytes. SHA-256 068b2d47af8bfa3f74ffac629dd27887575fa48c2b079a2b275e030bcab59b29