GeoDict Copilot (CVT engineering) simulates the microstructure; MESSAI publishes the result next to 23k papers of literature values, tagged “simulated”.
Each card shows the structure a pore-scale run would characterise and the five properties MESSAI would receive for it.
| Property | Value · Range |
|---|---|
| Porosity | 0.930.9–0.96 |
| Tortuosity | 1.81.4–2.6 |
| Specific surface | 30,000m²/m³10,000–100,000 |
| Permeability | 3.00e-10m²1.00e-10–1.00e-9 |
| Thickness | 5mm4–6 |
Source: Manufacturer datasheet range (SGL / Sigracell felt series); specific surface is a geometric estimate from fibre diameter; illustrative
| Property | Value · Range |
|---|---|
| Porosity | 0.780.75–0.82 |
| Tortuosity | 1.81.4–2.4 |
| Specific surface | 120,000m²/m³50,000–200,000 |
| Permeability | 5.00e-12m²1.00e-12–1.00e-11 |
| Thickness | 190µm170–210 |
Source: Manufacturer datasheet range (Toray TGP-H series, nominal 190 µm grade); specific surface is a geometric estimate; illustrative
| Property | Value · Range |
|---|---|
| Porosity | 0.80.75–0.85 |
| Tortuosity | 1.51.2–2 |
| Specific surface | 100,000m²/m³50,000–200,000 |
| Permeability | 2.00e-11m²1.00e-11–1.00e-10 |
| Thickness | 350µm300–500 |
Source: Typical woven PAN carbon cloth range (0.3 to 0.5 mm); permeability is an order-of-magnitude estimate between paper and felt; illustrative
| Property | Value · Range |
|---|---|
| Porosity | 0.960.95–0.98 |
| Tortuosity | 1.51.2–2 |
| Specific surface | 16,000m²/m³5,000–50,000 |
| Permeability | 1.00e-8m²1.00e-9–1.00e-7 |
| Thickness | 30mm20–100 |
Source: Typical graphite fibre brush range; specific surface quoted per brush volume; thickness is the brush envelope diameter and varies widely by design; illustrative
| Property | Value · Range |
|---|---|
| Porosity | 0.450.38–0.52 |
| Tortuosity | 1.51.3–2 |
| Specific surface | 1,800m²/m³800–3,600 |
| Permeability | 7.00e-9m²1.00e-9–4.00e-8 |
| Thickness | 50mm20–200 |
Source: Packed bed of 1 to 4 mm granules; specific surface is the geometric 6(1-ε)/d_p (BET micropore area is ~10⁶× larger and not resolved here); permeability from Kozeny-Carman; thickness is bed depth; illustrative
CVT’s Python post-processing writes JSON in MESSAI’s existing CalibrationSet shape: for each parameter a value, a unit, a standard error, the rig it came from, and the source file.
The platform’s applier only lets a file touch the parameters that are allow-listed for its target. Anything else is rejected, not silently dropped.
{
"schemaVersion": "0.1.0",
"id": "00000000-0000-4000-8000-0000000000a1",
"label": "Illustrative current-efficiency calibration (simulated, not a result)",
"createdAt": "2026-10-23T00:00:00.000Z",
"rig": {
"label": "GeoDict FlowDict spacer sweep (illustrative)",
"campaign": "Proposed pilot - no run has been performed",
"operator": "CVT engineering (proposed)",
"notes": "ILLUSTRATIVE. These values are synthetic and were not produced by GeoDict or any rig. They show the shape of the exchange file for a proposed pore-scale spacer simulation of the pilot design: nine current-density points swept through the limiting current, source: simulated."
},
"structure": {
"membraneStackUp": "bipolar:Bipolar membrane (generic)|cation-exchange:Cation-exchange membrane (generic)|anion-exchange:Anion-exchange membrane (generic)",
"permselectivities": "bipolar:0.95|cation-exchange:0.97|anion-exchange:0.95",
"acidChannelGap_m": 0.0005,
"baseChannelGap_m": 0.0005,
"spacerVoidFraction": 0.8,
"spacerShapeFactor": 1.6
},
"designFingerprint": "dd398efa",
"data": {
"target": "current-efficiency",
"points": [
{
"currentDensity_A_per_m2": 1000,
"currentEfficiency": 0.9125
},
{
"currentDensity_A_per_m2": 2500,
"currentEfficiency": 0.8671
},
{
"currentDensity_A_per_m2": 4000,
"currentEfficiency": 0.8194
},
{
"currentDensity_A_per_m2": 5500,
"currentEfficiency": 0.7496
},
{
"currentDensity_A_per_m2": 7000,
"currentEfficiency": 0.6835
},
{
"currentDensity_A_per_m2": 8500,
"currentEfficiency": 0.597
},
{
"currentDensity_A_per_m2": 10000,
"currentEfficiency": 0.603
},
{
"currentDensity_A_per_m2": 12000,
"currentEfficiency": 0.596
},
{
"currentDensity_A_per_m2": 14000,
"currentEfficiency": 0.601
}
]
},
"parameters": [
{
"name": "coIonLeakageCoefficient",
"value": 0.3496142212500261,
"standardError": 0.002010397427043742,
"unit": "-",
"atBound": false,
"lower": 0,
"upper": 0.999
},
{
"name": "limitingCurrentDensity_A_per_m2",
"value": 8499.951629356563,
"standardError": 72.56883799103873,
"unit": "A/m2",
"atBound": false,
"lower": 2800,
"upper": 1400000
}
],
"fit": {
"rSquared": 0.9993256310612163,
"residualStandardDeviation": 0.0035232456000414787,
"chiSquare": 0.00008689281690748147,
"degreesOfFreedom": 7,
"conditionNumber": 25.515331148789418,
"iterations": 41,
"converged": true
},
"residuals": [
0.004000490798639067,
-0.0030110479702989412,
0.0019039263891125202,
-0.004211305949715638,
0.0027724590248771053,
-0.002330495118100906,
0.0036695048818990994,
-0.003330495118100907,
0.0016695048818990976
]
}The Copilot already talks to Claude; if it can call MCP, these are available at /api/mcp.
Trace a literature number back to its paper and extracted snippet.
Compare a simulated value with published distributions, priors and contradictions.
Predict performance for a candidate electrode, or update a prior with new data.
Find similar studies and gaps around a new electrode structure.
Look up microbes and electrode materials by name.
Spacer mesh → flow + mass transfer (GeoDict)
Limiting current density → membrane area (@messai/dac)
Membrane area → cost per tonne
The same pattern applies to direct air capture. The model’s own evidence document attributes 84.5% of commercial cost to bipolar membrane price and life (docs/dac-evidence-gaps.md).
Open the DAC workspace