Ingest · normalize · tier
Stage 01
Arc Signal Transformer
Raw industrial signal ingested, normalized, timestamped, and tiered before any model touches it.
MESO-1 · Foundation model · Industrial forecasting
The inference engine beneath every Forge decision.
MESO-1 is Palisade's in-house foundation model for industrial forecasting. It ingests structured signal from Arc, fires continuous probabilistic inference across eight industrial domains, and emits forecasts — directional reads, magnitude bands, confidence floors — that route directly into the operator's day through Forge.
Signal domains
8
Observations / day
~2.4M
Forecast window
9.6 d
Conviction floor
60%
Capabilities · Live inference
What MESO-1 produces, every minute it runs.
Each capability is a probabilistic surface — magnitude, direction, confidence — surfaced before broad market confirmation. The outputs are routable, auditable, and consumed directly by Forge.
μ
+1.4σ
Δp
+2.2%
τ
6.4 d
Inference pipeline
Four stages. Every prediction traceable to the signal that produced it.
Each stage emits structured artifacts — versioned, observable from inside Forge, and reproducible without re-deriving the upstream state.
8 domains · 240+ features
Stage 02
Feature Extraction
Domain-specific feature engineering across eight signal classes — supervised, evaluated, and versioned end-to-end.
Live · continuous inference
Stage 03
MESO-1 Inference
Continuous probabilistic prediction. Magnitude, direction, and confidence emit as a live state, not a batch report.
forge / mission control
HRC
+2.4σ
CRC
+1.1σ
Plate
−0.6σ
Routed to operator surfaces
Stage 04
Forge Surfaces
Operator-facing decision routing across Mission Control, Converse, Citadel, and Research.
Substrate · Material truth
The lot before the trade. The metal before the model.
MESO-1 trains against material reality — physical inventory, dispatch cadence, and clearing pressure — not just price tape.
Plate 03 · Steel lot
Signal domains
Eight domains. One inference surface.
Every domain passes through the Arc Signal Transformer before MESO-1 fires. Primary domains carry the load; secondary domains condition the prediction.
01
Sentinel-1 SAR satellite imagery
primary 6 h revisit02
AIS vessel & freight tracking
primary Real-time03
Global steel exchange indices
primary Real-time04
Distributor inventory & quote graph
primary Real-time05
Municipal bonds & infrastructure permits
secondary Daily06
Energy grid load data
secondary Hourly07
Railroad & drilling permits
secondary Daily08
Weather & port congestion feeds
secondary 3 hPalisade Research · Technical Report
MESO-1 / v1.4 · Edition 2026.05
MESO-1: A Foundation Model for Industrial Forecasting
8 signal domains ~2.4M obs / day 9.6 d window
Research · Maturity
Built by an industrial research group, not a dashboard team.
MESO-1 is the inference substrate of a research program. Every release ships a versioned technical report — model card, training corpus boundaries, evaluation methodology, and known failure modes. Operators and counterparties read the same primary source the team does.
v1.4 · 2026.05
Current production
Sentinel-1 SAR augmentation
v1.3 · 2026.02
Forecast window extended to 9.6 d
Confidence floor raised to 60%
v1.2 · 2025.11
Counterparty risk surface introduced
Continuous re-scoring
Output surfaces
Where MESO-1 lands in the operator's day.
Predictions never leave the system as a number on a page. They route to the surface where they become a decision.
01
Mission Control
Live position, exposure, and recommended actions tied to current forecasts.
02
Converse
Natural-language interface to MESO-1 inference and reasoning.
03
Research
Long-running agent programs grounded in MESO-1 outputs.
04
Citadel
Per-SKU risk, depreciation, and turnover scoring.
05
Exchange
Listing recommendations and counterparty routing.