Multi-domain simulation
and tactical visualization.

Overview

TheaterForge is a research and development firm focused on making complex multi-domain analysis available on the hardware analysts already have, inside software environments they are already permitted to run.

Our engine performs trajectory modeling, telemetry parsing, and orbital computation entirely within a standard browser runtime. Analysis executes locally on the client machine. No mission data is transmitted to an external server, and no installation is required.

Core Competencies

Deployment Model

There is no server component and no installation. The engine executes inside a standard browser on the machine the analyst already has, and mission data has nowhere to go.

Findings

Four results from committed, hashed artefacts. Each figure states its own limits.

Differentiators

• Zero-footprint deployment. Runs in a standard browser on the machine the analyst already has. No installation, no backend to accredit, no workstation configuration — materially lowering the barrier to fielding analysis capability inside restricted enclaves.
• Offline by default. Ships with its network gate closed and runs fully offline: no telemetry, no scenario upload, no external analysis service. Optional basemap imagery is the only network-capable feature and can be disabled or served locally.
• Reproducibility, with the limit tested. The engine has a single stochastic draw — the intercept roll — keyed to shooter, weapon, round index and target rather than to a global sequence. Adding or removing elements that do not interact with an engagement leaves that engagement byte-identical, verified by test. Changes that alter engagement order shift the round index and re-roll it; that limit is published with the evidence rather than implied away.
• One engine, two modes. The interactive console and the batch analysis harness execute the same code on the same physics cadence. No data-generation versus presentation divergence to reconcile.
• Safe to learn against. A measure whose event never occurs returns null, never a sentinel value, and requirement checks reject null — so an undetected threat can never score as comfortable margin, and an agent trained against the engine cannot learn to win by letting threats through.
• Screening scale, verified. The 32-run design shown under Findings is verified by automated test to be saturated, to keep every pair of columns orthogonal, to balance every column, and to recover a known effect from a synthetic response. Test output is committed and hashed.
• Throughput. Approximately 1,000–3,000× real time on a single laptop core, scenario-dependent. A twelve-design-point study covering six engagement-hours completes in roughly ten seconds.
• Engineering discipline evidenced, not asserted. 379 automated tests across 36 suites at tag v0.3.0, static type checking over both the application and the command-line analysis harness, generated software bill of materials, continuous integration, and a runnable timestep convergence study that reproduces its own verification result on demand.
• Honest verification posture. The system is verified, and one component — orbital propagation including J2 secular perturbation — has been cross-checked against NASA GMAT, with the residual disagreement documented. No component is claimed as validated. Byte-identical reproduction is claimed for the same code on the same runtime, and every artefact records the version used. Capability and limitation are stated together.

Company Data

Contact