AI-assisted modernization of C#/.NET mission-planning software, with behavioral-equivalence proof and continuous-ATO artifacts - runnable end-to-end on a synthetic, unclassified surrogate.
Developed by 577 Industries, Inc. under U.S. Navy SBIR topic DON26BZ01-NV013 ("AI-Assisted Modernization and Optimization of Theater Mission Planning Center (TMPC) Software"). This repository is the C#/.NET/JavaScript/SQL extension of the FORGE EVOLVE code-modernization framework. (FORGE EVOLVE's prior published results cover COBOL/Fortran/Ada/VB6 → Java/Python/Rust/TypeScript; the C#/.NET capability here is newly developed for the TMPC stack.)
This repo lets you run the modernization pipeline yourself, with no API keys and no network, and inspect both the legacy input and the modernized output.
git clone https://github.com/577Industries/forge-evolve-tmpc
cd forge-evolve-tmpc
make demo # offline, deterministic, ~minutesmake demo runs the full pipeline on the synthetic surrogate and writes evidence to results/:
| Phase-I ask (topic) | What the demo does | Evidence produced |
|---|---|---|
| 1. Analyze architecture, identify modernization needs | Roslyn/Tree-sitter discovery → dependency graph, complexity, business rules, crypto inventory | results/discovery-report.json, results/business-rules.ttl |
| 2. Conceptual AI models for analysis/refactor/cyber | CLAR lift → risk-scored microservice decomposition → multi-agent transform | results/migration-plan.json, results/clar/*.jsonld |
| 3. PoC AI refactoring on a representative component | Modernize a coupled C# component to .NET 8; prove behavioral equivalence | results/equivalence-report.json, tmpc-modern-mds/ |
| 4. Phase II prototype plans + cyber | cATO artifacts as CI byproducts | results/stig-findings.*.json, results/control-map.yaml, results/sbom.cdx.json, results/provenance.json, results/poam.csv |
What to read first: companion/companion.md maps each proposal claim to the
exact file/test that backs it. EXCLUSIONS.md states why this is safe to publish.
Honesty note. Every metric the demo prints is measured on the synthetic, unclassified surrogate in
surrogate/and is preliminary - not validated against government TMPC code and not validated by the Navy. The surrogate is engineered to exhibit the same classes of technical debt and shapes of mission-planning logic the government described, so results transfer in kind, not in fidelity. Seesurrogate/DEBT.md.
companion/companion.md- each proposal claim mapped to the exact file/test that backs it.REPRODUCIBILITY.md- reproduce every cited metric offline, from a clean clone.EXCLUSIONS.md- why this is unclassified, ITAR-clean, and safe to publish.governance/pre-registration.md- metrics, thresholds, and kill-gates frozen before any run.- Companion site: https://577industries.github.io/forge-evolve-tmpc/
ingest → Discovery → CLAR → Migration Planner → Tool Orchestrator → Transformation
→ Validation (behavioral equivalence) → Cyber/cATO overlay → Governance
- Discovery - Roslyn semantic analysis of C# (+ Tree-sitter for JS/SQL, FORGE VB6 grammar): dependency graph, Tarjan SCCs, complexity vectors, business-rule extraction (AST + LLM majority vote → RDF), cryptographic inventory.
- CLAR - Cross-Language Abstract Representation (
clar-spec/): four layers, decouples source parsing from target generation; flags precision-constrained values so coordinate/time-on-target math is emitted withdecimal/checked arithmetic. - Migration Planner - composite risk scoring + spectral clustering → candidate microservice boundaries + a topologically-ordered, risk-scored migration sequence (the Phase II roadmap).
- Tool Orchestrator - routes transform tasks via
@577-industries/model-router; runs offline (deterministic transcript replay - the default), local (sovereign/air-gapped), or cloud. - Transformation - emits modern, testable .NET 8 with secure-by-construction patterns.
- Validation - differential + property-based testing with mission-data-aware oracles (route feasibility, anti-meridian, time-on-target); discrete outputs to exact equality, continuous to bounded relative error; reports a Chernoff confidence bound.
- Cyber/cATO - STIG checks (before/after), NIST 800-53 mapping, CycloneDX SBOM, hashchain
provenance (
@577-industries/hashchain-audit), POA&M. - Governance - human-in-the-loop review gates + tamper-evident audit trail.
| Path | Contents |
|---|---|
src/ |
The FORGE EVOLVE for TMPC engine (.NET). ForgeEvolve.Contracts = frozen interfaces/DTOs. |
orchestrator/ |
TypeScript orchestration layer (consumes @577-industries/*). |
surrogate/ |
The synthetic, unclassified MDS-like component + golden corpus + legacy runtime. |
clar-spec/ |
CLAR JSON-LD schema + per-language profiles. |
evidence/ |
Claim→Evidence Traceability Matrix (CETM) + validator. |
results/ |
Demo outputs (a committed reference run + your local run). |
governance/ |
Pre-registration of metrics + human review gates. |
make build # dotnet build + npm build
make test # unit + integration tests with coverage
make audit # run the CETM validator (claim→evidence honesty check)
make sbom # generate the CycloneDX SBOMToolchain: .NET SDK 8 (pinned in global.json), Node 18+, Python 3.10+ (corpus generation only),
Docker (legacy surrogate runtime for differential testing). See REPRODUCIBILITY.md.