Roadmap · Phase 0 → Phase 6

Six-phase plan to take AgniKawach from foundational infrastructure through to GPU Monte Carlo radiation. Snapshot 2026-05-22. Status reflects code-in-tree, not aspiration.

0 Infrastructure foundation ~ 90 % SUBSTANTIALLY DONE

HazmatDB, CoolProp, RADCAL WSGGM tables, Mie tables (drafted), radiation solver, AKSource tensor + production LUT.

1 Tier-1 complete · QRA v1.0 · AKSource unified stack ~ 90 % SUBSTANTIALLY DONE

Integral models for all 8 consequence types + QRA / LOPA / regulatory packs + unified source-modelling tensor.

2 Tier-2 dispersion · geometry ~ 75 % CURRENT

VC-002 → EB-004 V&V cases; semi-implicit Brinkman; Boussinesq-MW; 5 in-tree patches.

Phase-2 close criteria. All four open dispersion gaps closed: anisotropic dz lift, AMR + turb spectral filter validated, anelastic buoyancy, EB wall-function landed.
3 Tier-2 fire — pool + jet ~ 30 % IN BUILD-OUT

Standalone radiation matured (P1 + 4-mode MC + WSGGM + soot). Patch 0008 combustion kernels (TFC/EDC/Detonation/Fireball) land 2026-05-18.

4 Tier-2 explosion · BLEVE · flash fire — % QUEUED

PeleC compressible reactive flow. PDR is required at coarse grid for industrial congestion.

5 QRA workbench v2.0 ~ 40 % PROTOTYPE

F-N curves, LOPA, GIS contours. Frontend prototype already in tree at qra_dashboard.html.

6 GPU Monte Carlo radiation — % DEFERRED · RESEARCH

Move MC ray-tracing to CUDA kernels. Sobol + hot-ray already vectorisable; the win is the tracking kernel.

Cross-cutting infra dependencies

Compute orchestration (AWS A100 spot runner) P0 BLOCKER

K8s + Karpenter spot job runner pulling scenario JSON, building PeleLMeX with USE_CUDA = TRUE, shipping plotfiles to S3. Required to unlock the ₹2 500 — 8 000 / hr A100 cost model. Not solver work but blocks Phase 3 graduation and the live frontend pipe (see coverage gaps).