Roadmap

🗺️ Roadmap

GRANITE v0.6.8 | ← Physics Formulations | Scientific Context →

---

Version Roadmap

Version	Target	Status	Key Deliverables
v0.6.5	Q2 2026	✅ Released	BBH stable t=500M; 4-level AMR; 92 tests; Python dashboard; B2_eq benchmark
v0.6.7	Q2 2026	✅ Released	107 tests (20 suites); VORTEX Gold Master; identity purge; CI stabilization
v0.6.7.2	Q2 2026	✅ Released	granite_analysis package overhaul; CLI hardening; 25 broken links fixed; docs subcommand; Quick Start rewrite
v0.6.8	Q2 2026	✅ Released	"Red Team" audit remediation (4 sprints); CMake scoping; Rule-of-Five; namespace modernization; -ffast-math removal; AMR race fix; Wiki sync
v0.7.0	Q4 2026	🔄 In Progress	GPU CUDA kernels; checkpoint-restart; full dynamic AMR regrid; M1 wired into RK3
v0.8.0	Q1 2027	📋 Planned	Tabulated nuclear EOS + reaction network; SXS BBH waveform validation
v0.9.0	Q2 2027	📋 Planned	Full SXS catalog validation (~60 BBH configs); multi-group M1 radiation; Python steering API
v1.0.0	Q3 2027	🎯 Target	B5_star production run + paper; full community release; all OS support

---

Tier-1 Blockers for v0.7

These three items are confirmed blockers that must be resolved before v0.7 release:

Blocker	File	Issue
AMR dynamic regridding	src/amr/amr.cpp	regrid() evaluates criteria but doesn't rebuild hierarchy at runtime
Checkpoint-restart	src/io/hdf5_io.cpp	loadCheckpoint() stub exists but not implemented
M1 in RK3 loop	src/core/evolution_loop.cpp	M1 modules built and tested but never called in the time loop

---

GPU Porting Plan (v0.7)

Target hardware: vast.ai H100 SXM instances ($2–3/hr as of 2026)

Porting priority:

1. CCZ4 RHS hot loop (80% of compute time)
2. MP5 reconstruction
3. HLLD Riemann solver
4. CT update

Prerequisites: All CPU kernels must be fully validated first. GPU kernels will be verified against CPU results on test cases.

---

B5_star Scaling Path

v0.6.8:  128³ desktop (i5-8400, 16 GB)
           ↓ GPU kernel porting complete
v0.7:   GPU prototype (vast.ai H100, 256³)
           ↓ Strong/weak scaling benchmarks
v0.8:   Cluster production (NERSC/ARCHER2/Jülich, 512³)
           ↓ Full AMR hierarchy + B5_star config validated
v1.0:   Flagship production (~12 levels, ~2 TB RAM, ~5×10⁶ CPU-hours)

---

Known Limitations (v0.6.8)

Full table: see Simulation Health & Debugging: Known Diagnostic Limitations

Limitation	v Version Fix
loadCheckpoint() not implemented	v0.7
M1 radiation not wired into RK3	v0.7
Dynamic AMR regridding partial	v0.7
Phase labels time-based, not separation	v0.7
alpha_center reads from AMR level 0	v0.7
macOS / Windows native unsupported	v0.8

---