Development roadmap

M0 — Reproducible baseline

• Remove out-of-bounds access and NaN values.
• Correct soil-layer center and interface geometry.
• Establish initialize → advance → diagnostics → finalize.
• Introduce forcing time stepping, real64, SI units and input validation.
• Add debug/release builds, tests, restart output and conservation residuals.

Exit gate: an unfrozen annual case runs without invalid values and passes water/energy, time-step and grid-convergence thresholds.

M1 — Conservative frozen-soil thermo-hydrology

• Implement unfrozen-water and freezing-characteristic relationships.
• Couple freeze-induced suction, ice impedance and latent heat.
• Add nonlinear iteration and adaptive time stepping.
• Verify against Stefan solutions, INTERFROST-style benchmarks and laboratory experiments.

M2 — Minimum land-surface boundary

• Rain/snow partitioning and a minimal snow scheme.
• Surface energy closure, bare-soil evaporation and root uptake.
• Replaceable upper and lower water/energy boundary conditions.

M3 — Vapor and dry air

• Add FSoilTH-V and FSoilTH-VA configurations.
• Quantify when additional gas-phase physics changes observable behavior.

M4 — Ground ice and deformation

• Separate pore ice, excess ice and segregation ice.
• Begin with excess-ice melt and thaw settlement.
• Add frost-heave mechanics only with an explicit validation strategy.

M5 — Multi-site and land-model integration

• Evaluate seasonal and permafrost sites on the Tibetan Plateau and beyond.
• Combine surface-flux networks with soil and permafrost observations.
• Establish offline CTSM comparison, then an optional FSoilTH soil backend.

The milestone order is intentional: new complexity does not proceed ahead of a reproducible, conservative and verified core.