ai_workflow_best_practices.md - Experience report on the multi-AI method: which practices were adopted, what changed after each, what they cost, and a five-item minimum adoption set — written to be readable outside this project
lessons_learned.md (repository only, not rendered on the site) - The tripwire layer: operating rules distilled from the incident catalog, indexed by the activity you are starting and by the symptom you are staring at. Layer 1 is what a session loads; Layer 2 carries one evidence card per rule. The file states its own rule and incident counts, and the reconciliation between clustered incidents and catalog rows, in its header
lessons_learned_evidence.md (repository only, not rendered on the site) - The cataloged incidents behind those rules, one row each, every INC-NNN citation resolving to the dated devlog/ entry it was mined from — a reference appendix, not a document to read through
User Reference
getting_started.md - Build BELFEM and run your first simulation: configure/build, the example workflow (gmsh mesh generation, then belfem), restart and material-database gotchas, deck validation with belfem-conf
input_file_reference.md - Living reference of the input.conf contract: syntax, every parsed section/key with types, units, defaults and parse-site citations, aliases, dead keys, and pitfalls. Extend it in the same session as any input-feature change.
mpi_support.md - Open MPI is the only supported MPI: why MPICH and Intel MPI are refused at configure time, what the -DALLOW_UNTESTED_MPI=ON override costs, and why the Open MPI link flags in the MUMPS and MKL configs must not be "made portable"
parallel_execution.md — choosing MPI ranks, OpenMP threads and allocator settings for a production run: assembly scales with ranks (measured 3.6–3.8×), the factorization does not, and memory is set by a rank-independent factor
Module-Specific Documentation
Infrastructure
Core - Fundamental utilities (logging, timing, types, constants, string tools)
1.18.0