Real-Data Readiness
PartialWhat must become real before a pilot can be described as operating on project data rather than demonstration data.
Readiness
| Component | Current source | Real source needed | Status | Required work | Dependency | Risk | Priority |
|---|---|---|---|---|---|---|---|
| Model data | IFC/STEP upload to R2 + sita_files / versions | Pilot project models with declared units and LOD | partial | Confirm ingest on a real façade IFC; keep project assignment server-side | R2, worker, Neo4j | Unit / LOD mismatch | P0 |
| Geometry | AABB RCC8 default; OCCT behind flag | Exact BRep relations on production ingest | partial | Decide whether to enable SITA_GEOMETRY_ENGINE=occt for the pilot | Worker image with OCC | Re-ingest rewrites citable :SPATIALLY edges | P0 |
| Graph | Per-file Neo4j graph; :CONTAINS unrepresented | Honest containment + project-scoped KG | partial | O9 ingest change for :CONTAINS; project-level linking | Worker ingest | Cross-model identity is designed, not connected | P1 |
| Structural Memory | Live Structural Memory UI over serialised graph | Same, on a real model | implemented | Validate coverage metrics on the pilot file | Graph | Full-graph serialisation cost | P0 |
| Materials | 1000 synthetic material rows | Cited EPD / Material2050 or equivalent | mock | Replace seeds; keep data_status contract | Reference repositories | Agents will still look live if labels are ignored | P0 |
| Carbon / EPD | Synthetic GWP A1–A3 on seed materials | Verified EPD records | mock | Load real_parsed / real_verified corpus | Materials | Report carbon is currently synthetic_data | P0 |
| Cost | 1000 synthetic cost rows + supplements | Regional rates with provenance | mock | Replace cost seeds; keep unit normalisation | Quantity extraction | Mixed units need element quantities | P1 |
| Lifecycle | Synthetic replacement intervals | Manufacturer / standard service lives | mock | Replace lifecycle seeds | Break-even module | Crossing year is honest only if intervals are real | P1 |
| Regulations | NCC-inspired synthetic clauses | Licensed code text with clause IDs | mock | Do not present sample clauses as compliance | Legal source | Professional liability | P0 |
| Manufacturer data | 3×50 DEMO-SCH/KAW/REY records | Parsed manufacturer PDFs + product metadata | planned | Pipeline described in docs/sita-manufacturer-catalogue-ingestion is unimplemented | Catalogue matcher | Cross-section must not be described as catalogue comparison | P2 |
| Agent execution | Heuristic specialists + optional LLM narrative | Same architecture over real reference rows | partial | Keep heuristics honest; do not claim LLM core logic | Reference data | Narrative looks like evidence | P1 |
| Decision persistence | Decision Runtime tables, append-only HITL | Same | implemented | Apply Alembic on deployed Postgres | DATABASE_URL | Migrations never applied to deployed DB | P0 |
| Reports | Reads Decision Memory; HTML export + digest | Same over non-synthetic metrics | partial | Keep DemoChip whenever synthetic_data is present | Decision Memory | A polished report can launder demo numbers | P0 |