tgss-mem

Geometric Memory Manager — TTI.TOOL.TGSS-MEM-001

Memory #93 CURRENT

Sandbox-first build with dashboard wire-up validates For multi-sprint Tribernachi specs, build vertical slice (schema → store → MCP tool → tests → dashboard) end-to-end in sandbox before handing to dev team Validated 2026-05-03 on Sprint 2 of TTI.TOOL.TGSS-MEM-002 v1.0 (validity windows + memory_edges). Mark explicitly chose this over delegating straight to the dev team and confirmed "working perfectly" once the dashboard wire-up landed. **The pattern that worked:** 1. Write the full spec first (TGSS-MEM-002 v1.0), get explicit go-ahead on schema reconciliation, ordering, fold-and-retire decisions. 2. Cut a sprint branch (`sprint-N--edges`) on the actual repo (not a worktree). 3. Read the existing reference impl (e.g., the Python `pria_core/order_trace.py` for parity) before porting. 4. Build vertically — schema migration → port primitives → store layer → MCP tool — so each layer can be unit-tested in isolation before integration. 5. Wire the dashboard as the LAST step. Dashboard is the user-facing demo surface; landing it confirms the full stack composes correctly and gives Mark a click-around UI for sanity-check before commit. 6. Smoke-test routes via curl (`200 OK` per route) + content-grep for the new fields. The user gets confidence without having to manually verify each route. **Why:** This pattern exposes integration breaks (e.g., schema/migration mismatches, MCP tool signature drift, dashboard rendering of new fields) BEFORE the dev team inherits them. Ships a working artifact for Mark to inspect, not a spec promising one. Mark prefers seeing "270/270 tests pass + dashboard works" over "spec ready for dev team" because it removes the bug-discovery latency that would otherwise land mid-sprint. **How to apply:** - For any TGSS-MEM / TCE / Memory-Layer sprint, default to this loop unless Mark says otherwise. - Always confirm scope before starting (vertical slice, files touched, tests). - Use TaskCreate/TaskUpdate to track sub-steps so progress is visible. - Run the dashboard against the seeded sandbox DB; curl smoke-test all touched routes + JSON APIs; pass control back with explicit "next moves" options. - Branch stays uncommitted until Mark says commit — dashboard click-through is the review gate. — [feedback_sandbox_first_with_dashboard.md]

CompositeFCF1CC3AA78D074F7
Project prime13
Domain prime59
Type prime61
Importance1.000000 (CRITICAL)
Decay epoch0
Created2026-05-04 15:46:49
Valid from(unset)
Valid toNULL — still believed true

Outgoing Edges

No outgoing edges.

Link to Another Memory

Back to memories