fix: scope workspace registry smoke cleanup

harness/README.md

@@ -22,16 +22,19 @@ The `tht` command is now on PATH. Node ≥ 20 is needed for the gate JS tests
 
 ```bash
 cp .env.example .env
-# fill in: THT_PROFILE, THT_DB_*, THT_DWH_API_KEY, THT_VEC_API_KEY,
-#          THT_VEC_WRITE_API_KEY, THT_SSL_CA, THT_OLLAMA_URL, ...
+# fill in: THT_PROFILE, THT_DB_*, THT_DWH_API_KEY, THT_SSL_CA, ...
 ```
 
 Keys are never logged; URLs are fine. Rotate any key that appeared in chat.
 
 ### `workspaces/<name>.yaml`
 
-A workspace wires the relational DWH + the pgvector (dual-key) + embeddings + evidence.
-See `workspaces/tht.example.yaml`. `${THT_*}}` tokens expand from `.env`.
+A schema-v3 workspace wires the external relational DWH to one internal Qdrant collection
+and the internal Ollama embedding model. Evidence paths remain workspace-local, while Qdrant
+stores the derived semantic projection for schema, Evidence, Memory, and solved-question
+records. The legacy files under `workspaces/` are retained as migration fixtures; new
+operator-facing descriptors live in the workspace Git registry. `${THT_*}` tokens expand
+from `.env`.
 
 > **DB support (MVP):** the `direct` transport supports **PostgreSQL only** (psycopg2
 > driver, `pg_*` catalog introspection, postgres-dialect sqlcheck/EXPLAIN). The central
@@ -130,7 +133,7 @@ tht/                 Python package (CLI + workflow + phase + decisions + db/res
 .pi/                 Pi project (settings, prompts, themes, extensions/tht-gate.js + gate/)
 workflow.yaml        single source of workflow truth (F2)
 workspaces/          workspace YAML definitions (D3)
-scripts/             reader/writer RPC SQL for pgvector (D11)
+scripts/             retained legacy SQL fixtures and workspace utilities
 tests/               L0 (testcontainers), L1 (logic + builders), L2 (real model + DB)
 docs/                testing guide + workflow editing
 ```

harness/docs/testing.md

@@ -59,20 +59,22 @@ the anti-bypass hooks. The glue depends on the Pi runtime (`pi.registerTool`,
 
 ## L2 — real LLM + real remote DB, manual / pre-release (NOT automated)
 
-**What:** end-to-end sessions with GLM 5.2 + the real Chirone DWH + pgvector, reached
-via REST over VPN. Plus the gate-glue validation (the part L1 cannot reach).
+**What:** end-to-end sessions with GLM 5.2 + the real Chirone DWH, plus the internal
+Qdrant/Ollama semantic services started by the ThothII stack. Plus the gate-glue
+validation (the part L1 cannot reach).
 
 **Dependencies (all required, skip cleanly if missing):**
 - LLM: Pi configured locally with GLM 5.2.
-- DB: the remote Supabase endpoints (DWH read-only + pgvector reader/writer), via VPN.
-- `harness/.env` populated with the API keys + CA path.
+- DB: the remote DWH endpoint, via VPN when required.
+- ThothII stack: internal Qdrant and Ollama services reachable from `core`.
+- `harness/.env` populated with the required DWH/model API keys + CA path.
 
 **Coverage (honest):** validates the assumption L1 cannot — that GLM 5.2 produces tool
 calls the gate accepts, that the skill's prompts lead to the expected interaction shape,
 that the gate glue handles real tool-call sequences (incl. Altro/Rifiuta/rollback),
 that value grounding and formula approval surface correctly on the real schema, that
-`memory save-one` upserts to the real pgvector. **Closes the skill→LLM→gate loop AND
-exercises the gate glue.**
+`memory save-one` upserts to the configured semantic store. **Closes the
+skill→LLM→gate loop AND exercises the gate glue.**
 
 **Honest limitation:** L2 is non-deterministic (the model may behave differently across
 runs) and slow/costly. It is a **pre-release safety net, not a regression gate**.

harness/tests/test_adapter_command_regressions.py

@@ -62,8 +62,8 @@ def test_memory_command_writes_through_factory_vector_store(monkeypatch):
     captured = []
     original_upsert = store.upsert
     store.upsert = lambda table, rows: captured.extend(rows) or original_upsert(table, rows)
-    # Server deployments write directly to pgvector and intentionally do not
-    # configure the workstation-only REST writer key.
+    # Legacy server deployments wrote directly through the factory and intentionally
+    # did not configure the workstation-only REST writer key.
     cfg = SimpleNamespace(profile="server", embeddings=object(), vector_write_rest=None)
     manifest = SimpleNamespace(id="s1")
     snapshot = SimpleNamespace(manifest=manifest, decisions=[], artifacts={})

harness/tests/test_memory_save_one.py

@@ -1,8 +1,8 @@
 """L1: tht memory save-one -- targeted upsert via the writer key (spec D11).
 
 The D11 deviation: instead of a full vectorstore resync (tht memory index / sync),
-a remote workstation with a writer key can push a SINGLE promoted decision to
-pgvector as a one-row upsert. This test pins the pure core of that behavior:
+the workflow can push a SINGLE promoted decision to the configured semantic store as
+a one-row upsert. This test pins the pure core of that behavior:
 - exactly one VectorRecord is built for the chosen decision_seq
 - the writer.upsert_records is called once with a single row
 - writer.sync is NEVER called (that is the full-resync path)

harness/tests/test_solved_search_cli.py

@@ -1,7 +1,7 @@
 """L1: `tht memory solved-search` — degrado gentile e mapping dei risultati.
 
-SKILL.md prescrive solved-search in F4/F6/F7 di OGNI sessione: a vectordb
-irraggiungibile (VPN giu', Ollama spento) il comando non deve morire con un
+SKILL.md prescrive solved-search in F4/F6/F7 di OGNI sessione: se lo store
+semantico è irraggiungibile (Qdrant/Ollama non disponibili) il comando non deve morire con un
 traceback grezzo ma degradare a un avviso di una riga su stderr, con stdout
 puro (`[]` in modalita' --json) ed exit 0, cosi' il modello prosegue senza
 exemplar. Il finalize-hook gestisce gia' lo stesso scenario in modo analogo.

harness/tht/memory.py

@@ -299,10 +299,10 @@ def memory_vector_record_for_decision(
 def save_one_memory(
     records: list[MemoryRecord], decision_seq: int, *, store, embedder
 ) -> int:
-    """Targeted one-row upsert of a promoted decision to pgvector via the writer key
+    """Targeted one-row upsert of a promoted decision to the configured semantic store
     (spec D11). This is NOT a full vectorstore resync: it embeds and pushes a single
-    record, so a workstation with a writer key can publish one memory without
-    rebuilding the index. Returns the upsert count (0 if no record matched or the
+    record, so a memory can be published without rebuilding the index.
+    Returns the upsert count (0 if no record matched or the
     record is already up to date).
 
     Hash dedup client-side (spec §5.4): the SHA-256 of the content is compared with

harness/tht/search/__init__.py

@@ -59,8 +59,8 @@ def aggregate_lsh_multi(hits: list[dict]) -> dict[str, list[dict]]:
     grouped: dict[str, list[dict]] = {}
     for (table, _), row in best.items():
         grouped.setdefault(table, []).append(row)
-    for table in grouped:
-        grouped[table].sort(key=lambda r: r["score"], reverse=True)
+    for rows in grouped.values():
+        rows.sort(key=lambda r: r["score"], reverse=True)
     return grouped
 
 
@@ -99,7 +99,7 @@ def combined_search(
     kinds: list[str] | None,
     query_vec: list[float] | None = None,
 ) -> list[SearchResult]:
-    """Fonde LSH (valori di campo) e pgvector con Reciprocal Rank Fusion.
+    """Fonde LSH (valori di campo) e ricerca semantica con Reciprocal Rank Fusion.
 
     `query_vec` permette di riusare un embedding gia' calcolato della stessa
     keyword (es. `tht search pack`, che fa piu' ricerche sulla stessa domanda)."""

harness/tht/vectorstore/store.py

@@ -52,11 +52,13 @@ def hit_from_metadata(similarity: float, metadata: dict | None) -> VectorHit:
 
 
 class VectorStore:
-    """Tabella pgvector table-scoped: scrittura diretta (loading) su una tabella dello schema
-    `vectors`. La lettura via REST avviene su `search_similar`; questo store serve al loading e
-    alla lettura diretta (dev/test). Il contratto della tabella remota richiede `id` (BIGSERIAL),
-    `embedding vector(N)` e `metadata jsonb`; le colonne extra (`record_key`, `kind`,
-    `content_hash`) servono solo al loader e non sono esposte dalla REST."""
+    """Legacy table-scoped vector store retained for compatibility fixtures.
+
+    New operational semantic storage is handled by the Qdrant adapter. This class preserves
+    the older SQL-table contract used by historical tests and migration checks: `id`
+    (BIGSERIAL), `embedding vector(N)`, and `metadata jsonb`; the extra columns
+    (`record_key`, `kind`, `content_hash`) serve only the loader and are not exposed by REST.
+    """
 
     def __init__(
         self, engine: Engine, schema: str = "vectors", table: str = "records", dim: int = 768

harness/tht/workspace.py

@@ -4,8 +4,9 @@ Reads workspaces/<name>.yaml, expands ${VAR} from env, validates via the Config
 (ported from the reference implementation). Future migration to a DB store would replace only this module.
 
 La struttura YAML rispecchia esattamente tht/config.py:
-  database + rest (DWH), vector_rest + vector_write_rest (pgvector, doppia key top-level),
-  vector_db (loading diretto, server-only), embeddings, evidence, execution.
+  database/rest o resources.dwh per il DWH, resources.vector/resources.embeddings
+  per Qdrant/Ollama interni, più evidence ed execution. I vecchi campi vector_db e
+  vector_rest restano solo per leggere fixture legacy durante la migrazione.
 """
 from __future__ import annotations