Embeddings timeout was 120s, causing multi-minute hangs when Ollama was
down during F4/F6/F7 solved-search. Now: connect_timeout=5s across all
HTTP clients (REST + Ollama), read_timeout reduced to 30s for embeddings,
and OllamaEmbeddings auto-restarts the server on ConnectionError before
degrading gracefully.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Lever 1: Join-graph via FK logics in annotations + suggest-fks command
- TableAnnotation.foreign_keys field stores curated logical FKs (DWH has no FK constraints)
- tht schema suggest-fks: mine from approved SQL, heuristics (time_key → dim_time),
same-name discovery + explicit --assume flag for multi-owner PKs
- mschema renders 【Foreign keys】 section populated; validation in merge.py
- SKILL.md F4 now reads FKs from mschema-text, no custom data_time_key logic
Lever 2: Context-pack consolidation at kickoff (tht search pack)
- Single embedding of question, reused for schema + evidence + solved searches
- One command: tht search pack <question> --session <id> → retrieval_pack.md
- Graceful degradation when Ollama/vector store unreachable (exit 0, empty sections)
- SKILL.md F1 prescribes as first call; reduces model thinking turns via pre-retrieval
Lever 3: Phase-summary recap v2 auto-construction from session ledger
- tht session show --json includes full decisions ledger
- tht phase meta --json exports 'emits' (substantive decision types per phase)
- Gate appends deterministic 【Decisioni registrate in questa fase】 section (appendLedgerSection)
- Model authors only summary + checks; recap table comes from persisted state (exact by construction)
- SKILL.md Disciplina 6: brief model output, gate fills the rest
Tests: 358 Python (including 10 FK + 3 pack + 1 session-ledger tests) + 111 JS gate tests, all pass.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Transcript analysis (session 2026-07-06-175012, GLM 5.2) showed F1 at 567s:
182s wasted on a useless `tht schema introspect` (re-introspecting the remote
DWH although physical.yaml was already materialized) plus ~220s of model
thinking inflated by ~7 exploratory turns (--help/find/cat). The actual
searches cost ~15s; reviewer gates (~145s, untouched) are the quality contract.
- schema_cmd.py: introspect now exits 0 with "OK (cache)" in ~1s when
physical.yaml exists; --refresh forces the real re-introspection.
Deterministic cross-model guarantee, verified live on psd (163 tables, 1.2s).
- SKILL.md: F1 toolbox (only `tht search find` + `tht schema render`; no
introspect/--help/filesystem browsing; batch all searches in one turn);
F4 step 1 is render-only with a one-shot introspect fallback.
- tht-gate.js: `tht schema introspect ... --refresh` added to FORBIDDEN
(maintenance stays shell-only, never in-session).
- tests: 4 new pytest cases (cache hit placement proven with fake credentials,
refresh bypass, corrupt-catalog fall-through, render fallback message) and
2 gate anti-bypass JS cases.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
WS1 of review-gates-v2: gives the JS gate (WS2) deterministic data to build the
cte_result v2 payload.
- CteTestRecord gains optional preview_rows (JSON-coerced, truncated cells);
test_cmd populates it from the bounded result rows.
- New read-only `tht cte info <name> --session <id> [--json]`: plan
index/total, persisted .sql, cte_plan_doc.json entry (if any), last
CteTestRecord. Exits 1 with a clean stderr message on missing
session/plan/name/sql.
- `tht cte plan --doc -` validates a chain-doc JSON (ctes[].name must match
--name, same order) and writes it to cte_plan_doc.json; cte_plan.json stays
a plain list[str] (load-bearing for tht.phase.next_cte). --doc is optional.
New sessions get a concise Italian-keyword `name` instead of the truncated
question. `tht session new` (when no --name is given) derives it via a new
`_extract_name` helper using YAKE (pure-Python, unsupervised, Italian, no LLM),
dropping generic query verbs and keeping the top keywords in reading order;
falls back to `_summarize` if YAKE is unavailable. `create_session` core keeps
its `name=None` default — the policy lives at the CLI layer.
TDD: tests/test_session_name.py (unit + CliRunner integration). Full harness
suite 269 passed; ruff clean.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
ctx.ui.input in `pi --mode rpc` correlates extension_ui_response on its own
top-level RPC id (crypto.randomUUID), not the descriptor id the gate carries
in `title`. SessionBridge replied with the descriptor id, so Pi silently
dropped the response and the model never resumed — every reviewer widget hung
after the human answered.
SessionBridge now stores Pi's top-level m.id (pendingPiId) and replies
extension_ui_response{ id: pendingPiId, value: <uiResponse> }; value still
carries the descriptor id so the gate's internal resp.id === descriptor.id
check still holds.
The fake-pi double had masked the bug by forcing m.id == descriptor.id; it now
mirrors real Pi (distinct randomUUID, correlate on it, drop unknown ids), with
a negative regression test. SKILL.md Phase 1 also now steers multi-answer
disambiguation to reviewer_decide (multiselect).
Tests: backend 67/67, tsc clean, fake-pi contract 2/2.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
workspace/provider/model/thinking now come from getSettings() injected into
sessionRoutes; the request body supplies only question+name. Also teaches
fake_pi_rpc to respond to set_model and set_thinking_level RPC commands so
tests that pass real model settings don't hang.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Harness:
- preview_cmd FILE positional arg made optional; when omitted with --session,
path is derived via _session_sql_file (mirrors export_cmd) — fixes the
deferred Task-5 bug where the backend passed sessions/<id>/sql_final.sql
relative to harnessDir, which broke for workspace-dependent paths.
- New pytest: test_preview_session_no_file_resolves_sql_final
Backend:
- ThtRunner.sqlPreview: drop positional file arg; use --session only
- New routes/sql.ts: POST /sessions/:id/sql/preview + /export
- New routes/meta.ts: GET /workspaces (yaml scan) + GET /models (injectable
seam + graceful fallback to {models:[]})
- app.ts: register sqlRoutes + metaRoutes; add listModels to BuildAppDeps
- tht-runner.test.ts: add sqlPreview argv assertion (no file path)
- test/routes-sql-meta.test.ts: 9 tests (sql preview/export + meta routes)
Tests: harness 233 passed; backend 29 passed; build clean.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
- test_list_sessions_required_keys: assert full spec key set
(adds summary/updated_at/author) so dropping any goes caught.
- test_cli_show_json_valid: assert "schema" in / "db_schema" not in
data to pin by_alias=True on the alias-sensitive field.
Production code unchanged. 8 passed; full suite 230 passed.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
When offset>0 the wrapper added an outer LIMIT N, so run_controlled's
_inject_limit bailed (a LIMIT IS present) and truncated was always False —
AGGrid could never detect more rows. Fix: for offset>0 probe with LIMIT (N+1)
OFFSET M, then compute truncated = len(rows) > N in do_run and slice back to N.
offset==0 path unchanged (delegates to extracted _run_transport helper). JSON
still reports the user's requested limit N and correct truncated. Adds 3 tests
exercising the real do_run offset>0 path (N+1 -> truncated True, N -> False,
offset==0 verbatim). 222/222 passing.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Add inject_limit_offset (tht/execute/limit.py) — pure subquery wrapper that
applies LIMIT/OFFSET non-destructively without clobbering user-supplied LIMITs.
Wire offset param into do_run (pre-processing when offset>0) and add --offset /
--json flags to preview_cmd; JSON mode emits pristine stdout with columns, rows,
execution_ms, truncated, limit, offset. 5 new tests (4 unit + 1 JSON-purity),
219/219 total passing (no regressions).
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Implementazione del piano di remediation progressiva sui difetti emersi
dall'analisi dell'harness. Tutto verificato: 214 test Python (incl. L0 su
Postgres reale), 14 test JS del gate, ruff pulito.
Blocco 1 (CRITICA, integrazione gate↔CLI):
- phase advance: gate usa --auto + exit 6; reviewer_confirm kind:phase fa
advance esplicito che applica i prerequisiti (prima non avanzava per le
fasi a conferma umana).
- cte plan riceve i --name dal gate (param names); set-question con id
posizionale; skill `tht search find`; nuovo comando `tht memory save-one`
con dedup hash client-side in save_one_memory.
Blocco 2 (D15, stato post-rollback):
- campo `phase` su DecisionRecord + effective_decisions phase-aware per i
subject "a nome" (cte_approved ecc.); _compute_promotions e finalize sulla
vista effective; finalize confronta col piano CTE effettivo, non glob;
`decision add --retracts` + comando `decision retract`.
Blocco 3 (D7 read-only + D6 manifest):
- assert_read_only su tutti e quattro i codepath (direct + REST);
- manifest author/summary/updated_at/updated_by/schema_version popolati +
helper touch_manifest sulle mutazioni.
Blocco 4-5 (D14a/D14b):
- decision_min_phase data-driven via `emits:` in workflow.yaml;
- formula evidence: status auto, search_formulas, gruppo CLI `tht formula`,
`search find --kind formula`, load_evidence_dir salta i .sql.md.
Blocco 6 (robustezza):
- taskdoc slice promoted_tables + bound enforced; report escaping/bound +
rsplit note; filtro kind reader REST/direct; conteggio upserted robusto;
guard REST run_query non-list; LSH disallineato -> LshIndexError.
Blocco 7 (pulizia):
- dead code gate e KIND_TO_TABLE morto rimossi; doc Postgres-only
(README + connection.py).
Blocco 0 (parziale): test di compatibilità firma gate↔CLI
(tests/integration). Rinviati: fake-Pi runtime completo, artifact-gate da
disco (#23), parità eligibility REST/direct (#28), unificazione
reserved-labels (#30), memory_rejected da deselezione (#33).
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Correzione del gap ereditato (resosi NECESSARIO dal drop del registry, spec 5): il
metadata del VectorRecord memory ora porta subject/detail/rationale oltre a
type/session_id/tables/concepts. pack_metadata li serializza nel jsonb via
**record.metadata. search_similar proietta metadata completo -> la F2 ricostruisce
la decisione direttamente dall'hit, senza lookup registro.
L1: 4 test (subject/detail/rationale presenti, campi esistenti preservati,
no cross-contamination multi-record, save_one_memory propaga il metadata alla riga).
Suite: 165 passed.
require_phase_or_exit: guard riscritto vs Workflow (load_workflow().phase_name invece
della costante PHASE_NAMES drift). Exit 1 se la sessione e' sotto soglia. Usato da
cte/decision/datamart cmd.
Comandi phase (portati + adattati al modello ThothII, non copia cieca):
- advance: persiste phase_approved; --auto exit 6 se la fase non e' completa
(contratto col gate)
- reopen: persiste phase_reopened + teardown_to_phase degli artefatti oltre il target
- show: stato sessione (fase corrente, ultime decisioni)
session_dir helper tenuto qui (mirror di session_cmd) per evitare circular import.
_cfg() fa fallback a THT_WORKSPACE env finche' _load_config_or_exit (Onda 1.4) non
sara' portato.
L1: 4 test require_phase_or_exit (allow at/above, exit below, message con nome fase
dal workflow). Suite: 161 passed.
Aggiunge il metodo che i cmd CLI useranno al posto della vecchia costante
SCHEMA_LINKING_PHASE (drift fix Onda 1). Ritorna il num della fase il cui
artifacts_out contiene schema_linking.json, default 5 se nessuna la dichiara.
L1: 4 test (fase reale F4, posizione arbitraria, default 5, artefatti multipli).
Suite: 157 passed.
44 test L1 sui 3 moduli backend con logica non banale (opzione 2 della user review):
- sqlcheck.validate_sql (16 test): parse/single-statement, read-only enforcement
(INSERT/UPDATE/DELETE/CREATE/DROP/ALTER/TRUNCATE/GRANT rifiutati, WITH/UNION ok),
forbidden functions (dblink default blacklist, custom set, allowed not flagged),
object-existence (tabella inesistente, CTE non flaggata, perimetro promoted warning,
colonna inesistente con alias). Documenta una limitazione reale: le funzioni
aggregate specializzate (count/sum/coalesce) NON sono catturate dal name-matcher
perche' sqlglot modella .name come argomento, non come nome funzione.
- ctetest (14 test): has_trailing_select (semantica controintuitiva: True = violazione),
last_cte_name, build_test_sql, ledger I/O (load/append roundtrip, JSON-array e
JSONL tolleranti, corrupt-ledger raise).
- execute._inject_limit (6 test): LIMIT iniettato quando assente (limit+1 per
troncamento), rispettato quando presente, non iniettato su non-query, UNION/WITH ok.
Suite: 153 passed (109 + 44). Bonus: __psd_probe__ -> __tht_probe__ (riferimento
cliente neutralizzato in ctetest).
Bug trovato provando la connessione reale col .env: il write endpoint vive su un
PATH DEDICATO /vector/write/v1/ (non /vector/v1/), e il modello Config ha write_rest/
vector_write_rest a TOP-LEVEL (non nidificati in vector_db).
- .env.example: aggiunge THOTH_VEC_WRITE_REST_URL (path dedicato del writer, con
avviso che le due chiavi valgono su path separati).
- workspaces/chirone-test.yaml: riscritto allineato a chirone.example.yaml + config.py
(vector_rest/vector_write_rest top-level; write_rest punta a THOTH_VEC_WRITE_REST_URL).
- tests/l2/*: corretti gli accessi strutturali (ws.vector_write_rest invece di
ws.vector_db.write_rest; ws.vector_rest invece di ws.vector_db.rest).
test_value_grounding_real skip-when-import-fails su nsp.lshindex (modulo deferred da B3).
Verificato end-to-end: save_one_memory (embeddings -> writer REST /vector/write/v1/
-> upsert pgvector -> read-back reader) PASSED. Suite L0+L1: 109 passed. Suite L2:
4 passed, 1 skipped (lshindex deferred).
Nota operativa: THOTH_SSL_CA va lasciato VUOTO sulla workstation (cert GoDaddy
pubblico in certifi). I campi direct-transport (THOTH_DB_*, THOTH_VEC_PASSWORD)
sono obbligatori per il modello ma inutilizzati in transport=rest: riempiti con
dummy nel .env locale (come faceva ChironeWp3).
Two fixes found while unblocking the L2 setup:
1. conftest: THOTH_SSL_CA is NOT an L2 prerequisite. The DWH endpoint presents a
public cert (*.policlinicosandonato.it, signed by GoDaddy), already in the
certifi bundle, so the REST clients validate TLS with verify=True -- no CA file
needed. The ssl_ca line was commented out in ChironeWp3's nsp.yaml too.
2. test_workspace: the profile-default assertion collided with the operator's real
harness/.env once load_dotenv (D3) started injecting THOTH_PROFILE into the
process env. The test now dels THOTH_PROFILE to assert the actual *default*
(server), regardless of what the operator set in .env.
Suite: 109 passed.
Pre-release, non-deterministic tests (marker l2, skipped without .env + VPN). They
close the gaps L1 leaves open: real value grounding on the live schema, real memory
save-one upsert to pgvector, and the full GLM 5.2 -> gate conversation on the
'ablazione' question (which exercises D14 value grounding + formula on a multi-
column case + the gate glue L1 cannot reach).
workspaces/chirone-test.yaml points at the remote endpoints (DWH read-only +
pgvector dual-key, TLS self-signed); secrets via ${THOTH_*}.
- test_session_ablazione: precondition checks (workspace loads, env present, pi on
PATH) + the documented manual run protocol (human-in-the-loop; scripted-answers
variant is a follow-up). Default run skips cleanly.
- test_value_grounding_real: 'ablazione' grounds to multiple columns on the real
schema (D14a non-collapsing), needs a built LSH index.
- test_memory_save_one_real: save_one_memory upserts one row via the writer key
(D11) and search_similar retrieves it via the reader key.
Operator runs before release (pytest -m l2). Default run: 109 passed, 5 skipped.
conftest loads harness/.env once (session, autouse) via python-dotenv, and exposes
an l2_env fixture that SKIPS (not fails) when any L2 prerequisite var is missing/
empty: THOTH_DWH_API_KEY, THOTH_VEC_API_KEY, THOTH_VEC_WRITE_API_KEY, THOTH_SSL_CA.
So the default run (pytest = L0+L1, addopts '-m not l2') stays green without .env;
only pytest -m l2 (pre-release, with .env + VPN) exercises them. l0/l2 markers were
registered in A9. tests/l2/ package created for the L2 tests (D4, D5).
Pure-logic smoke (no LLM, no DB) that builds a synthetic ledger by hand and asserts
the Phase-A substrate stays coherent: full F1->F8 walk reaches terminal phase
(max+1); rollback truncates the effective view (stale phase-7 decision excluded
after reopen to F4) and resets current_phase; teardown deletes artifacts beyond the
target while preserving the target phase's; re-approve after rollback advances
correctly; taskdoc stays under byte budget across all phases; decision_retraction
excludes the retracted seq + the marker itself from effective_decisions.
This is the CI-runnable coherence net for the L2 session test (which exercises the
LLM->gate loop that L1 cannot).
D13 instructs the model to evaluate Altro/Rifiuta/steering free text in context,
act on it, re-ask if ambiguous, and record the user's words in the decision
rationale. The actual interpretation is model behavior enforced by the skill prose
+ gate, validated at L2; this test pins the RECORDING contract the gate relies on:
free text from 'Altro' round-trips into the decision rationale and survives
persistence, never silently discarded.
L1: test_freetext_interpretation (4 tests) -- Altro text preserved, persistence
roundtrip (exact), multiline steering intact, empty rationale allowed.
The skill prose ('Interpretazione del testo libero') ports with the .pi/ skill
in Phase D.
New evidence/formula_store.py: ConceptFormula (concept, columns, sql, status,
sources) as a frontmatter-YAML + SQL-body unit, stored one-file-per-formula under
<formulas>/<slug>-<n>.sql.md. save_formula is append-only (competing drafts and
reviewed versions coexist); retrieve_formula(concept) returns all of them so the
gate can surface candidates and let the reviewer choose.
concept_formula_approved / concept_formula_rejected added to DecisionType
(records the reviewer's choice; approved formulas travel with schema-linking).
L1: test_formula (7 tests) -- retrieval by concept, save/reload roundtrip (SQL
body preserved, frontmatter well-formed), multiple formulas per concept, empty
on no-match / missing dir, decision-type existence, default draft status.
Deferred: --kind formula on nsp search (needs search_cmd porting) wires
retrieve_formula into the CLI; lands with the search command.
marcopan
