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.
Ports search/__init__.py (combined_search/RRF/aggregate) renamed psdwp3->nsp.
New aggregate_lsh_multi (the D14a deviation): groups LSH hits by table keeping
EVERY column where a value appears -- NOT collapsed to a single best column.
The old _aggregate_lsh hid alternative groundings (e.g. 'ablazione' matching both
a boolean flag and a free-text patologia field). aggregate_lsh_multi exposes all
columns so the value-grounding widget lets the reviewer choose the anchor(s).
Within one (table, column) the best-scored value is kept; columns ordered by score.
value_grounded added to DecisionType (records the reviewer's anchor choice).
L1: test_value_grounding (6 tests) -- multi-column exposure, grouping, within-column
best-value, ordering, empty, and the value_grounded decision-type existence.
Deferred: lshindex/ (needs vendor/thoth_lsh) and the L0 test_rrf.py land with the
nsp lsh build command + index-building path; not needed for the pure L1 core here.
The D11 deviation is a single-row pgvector upsert, not a full vectorstore resync.
Ports memory.py + session/{store,artifacts} + textutil (deps of memory), renamed
psdwp3->nsp. Decision import paths rewired from nsp.session.decisions to nsp.decisions
(our A4 port lives at the top level). session/models.py left UNCHANGED to preserve
the Phase-A ThothII additions (D12/D15 author/summary, D14a grounded_values,
D14b concept_formulas).
New in memory.py:
- memory_vector_record_for_decision(records, decision_seq): the single VectorRecord
for a chosen decision (reuses memory_vector_records, filtered to one).
- save_one_memory(records, decision_seq, writer, embedder): embeds one record and
calls writer.upsert_records('memory', [row]) -- NEVER writer.sync (that's the
full-resync, server-side-only path). Returns the upsert count.
L1: test_memory_save_one (5 tests) pins the contract -- single row, one upsert
call, sync never called, None/0 for unknown seq.
Deferred: the full nsp memory save-one CLI command (config/session loading + the
workstation write-guard) lands when memory_cmd.py is ported alongside the other
CLI commands. The pure D11 core is what L1 can honestly cover here.
Ports vectorstore/{rest_client,rest_writer,store,reader,embeddings,records},
evidence/model (leaf dep of records), and cli/_guards (require_vector_write_allowed
workstation write-guard). Renamed psdwp3->nsp, verbatim.
VectorRestClient gains an api_key property so reader/writer clients carry their
distinct keys visibly (spec D11: vector_reader / vector_writer on the same endpoint).
scripts/create_vector_reader_rpc.sql is NEW: the reader RPCs (search_similar,
list_tables) lived server-side in Supabase and were never versioned. Authored now
mirroring the writer allowlist pattern (table allowlist, security definer, revoke
from anon/authenticated, grant to vector_reader only). Writer RPC ported verbatim.
L1: test_vector_dual_key (7 tests) pins the dual-key construction + the workstation
write-guard (exit 4 without writer key).
nsp.cli app registers phase_app (the gate's workflow-fact source). phase meta
--json emits {schema_version, max_phase, phases:[{num,id,name,advance,artifacts_out}]}
from load_workflow() -- the single source of truth. F8/datamart is present (the
exact JS-drift bug in ChironeWp3, PHASE_NAMES truncated to 7, is structurally gone).
Scope: the 12 other command groups land in their porting tasks (A9 ports
db/mschema/_guards; B1 vectorstore; B3 search/lshindex). Eager-importing them now
would break the app on unported deps -- deferred to keep the suite green at each commit.
Genera un documento compatto per fase, derivato da artefatti + effective_decisions,
con byte budget enforced (target <20k token per un 35B/<200k). Mai incorpora
physical.yaml (~190k token, fatale). D15+D16 complementari: il brief delle decisioni
e' effective-aware, quindi post-rollback riflette lo stato corretto (le stale di
fasi > current_phase sono escluse).
6 tests (question+schema, no physical.yaml, budget ok/violato, header fase,
stale-excluded post-rollback). 32 total passing.
Cancella ogni artefatto la cui fase produttrice > target, usando artifacts_out di
workflow.yaml. Risolve il bug latente di ChironeWp3: ctes/*.sql orfani (non piu'
nel piano dopo un re-derive) restavano su disco e bloccavano finalize.
Da chiamare insieme all'append di phase_reopened per mantenere stato coerente.
6 tests (target 4/1/7, empty, missing, orfani CTE). 26 total passing.
The single most important architectural fix vs ChironeWp3: ALL helpers consult
effective_decisions() instead of raw list_decisions(), so the reopen-aware view is
consistent everywhere (fixes the bug where approved_ctes/advance_problems conflated
stale pre-reopen decisions with new ones).
Model (corrected during TDD):
- current_phase folds the audit (excluding retracted) with guard 'n == cur' --
already reopen-aware (old phase_approved:N after reopen to M<N don't advance).
- effective_decisions = decisions whose phase <= current_phase. A sql_approved at
phase 7 is stale when current_phase=4 after a rollback to F4, even if in the ledger.
Rollback to F4 does NOT invalidate decisions of phases 1-3 (they stay effective).
- decision_retracted markers excluded (audit-only).
Also: session/models.py ported (SchemaLinking + Candidate with grounded_values D14a
+ concept_formulas D14b). MAX_PHASE/PHASE_NAMES read from workflow.yaml via
load_workflow() (no duplication). Strada 2: ladder if-phase-N kept for now,
generic prerequisites evaluator (F2 full) deferred.
7 phase tests + 20 total passing.
- ported from ChironeWp3 (22 DecisionType, append-only jsonl, monotonic seq)
- added decision_retracted type + retracts field for step-level rollback (D15):
the retracted decision stays in the audit log, effective_decisions() (Task A5)
will exclude it from the active view
- 5 tests: retract marker + monotonic seq + retracts default + empty session +
literal includes retracted. All 13 harness tests pass.
marcopan