[codex] Add Mistaber XLOG YD 87:1 slice

tests/xlog/test_sefaria_mcp_fetch.py

@@ -0,0 +1,68 @@
+import importlib.util
+from pathlib import Path
+
+
+MODULE_PATH = Path(__file__).resolve().parents[2] / "xlog" / "tools" / "sefaria_mcp_fetch.py"
+SPEC = importlib.util.spec_from_file_location("sefaria_mcp_fetch", MODULE_PATH)
+sefaria_mcp_fetch = importlib.util.module_from_spec(SPEC)
+assert SPEC.loader is not None
+SPEC.loader.exec_module(sefaria_mcp_fetch)
+
+
+def test_build_linked_source_selection_records_selected_and_skipped_links():
+    links = [
+        {
+            "category": "Commentary",
+            "ref": "Turei Zahav on Shulchan Arukh, Yoreh De'ah 87:1",
+            "type": "commentary",
+        },
+        {
+            "category": "Halakhah",
+            "ref": "Mishneh Torah, Forbidden Foods 9:1",
+            "type": "ein mishpat / ner mitsvah",
+        },
+        {
+            "category": "Commentary",
+            "ref": "Yad Avraham on Shulchan Arukh, Yoreh Deah 87:3",
+            "type": "commentary",
+        },
+    ]
+
+    selection = sefaria_mcp_fetch.build_linked_source_selection(
+        slice_id="smicha86_u01_s87_seif_1",
+        links=links,
+        selected_refs={
+            "Turei Zahav on Shulchan Arukh, Yoreh De'ah 87:1": (
+                "qualifies the first-pass cooking/eating/benefit split"
+            ),
+            "Mishneh Torah, Forbidden Foods 9:1": (
+                "upstream Rambam codification for Torah-level meat-milk prohibitions"
+            ),
+        },
+        source_links_artifact=(
+            "xlog/evidence/smicha86/u01/yd_87/seif_1/sources/seif-links.json"
+        ),
+        evidence_dir=Path("xlog/evidence/smicha86/u01/yd_87/seif_1"),
+        generated_at="2026-05-31T00:00:00+00:00",
+    )
+
+    rows = {row["ref"]: row for row in selection["links"]}
+
+    taz = rows["Turei Zahav on Shulchan Arukh, Yoreh De'ah 87:1"]
+    assert taz["selected"] is True
+    assert taz["family"] == "taz_turei_zahav"
+    assert taz["selection_reason"].startswith("qualifies")
+    assert taz["skip_reason"] == ""
+    assert taz["fetched_text_artifact"].endswith(
+        "sources/linked/turei_zahav_on_shulchan_arukh_yoreh_deah_87_1.json"
+    )
+
+    rambam = rows["Mishneh Torah, Forbidden Foods 9:1"]
+    assert rambam["family"] == "mishneh_torah_rambam"
+    assert rambam["selected"] is True
+
+    skipped = rows["Yad Avraham on Shulchan Arukh, Yoreh Deah 87:3"]
+    assert skipped["selected"] is False
+    assert skipped["fetched_text_artifact"] is None
+    assert skipped["skip_reason"]
+    assert selection["gate"]["status"] == "source_chain_expansion_required"

xlog/GOAL_BUNDLE.md

@@ -0,0 +1,215 @@
+# GDSD Goal Bundle: Mistaber On XLOG
+
+## Mission
+
+Re-implement Mistaber from scratch as a corpus of `.xlog` programs that encode
+Jewish legal sources with executable validation, ontology coherence, source
+provenance, and non-blocking human semantic review.
+
+This bundle adapts Goal-Driven Software Development (GDSD) and Goal-Question-
+Metric (GQM) into an agent-operated encoding system. It is designed for long
+running continuous work: each source slice becomes a small, validated XLOG
+change with evidence and a human validation gate, while later slices continue in
+parallel.
+
+## Strategic Outcome
+
+Mistaber becomes an XLOG-native legal reasoning system where:
+
+- the corpus is encoded as audited `.xlog` programs;
+- rules are executable and explainable;
+- source references come from Sefaria MCP;
+- commentary and upstream sources are fetched before human review and drive a
+  mandatory second-pass re-encoding for every seif;
+- ontology reuse is enforced automatically when confidence is high;
+- predicate drift is detected before it pollutes the core model;
+- humans validate semantic correctness before merge;
+- agents can keep encoding without waiting for each human response.
+
+## Scope
+
+In scope:
+
+- XLOG program architecture for codex, ontology, scenarios, and provenance.
+- Continuous agent workflow for source ingestion, encoding, validation, review,
+  and merge.
+- Ontology grounding and predicate reuse controls.
+- GQM metrics and evidence requirements.
+- PR/discussion templates for human validation.
+
+Out of scope for this bundle:
+
+- Porting the old ASP implementation directly.
+- Treating old HLL syntax as the target language.
+- Claiming a law is accepted without human semantic validation.
+- Maintaining ungrounded predicate aliases for convenience.
+
+## Roles
+
+### Encoding Agent
+
+Selects the next source slice, retrieves sources through Sefaria MCP, proposes
+semantic decomposition, grounds all atoms, writes `.xlog`, runs validation, and
+opens the review artifact.
+
+### Ontology Steward
+
+Maintains the canonical predicate catalog, synonym map, drift policy, and merge
+rules for new public concepts. This role may be an agent, but public ontology
+changes require human visibility in the review artifact.
+
+### Validation Agent
+
+Runs static, runtime, explanation, scenario, and regression gates. It writes
+evidence manifests and marks machine validation status.
+
+### Human Reviewer
+
+Validates that the encoded law faithfully represents the cited source material.
+This role approves, requests changes, rejects, or marks a slice as needing
+rabbinic/domain escalation.
+
+### Merge Automation
+
+Polls GitHub PRs/discussions without reminders, detects human verdicts, reruns
+gates, verifies scope isolation, and merges only approved slices.
+
+## Goal Tree
+
+### G0: Continuous Corpus Encoding System
+
+Build a sustainable loop that converts source slices into validated XLOG
+programs without blocking later slices on pending human review.
+
+Acceptance:
+
+- every slice has a durable branch, PR or discussion, evidence directory, and
+  ledger entry;
+- pending human validation does not stop future slice selection;
+- merge is impossible before human approval and green validation.
+
+### G1: Source Comprehension
+
+Acquire and understand source material from authoritative corpus APIs.
+
+Acceptance:
+
+- all source text and metadata come through Sefaria MCP;
+- source fetch logs include reference, version metadata when available, language,
+  retrieval timestamp, and MCP tool trace;
+- semantic claims are linked to source spans or exact references.
+
+### G2: Ontology Grounding
+
+Prevent concept drift and predicate pollution while preserving coverage.
+
+Acceptance:
+
+- each public predicate is grounded to a canonical ontology entry;
+- top-k atom and subgraph search is recorded before new predicate creation;
+- high-confidence matches automatically reuse canonical predicates;
+- uncertain cases are blocked from public ontology expansion until reviewed.
+
+### G3: XLOG Encoding
+
+Encode the source slice as executable XLOG modules.
+
+Acceptance:
+
+- encoded rules are idiomatic XLOG;
+- private helper predicates stay local unless promoted through ontology review;
+- provenance facts connect rules to source references;
+- no old Mistaber DSL is introduced as a compatibility layer.
+
+### G4: Mechanical Validation
+
+Prove that the encoding runs, explains, and handles scenarios.
+
+Acceptance:
+
+- static syntax and style gates pass;
+- `xlog run` outputs expected values for all scenarios;
+- `xlog explain` JSON or equivalent provenance evidence is captured;
+- regression checks prove no unrelated law slice changed.
+
+### G5: Human Semantic Validation
+
+Require human approval for every encoded law without blocking the continuous
+encoding pipeline.
+
+Acceptance:
+
+- a PR or discussion is created with source links, citations, formulas, rules,
+  scenarios, evidence links, and an explicit verdict section;
+- automation polls for verdicts on a schedule;
+- approved slices are merged only after validation reruns;
+- requested changes create a new encoding iteration and evidence refresh.
+
+### G6: Integration And Closure
+
+Keep the system coherent as the corpus grows.
+
+Acceptance:
+
+- ontology catalog, aliases, scenario index, and evidence manifests are updated;
+- merged slices are traceable from source to rule to scenario to human verdict;
+- metrics show drift, reuse, validation, and review health.
+
+## Goal Lifecycle
+
+```text
+candidate
+  -> sourced
+  -> source_chain_expanded
+  -> comprehended
+  -> commentary_comprehended
+  -> grounded
+  -> encoded
+  -> machine_validated
+  -> awaiting_human
+  -> human_approved
+  -> merge_validated
+  -> merged
+```
+
+Allowed non-terminal states:
+
+- `human_changes_requested`
+- `ontology_review_required`
+- `source_chain_expansion_required`
+- `source_blocked`
+- `machine_validation_failed`
+- `domain_escalation_required`
+
+## Non-Blocking Principle
+
+Human approval blocks merge, not encoding.
+
+The agent must continue with the next eligible candidate slice after creating a
+complete review artifact for the current slice. It must not wait idly for a
+human reply. The review monitor is responsible for checking existing PRs and
+discussions on every cycle and on a scheduled cadence.
+
+## Merge Gate
+
+A branch may merge only when:
+
+- source evidence exists;
+- commentary and upstream source expansion is complete;
+- second-pass re-encoding from the expanded source chain is complete;
+- ontology grounding is complete;
+- machine validation is green;
+- human verdict is approved;
+- branch scope matches exactly one law slice or one explicitly approved bundle;
+- final validation rerun passes after the human-approved revision.
+
+## Stop Conditions
+
+Stop the current slice and continue elsewhere when:
+
+- Sefaria MCP cannot retrieve required source text;
+- ontology grounding has no confident candidate and no reviewable proposal;
+- XLOG execution exposes a runtime blocker outside the slice scope;
+- the source requires domain escalation before a faithful encoding can be made.
+
+Do not stop the whole pipeline for pending human validation.