"""W12 structural tests — generic synthetic fixtures only. No scenario-specific tokens (rooms/props/states/colors/narrative literals) in test methodology. Production checkpoint loader is exercised only by the W11 test in the sibling script test file; W12 tests use synthetic fixtures. """ from __future__ import annotations import json import sys from pathlib import Path import pytest _REPO_ROOT = Path(__file__).resolve().parents[3] _SCRIPTS_DIR = _REPO_ROOT / "backend" / "scripts" if str(_SCRIPTS_DIR) not in sys.path: sys.path.insert(0, str(_SCRIPTS_DIR)) def _synthetic_adapter() -> dict: return { "background_catalog": { "L01B01": {"depends_on_fp": ["FP1"], "depends_on_bg": [], "sub_location_label": "zone_a", "state_label_raw": "base", "loc_id": "L01", "space_key": "S01", "applies_to_shots": ["S1_Shot1"]}, "L01B02": {"depends_on_fp": ["FP1"], "depends_on_bg": ["L01B01"], "sub_location_label": "zone_a", "state_label_raw": "alt", "loc_id": "L01", "space_key": "S01", "applies_to_shots": ["S1_Shot2"]}, } } def _synthetic_w11_brief() -> dict: return { "group_set_briefs": { "G1": { "group_id": "G1", "fp_blocks": [{"fp_id": "FP1", "png_available": False, "production_camera_bg_ids": []}], "spatial_zones": [ {"zone_id": "Z1", "label_source_derived": "zone_a", "supporting_numbered_elements": [1]}, ], "openings_and_transitions": [], "persistent_fixtures": [], "state_transition_zones": [], "camera_axes": [ {"axis_key": "ax1", "from_zone": "Z1", "to_zone": "Z1", "reason": "within zone"}, ], "default_avoidance": "", "conflicts_and_assumptions": [], "director_set_summary": "", }, }, "per_bg_set_selections": { "L01B01": {"bg_id": "L01B01", "group_id": "G1", "fp_id": "FP1", "applies_to_shots": ["S1_Shot1"], "relevant_zone_ids": ["Z1"], "relevant_numbered_elements": [1], "matched_production_camera": None, "camera_axis_used": "ax1", "continuity_intent": "", "state_delta": "", "reconciliation_notes": ""}, "L01B02": {"bg_id": "L01B02", "group_id": "G1", "fp_id": "FP1", "applies_to_shots": ["S1_Shot2"], "relevant_zone_ids": ["Z1"], "relevant_numbered_elements": [1, 2], "matched_production_camera": None, "camera_axis_used": "ax1", "continuity_intent": "", "state_delta": "", "reconciliation_notes": ""}, }, "final_plate_prompt_candidates": { "L01B01": {"bg_id": "L01B01", "final_plate_prompt_text": "base text", "source_grounded_anchors_used": [], "set_brief_grounded": True}, "L01B02": {"bg_id": "L01B02", "final_plate_prompt_text": "alt text", "source_grounded_anchors_used": [], "set_brief_grounded": True}, }, } def _synthetic_fp_context() -> dict: return { "fp_prompt_path": "/fake", "fp_render_path": "/fake", "fp_prompt_status": "ok", "fp_render_status": "ok", "floor_plans": { "FP1": {"fp_id": "FP1", "group_id": "G1", "depends_on_fp": [], "key_elements": [], "numbered_elements": [ {"number": 1, "category": "area", "label": "z", "position_hint": "c"}, ], "camera_recommendations": [], "t2i_prompt": "", "applied_shots": [], "png_path": "", "png_exists": False}, }, } def _w12_ok_payload() -> dict: return { "candidate_floor_plans": { "FP1": { "fp_id": "FP1", "group_id_pointer": "G1", "candidate_diagram_t2i_prompt": "flat schematic plan ...", "candidate_key_elements": [], "candidate_numbered_elements": [ {"number": 1, "label": "z", "category": "area", "position_hint": "c", "zone_id_pointer": "Z1"}, {"number": 2, "label": "o", "category": "opening", "position_hint": "n", "zone_id_pointer": "Z1"}, {"number": 3, "label": "f", "category": "furniture", "position_hint": "s", "zone_id_pointer": "Z1"}, ], "candidate_camera_recommendations": [ {"bg_id": "L01B01", "sub_location": "zone_a", "camera_position": "near #1", "camera_height": "1.6m", "lens_hint": "35mm", "framing_notes": ""}, {"bg_id": "L01B02", "sub_location": "zone_a", "camera_position": "near #2", "camera_height": "1.6m", "lens_hint": "35mm", "framing_notes": ""}, ], "reconciliation_notes_vs_production": "smaller set", }, }, "per_bg_render_reference_instructions": { "L01B01": {"bg_id": "L01B01", "group_id": "G1", "fp_id": "FP1", "applies_to_shots": ["S1_Shot1"], "floor_plan_ref_role": "layout_only", "use_numbered_elements": [1, 2], "ignore_numbered_elements": [3], "camera_axis_used": "ax1", "camera_axis_source": "w11_axis", "visible_zone_scope": ["Z1"], "prior_bg_ref_role": "none", "render_prompt_appendix": "layout_only. use #1,#2. ignore #3. axis=ax1. prior=none.", "final_prompt_assembly_preview": "base text\n\nlayout_only. use #1,#2. ignore #3."}, "L01B02": {"bg_id": "L01B02", "group_id": "G1", "fp_id": "FP1", "applies_to_shots": ["S1_Shot2"], "floor_plan_ref_role": "layout_only", "use_numbered_elements": [1, 3], "ignore_numbered_elements": [2], "camera_axis_used": "ax1", "camera_axis_source": "w11_axis", "visible_zone_scope": ["Z1"], "prior_bg_ref_role": "same_room_state_transition", "render_prompt_appendix": "layout_only. use #1,#3. ignore #2. axis=ax1. prior=same_room_state_transition.", "final_prompt_assembly_preview": "alt text\n\nlayout_only. use #1,#3. ignore #2."}, }, } def test_w12_candidate_camera_recs_cover_all_current_bgs(): """Stage B coverage split: every adapter fp must have a candidate fp entry, and every adapter-mapped bg within that candidate fp must have a camera_rec.""" from experiment_floor_plan_generation_slice import _build_w12_compatibility_report w12 = _w12_ok_payload() rep = _build_w12_compatibility_report( w12=w12, adapter_plan=_synthetic_adapter(), w11_brief=_synthetic_w11_brief(), fp_context=_synthetic_fp_context(), production_diff_empty=True, db_write_count=0, image_import_seen=False, prev_run_id="fakeW11", missing_inputs=[], stage_status="generated", ) assert rep["invariants"]["candidate_floor_plans_cover_all_adapter_fps"]["pass"] is True assert rep["invariants"]["candidate_camera_recs_cover_all_bgs_within_candidate_fps"]["pass"] is True # Drop one camera_recommendation to force per-fp coverage failure. w12_bad = json.loads(json.dumps(w12)) w12_bad["candidate_floor_plans"]["FP1"]["candidate_camera_recommendations"].pop() rep_bad = _build_w12_compatibility_report( w12=w12_bad, adapter_plan=_synthetic_adapter(), w11_brief=_synthetic_w11_brief(), fp_context=_synthetic_fp_context(), production_diff_empty=True, db_write_count=0, image_import_seen=False, prev_run_id="fakeW11", missing_inputs=[], stage_status="generated", ) assert rep_bad["invariants"]["candidate_camera_recs_cover_all_bgs_within_candidate_fps"]["pass"] is False # Drop the whole candidate fp to force the fp-coverage invariant to fail. w12_no_fp = json.loads(json.dumps(w12)) w12_no_fp["candidate_floor_plans"].pop("FP1") rep_no_fp = _build_w12_compatibility_report( w12=w12_no_fp, adapter_plan=_synthetic_adapter(), w11_brief=_synthetic_w11_brief(), fp_context=_synthetic_fp_context(), production_diff_empty=True, db_write_count=0, image_import_seen=False, prev_run_id="fakeW11", missing_inputs=[], stage_status="generated", ) assert rep_no_fp["invariants"]["candidate_floor_plans_cover_all_adapter_fps"]["pass"] is False def test_w12_use_numbered_elements_subset_invariant(): """use_numbered_elements ⊆ candidate_numbered_elements.number for each bg.""" from experiment_floor_plan_generation_slice import _build_w12_compatibility_report w12 = _w12_ok_payload() rep = _build_w12_compatibility_report( w12=w12, adapter_plan=_synthetic_adapter(), w11_brief=_synthetic_w11_brief(), fp_context=_synthetic_fp_context(), production_diff_empty=True, db_write_count=0, image_import_seen=False, prev_run_id="fakeW11", missing_inputs=[], stage_status="generated", ) assert rep["invariants"]["per_bg_use_numbered_elements_subset_of_candidate"]["pass"] is True w12_bad = json.loads(json.dumps(w12)) w12_bad["per_bg_render_reference_instructions"]["L01B01"]["use_numbered_elements"] = [1, 99] rep_bad = _build_w12_compatibility_report( w12=w12_bad, adapter_plan=_synthetic_adapter(), w11_brief=_synthetic_w11_brief(), fp_context=_synthetic_fp_context(), production_diff_empty=True, db_write_count=0, image_import_seen=False, prev_run_id="fakeW11", missing_inputs=[], stage_status="generated", ) assert rep_bad["invariants"]["per_bg_use_numbered_elements_subset_of_candidate"]["pass"] is False def test_w12_use_and_ignore_disjoint_invariant(): """use ∩ ignore must be empty per bg.""" from experiment_floor_plan_generation_slice import _build_w12_compatibility_report w12 = _w12_ok_payload() rep = _build_w12_compatibility_report( w12=w12, adapter_plan=_synthetic_adapter(), w11_brief=_synthetic_w11_brief(), fp_context=_synthetic_fp_context(), production_diff_empty=True, db_write_count=0, image_import_seen=False, prev_run_id="fakeW11", missing_inputs=[], stage_status="generated", ) assert rep["invariants"]["per_bg_use_and_ignore_disjoint"]["pass"] is True w12_bad = json.loads(json.dumps(w12)) w12_bad["per_bg_render_reference_instructions"]["L01B01"]["use_numbered_elements"] = [1, 2] w12_bad["per_bg_render_reference_instructions"]["L01B01"]["ignore_numbered_elements"] = [2, 3] rep_bad = _build_w12_compatibility_report( w12=w12_bad, adapter_plan=_synthetic_adapter(), w11_brief=_synthetic_w11_brief(), fp_context=_synthetic_fp_context(), production_diff_empty=True, db_write_count=0, image_import_seen=False, prev_run_id="fakeW11", missing_inputs=[], stage_status="generated", ) assert rep_bad["invariants"]["per_bg_use_and_ignore_disjoint"]["pass"] is False def test_w12_camera_axis_resolves_split_by_source(): """camera_axis_used resolution branches on camera_axis_source. 'w11_axis' requires axis_key membership in the bg's group's W11 camera_axes; 'candidate_camera_recommendation' requires the bg_id to appear in the candidate fp's camera_recommendations and accepts a free-form axis slug.""" from experiment_floor_plan_generation_slice import _build_w12_compatibility_report # Baseline: both bgs use w11_axis with matching axis_key — pass. w12 = _w12_ok_payload() rep = _build_w12_compatibility_report( w12=w12, adapter_plan=_synthetic_adapter(), w11_brief=_synthetic_w11_brief(), fp_context=_synthetic_fp_context(), production_diff_empty=True, db_write_count=0, image_import_seen=False, prev_run_id="fakeW11", missing_inputs=[], stage_status="generated", ) assert rep["invariants"]["per_bg_camera_axis_resolves"]["pass"] is True # Switch L01B02 to candidate_camera_recommendation with a free-form slug # — pass because L01B02 IS in candidate_camera_recommendations. w12_cand = json.loads(json.dumps(w12)) w12_cand["per_bg_render_reference_instructions"]["L01B02"]["camera_axis_used"] = "free_form_local_axis" w12_cand["per_bg_render_reference_instructions"]["L01B02"]["camera_axis_source"] = "candidate_camera_recommendation" rep_cand = _build_w12_compatibility_report( w12=w12_cand, adapter_plan=_synthetic_adapter(), w11_brief=_synthetic_w11_brief(), fp_context=_synthetic_fp_context(), production_diff_empty=True, db_write_count=0, image_import_seen=False, prev_run_id="fakeW11", missing_inputs=[], stage_status="generated", ) assert rep_cand["invariants"]["per_bg_camera_axis_resolves"]["pass"] is True # w11_axis with unknown axis_key — fail. w12_bad_w11 = json.loads(json.dumps(w12)) w12_bad_w11["per_bg_render_reference_instructions"]["L01B01"]["camera_axis_used"] = "no_such_axis" rep_bad_w11 = _build_w12_compatibility_report( w12=w12_bad_w11, adapter_plan=_synthetic_adapter(), w11_brief=_synthetic_w11_brief(), fp_context=_synthetic_fp_context(), production_diff_empty=True, db_write_count=0, image_import_seen=False, prev_run_id="fakeW11", missing_inputs=[], stage_status="generated", ) assert rep_bad_w11["invariants"]["per_bg_camera_axis_resolves"]["pass"] is False # candidate source but bg_id missing from candidate_camera_recommendations # — fail. Drop L01B02 from candidate_camera_recommendations, then flip its # axis source to candidate_camera_recommendation. w12_bad_cand = json.loads(json.dumps(w12)) fp = w12_bad_cand["candidate_floor_plans"]["FP1"] fp["candidate_camera_recommendations"] = [ cr for cr in fp["candidate_camera_recommendations"] if cr["bg_id"] != "L01B02" ] w12_bad_cand["per_bg_render_reference_instructions"]["L01B02"]["camera_axis_source"] = "candidate_camera_recommendation" w12_bad_cand["per_bg_render_reference_instructions"]["L01B02"]["camera_axis_used"] = "free_form_local_axis" rep_bad_cand = _build_w12_compatibility_report( w12=w12_bad_cand, adapter_plan=_synthetic_adapter(), w11_brief=_synthetic_w11_brief(), fp_context=_synthetic_fp_context(), production_diff_empty=True, db_write_count=0, image_import_seen=False, prev_run_id="fakeW11", missing_inputs=[], stage_status="generated", ) assert rep_bad_cand["invariants"]["per_bg_camera_axis_resolves"]["pass"] is False # floor_plan_ref_role='not_used' branch: candidate fp is irrelevant; axis # must just be a valid axis_key in the bg's group's W11 camera_axes. w12_not_used = json.loads(json.dumps(w12)) w12_not_used["per_bg_render_reference_instructions"]["L01B02"]["floor_plan_ref_role"] = "not_used" w12_not_used["per_bg_render_reference_instructions"]["L01B02"]["camera_axis_source"] = "candidate_camera_recommendation" w12_not_used["per_bg_render_reference_instructions"]["L01B02"]["fp_id"] = "" w12_not_used["per_bg_render_reference_instructions"]["L01B02"]["camera_axis_used"] = "ax1" rep_not_used = _build_w12_compatibility_report( w12=w12_not_used, adapter_plan=_synthetic_adapter(), w11_brief=_synthetic_w11_brief(), fp_context=_synthetic_fp_context(), production_diff_empty=True, db_write_count=0, image_import_seen=False, prev_run_id="fakeW11", missing_inputs=[], stage_status="generated", ) assert rep_not_used["invariants"]["per_bg_camera_axis_resolves"]["pass"] is True # w11_axis branch: shrinking visible_zone_scope while keeping a # use_numbered_elements entry whose candidate zone_id_pointer is # OUTSIDE the chosen axis endpoints must FAIL (the renderer would draw # an out-of-axis zone). visible_zone_scope ∪ used_element_zone_ids ⊆ # endpoints. Mutate a candidate element to belong to a foreign zone. w12_cross_zone = json.loads(json.dumps(w12)) fp = w12_cross_zone["candidate_floor_plans"]["FP1"] for e in fp["candidate_numbered_elements"]: if e["number"] == 2: e["zone_id_pointer"] = "Zforeign" break # L01B01 uses [1,2]; #2 now points to Zforeign which is not in axis # endpoints {Z1}. visible_zone_scope is still ['Z1'] but use covers # Zforeign — must fail. rep_cross = _build_w12_compatibility_report( w12=w12_cross_zone, adapter_plan=_synthetic_adapter(), w11_brief=_synthetic_w11_brief(), fp_context=_synthetic_fp_context(), production_diff_empty=True, db_write_count=0, image_import_seen=False, prev_run_id="fakeW11", missing_inputs=[], stage_status="generated", ) assert rep_cross["invariants"]["per_bg_camera_axis_resolves"]["pass"] is False # Same payload with source switched to candidate_camera_recommendation # for L01B01 — the zone subset rule no longer applies and L01B01 is in # candidate_camera_recommendations; the bg must pass. w12_cross_fallback = json.loads(json.dumps(w12_cross_zone)) w12_cross_fallback["per_bg_render_reference_instructions"]["L01B01"]["camera_axis_source"] = "candidate_camera_recommendation" w12_cross_fallback["per_bg_render_reference_instructions"]["L01B01"]["camera_axis_used"] = "local_zone_oblique" rep_cross_fallback = _build_w12_compatibility_report( w12=w12_cross_fallback, adapter_plan=_synthetic_adapter(), w11_brief=_synthetic_w11_brief(), fp_context=_synthetic_fp_context(), production_diff_empty=True, db_write_count=0, image_import_seen=False, prev_run_id="fakeW11", missing_inputs=[], stage_status="generated", ) assert rep_cross_fallback["invariants"]["per_bg_camera_axis_resolves"]["pass"] is True def test_w12_html_section_order_and_first_screen_priority(tmp_path): """HTML first screen must show per-bg instruction (Stage C) before candidate floor plans (Stage B), before group briefs (Stage A), with production baseline collapsed below invariants.""" from experiment_floor_plan_generation_slice import ( _build_w12_compatibility_report, _render_w12_html, ) w12 = _w12_ok_payload() rep = _build_w12_compatibility_report( w12=w12, adapter_plan=_synthetic_adapter(), w11_brief=_synthetic_w11_brief(), fp_context=_synthetic_fp_context(), production_diff_empty=True, db_write_count=0, image_import_seen=False, prev_run_id="fakeW11", missing_inputs=[], stage_status="generated", ) run_meta = { "run_id": "RID1", "stage": "w12_floor_plan_generation_slice", "run_status": "succeeded", "exit_code": 0, "derived_from": "fakeW11", "model_used": None, "stage_a_source_mode": "derived_from_w11_plus_w3_source", "source_language_mode": "not_available", "image_generation_count": 0, "image_generation_backend": "gpt-image-2", } _render_w12_html(run_meta, w12, rep, _synthetic_w11_brief(), _synthetic_fp_context(), tmp_path) html = (tmp_path / "index.html").read_text() sections = [ "Per-bg render reference instruction", "Floor-plan candidate", "Group briefs carried from W11", "Invariants", "Production baseline floor_plan_prompt JSON", ] positions = [html.find(s) for s in sections] assert all(p > 0 for p in positions), positions assert positions == sorted(positions), positions # Production baseline must live inside a collapsed
: the # immediately preceding
opener must come before its summary text. baseline_pos = html.find("Production baseline floor_plan_prompt JSON") details_pos = html.rfind("
", 0, baseline_pos) assert details_pos > 0 and details_pos < baseline_pos # No image section. assert "