OjGkUdZddlmZddlmZmZmZmZmZm Z e hdZ de d<e ddhZ de d <e e zZde d <d Zd e d <ddddZde d<dZde d<GddeZd*dZ d+dZ d,dZ d-dZ d.dZd/dZ d0dZ d1d Z d2d!Z d3d"Z d4d#Z d5d$Z d6d%Z d7d&Z!d'ed( d8d)Z"y')9u W20A: base location dossier — pure deterministic builder. Consumes existing W19A v6 / W19B-1 outputs and produces a per-fp_id dossier that downstream W20B (shot-aware LLM planner) consumes. LLM / image / VLM / DB / ImageAsset write 0. **The VLM 10x10 readback gate is represented in the artifact but never invoked in W20A — only the synthetic_placeholder path is emitted.** Real VLM wiring is deferred to a future wave behind explicit approval. Dossier shape (per fp_id): { "fp_id": str, "fp_image_path": Optional[str], # from floor_plan_render cp "grid_size": [10, 10], "dwelling_identity": { "structure": { unit_markers, opening_markers, fixed_fixture_markers, anchor_furniture_markers, marker_count_per_layer }, "materials": { "diagnostics": [...] }, # LLM-emit slot "fixed_elements_inventory": [...], # base_persistent_fixture "standard_of_living_band": "modest_residential", "lighting_identity": { "diagnostics": [...] }, # LLM-emit slot }, "fp_geometry_candidates": { "grid_size": [10, 10], "camera_cell_candidates_per_unit": {}, # populated by VLM "look_at_cell_candidates_per_unit": {}, # populated by VLM "view_cone_lens_enum_map": {wide:90, normal:50, telephoto:25}, "visible_units_candidates": {}, # populated by VLM "visible_openings_candidates": {}, # populated by VLM "wall_door_invalidation_diagnostics": [], "diagnostics": [...], }, "fp_geometry_vlm_readback": { "status": "synthetic_placeholder", "grid_size": [10, 10], "observed_markers": None, "missing_markers": None, "extra_markers": None, "confidence": None, "gate_decision": "synthetic_pass", "diagnostics": [...], }, "base_marker_inventory": [...], # base_* "overlay_marker_inventory": [...], # state_overlay_* per (bg_id, number) "per_bg_render_facts_by_bg_id": { # W20A revision (Codex) # exact-ID per-bg facts W20B planner consumes without re-reading # raw W19 / master_plan checkpoints. integer / string equality only. "": { "bg_id", "fp_id", "target_unit_marker_numbers", "dominant_target_unit_marker_number", "use_numbered_elements", "ignore_numbered_elements", "base_marker_numbers_to_reference", "transient_marker_numbers_to_describe", "ignored_state_overlay_marker_numbers", "clean_background_expected", "applies_to_shots", "depends_on_bg", "diagnostics", }, }, "anchor_selection_metadata": { "candidate_bg_ids": [...], "selection_diagnostics": [...], "selected_anchor_bg_id": None, # W20B owns selection }, "diagnostics": [...], } W20 spec boundaries enforced: - exact-ID only (marker numbers / bg_id / fp_id exact equality). - no substring / regex / lexical inference over labels or position hints. - static contract carries no scenario-specific tokens — runtime label passthrough (label / position_hint) is data, not contract. - no BG-id keyed prose dict. - reference / camera selection NOT performed here (W20B). - anchor selection NOT performed here — only the candidate surface. ) annotations)AnyDict FrozenSetListOptionalTuple> base_openingbase_structural_unitbase_persistent_fixturebase_persistent_furniturezFrozenSet[str]BASE_LAYER_DECISIONSstate_overlay_plot_cuestate_overlay_transient_objectSTATE_OVERLAY_DECISIONSALL_LAYER_DECISIONS) rTuple[int, int]DEFAULT_GRID_SIZEZ2)widenormal telephotozDict[str, int]VIEW_CONE_LENS_ENUM_MAPmodest_residentialstrSTANDARD_OF_LIVING_BAND_DEFAULTceZdZdZy)BaseLocationDossierErrorz+Fail-closed signal for the dossier builder.N)__name__ __module__ __qualname____doc__`/Users/manta/Documents/Projects/TheRoad-I1/backend/app/modules/pipeline/base_location_dossier.pyr!r!s5r'r!c|xsijdxsi}|jDcic],\}}t|tr|jddk(r||.c}}Scc}}w)N floor_plansstatusokgetitems isinstancedict)fp_prompt_datafpsfidentrys r(_index_floor_plansr6sk  R $ $] 3 9rC))+ %JC eT "uyy':d'B U %  s1A"ci}|xsijdxsi}|jD]*\}}t|ts|jd||<,|S)u=Best-effort fp_id → png_path resolution. The floor_plan_render cp shape carries ``data.floor_plans[fp_id].png_path``. Dossier accepts a None when the path is missing — the gate registers a diagnostic but does not raise, because a dossier may legitimately be built before fp render completes. r*png_pathr-)fp_render_dataoutr3r4r5s r(_index_floor_plan_renderr;s]%'C  R $ $] 3 9rCiik U%& 99Z(C" Jr'ci}|xsijdxsi}|jD]\}}t|tr|jddk7r+|jdxsi}|jdxsgD]`}t|tr|jds%|jdxsg}|s=|j |dgj |b|S) Nplansr+r,plan backgroundsbg_id depends_on_fpr)r.r/r0r1 setdefaultappend)master_plan_datagroupedr=_gidr5r>bgdependss r(_collect_bg_specs_by_fprIs02G  # ( ( 1 7RE{{} e%&%))H*=*E yy &B((=)/R/Bb$'rvvgff_-3G   wqz2 . 5 5b 9 0 % Nr'c|xsijdxsi}i}|jD]M\}}t|ts|jd}t|ts9||j |i|<O|S)Noverlaysfp_id)r.r/r0r1rrB)overlay_payload_datarKrEr@payloadrLs r(_collect_overlays_by_fprOs~%*// ;ArH46G"..*w'4(  G$%% /65"%e, + Nr'c |jD]g\}}|jd}t|tst d|d|d|d|t vsEt d|d|d|dt t y) aRe-assert the W19B-1 partition enum on the dossier consumer side. Mirrors floor_plan_overlay_payload's exact-ID partition check. We re-do this here rather than trust the overlay payload alone, because the dossier consumes floor_plan_prompt v6 directly for its own marker inventory and a downstream change in the overlay step should not silently relax the dossier's contract. base_layer_decisionfp_id=z marker #z0 base_layer_decision missing or non-string (got )z base_layer_decision=z not in allowed enum N)r/r.r0rr!rsorted)rLmarkersnumnedecisions r(_validate_base_partitionrYs==?R66/0(C(* 3%0&&.\4  . .* 3%/D,3-./1 #r'c||jdd|jdd|jdd|jdddS)Nlabelcategory position_hintrQ)numberr[r]r^rQ)r.)rVrWs r(_marker_inventory_entryr`sH$FF:r*4!vv&;R@  r'c g}g}g}g}|jD]r\}}|jd}|dk(r|j|.|dk(r|j|E|dk(r|j|\|dk(sb|j|t|j|j|j|j||||t |t |t |t |ddS)NrQr r r r )r r r r ) unit_markersopening_markersfixed_fixture_markersanchor_furniture_markersmarker_count_per_layer)r/r.rCsortlen)rUunitsopeningsfixtures furniturerVrWds r(_structure_blockrnsEHHI==?R FF( ) & & LL  . OOC + + OOC - -   S !# JJL MMO MMO NN#!)$-$'JM'*8}),Y #  r'c g}t|D]b}||}|jddk7r|j||jdd|jdd|jdddd|S)u%Generic shape-kind inventory tied to base_persistent_fixture markers. Runtime labels (LLM-authored upstream) pass through verbatim — they may carry scenario-specific names because that content is data, not contract. W20 spec §4.1 / F-11 cover this static-vs-runtime split. rQr r[r\r^r])r_r[r^r])rTr.rC)rUr:rVrWs r(_fixed_elements_inventoryrp s{!#Cg S\ 66' (,E E  ,!#!<FF:r2    Jr'c g}|dk(r|jd|ds|jdt|iittiig|dS)uReturn the geometry-candidates shape with VLM-dependent slots initialized to empty containers. W20A intentionally does **not** populate camera_cell_candidates etc. from numbered_elements alone — position_hint is a label string and the user's no-literal-substring rule forbids parsing meaning out of it. The populated values arrive once the VLM 10x10 readback gate is wired up in a follow-up wave. For now the shape is declared so downstream W20B can rely on the keys being present. synthetic_placeholderzVLM 10x10 readback is synthetic_placeholder; camera / look-at / visible-units / visible-openings cell enumerations are left empty pending a real VLM wiring wave.rbu`no base_structural_unit markers — geometry candidates cannot be enumerated against zero units.) grid_sizecamera_cell_candidates_per_unit look_at_cell_candidates_per_unitview_cone_lens_enum_mapvisible_units_candidatesvisible_openings_candidates"wall_door_invalidation_diagnostics diagnostics)rClistr1r)rs structure vlm_statusrzs r(_fp_geometry_candidates_shaper~"sp  K,, 4 ^ $ 0 )_+-,.#'(?#@$&').0"  r'c .dt|dddddddgdS)uSynthetic-placeholder readback. W20A never invokes a real VLM. The gate decision is ``synthetic_pass`` — not ``ok`` — so downstream consumers can distinguish "VLM readback passed" from "no VLM readback was attempted". Wiring a real VLM provider would replace this function in a follow-up wave behind explicit approval; the gate decision then becomes ``ok`` / ``failed`` based on the VLM's observation set. rrNsynthetic_passz4VLM readback not invoked (W20A synthetic-only path).zwA real VLM provider must be wired and approved separately before this slot can transition out of synthetic_placeholder.)r+rsobserved_markersmissing_markers extra_markers confidence gate_decisionrz)r{rss r(#_vlm_readback_synthetic_placeholderrJs5*)_ ) B L   r'c g}|j|xsgD]V}t|ts|jd}t|tst|tsF|j |Xt t|S)aPull exact integer marker numbers from an overlay payload list field. Skips non-dict entries and any number that isn't already a strict int (no string-int coercion; the upstream W19B-1 validator already enforced integer-only on its way in). r_)r.r0r1boolintrCrTset)overlayfieldr:r5rVs r(_collect_overlay_marker_numbersrfsqCU#)r)%& ii! c4 3(<  3 * #c( r'c2i}tt|t|z}|D]U}|j|}|j|}g}||jd||jd|?|jdxsgD cgc]$} t | t rt | t s| &} } |jd} t | t rt | t s| } nd} |jdxsgD cgc]$} t | t rt | t s| &} } |jdxsgD cgc]$} t | t rt | t s| &}} t|d }t|d }t|d }t |jd d }ng} d} g} g}g}g}g}d }|g|jdxsgDcgc]}t |ts|}}|jdxsgDcgc]}t |ts|}}ng}g}||tt| | tt| tt||||||||d ||<X|Scc} wcc} wcc} wcc}wcc}w)u!Per-bg exact-ID facts the W20B shot-aware planner consumes. Joins the W19B-1 overlay payload (use/ignore arrays, base / transient marker partitions, target unit numbers, clean expectation) with the W19 background_master_plan bg spec (applies_to_shots, depends_on_bg) on ``bg_id`` exact-string equality. **No semantic inference** — integer / string equality only. Every bg_id seen on either side is surfaced once with a ``diagnostics`` list naming the missing side (if any). When overlay is missing the overlay-derived fields default to empty lists / None / False; when master_plan spec is missing the spec-derived fields default to empty lists. The W20B consumer can read the diagnostics to decide whether the bg has enough fact coverage to act on. Nz7missing floor_plan_overlay_payload entry for this bg_idz5missing background_master_plan bg spec for this bg_idtarget_unit_marker_numbers"dominant_target_unit_marker_numberuse_numbered_elementsignore_numbered_elementsbase_markers_to_reference)rtransient_markers_to_describeignored_state_overlay_markersclean_background_expectedFapplies_to_shots depends_on_bg) r@rLrrrr base_marker_numbers_to_reference$transient_marker_numbers_to_describe$ignored_state_overlay_marker_numbersrrrrz) rTrr.rCr0rrrr)rLoverlays_by_bgbg_specs_by_bg_idr: all_bg_idsr@rspecrzn target_units dominant_rawdominantuse_list ignore_listbase_marker_numstransient_numsignored_overlay_numscleansrrs r(_per_bg_render_factsrzs*&(CN+c2C.DDEJ $$U+ $$U+!# ?   I  <   G   "++&BCIrIIAa%jD.AI  #;;'KLL<-"<6*6"++&=>D"DDAa%jD.AD "++&@AGRGGAa%jD.AG  ?:  =>N$C>$ %@%HIELHHK! N#% E   HH%78>B> >a:aQTCU>  !HH_5;;;a 1c@R;  " M*0\1B*C2:%+CM%:(.s;/?(@0@4B4H). 0*& E If JK 4 s* )J)J)J J4JJ'Jcptd|jD}dg}|r||ddS|s|jdg|ddSddtfd|j Dtfd|jD}|jd t |d d ||dd d S)uSurface anchor candidates only — never pick. Per W20 spec §4.4 the anchor BG selection is LLM-owned. W20A surfaces the candidate set (every overlay whose ``clean_background_expected`` is True) and leaves selection to W20B. W20F5 fallback (Codex 2026-05-28 narrow wave): if no overlay has ``clean_background_expected=True`` but ≥1 overlay exists for the fp, fall back to the BGs with the fewest transient markers (ramped candidate set). The fallback is explicitly tagged in diagnostics so the LLM and downstream auditors know the candidate set was widened. Empty overlays_by_bg still produces empty candidates (fail-closed at the consumer). c3\K|]$\}}t|jddr|&yw)rFN)rr.).0r@rNs r( z,_anchor_candidate_surface..s04NE7  7? @ 4s*,z=anchor selection deferred to W20B (LLM-owned per W19J pivot).N)candidate_bg_idsselection_diagnosticsselected_anchor_bg_iduZno overlays exist for this fp — anchor candidate set is empty (fail-closed at consumer).c|jdxs|jdxsg}t|ttfr t |SdS)Nrrr)r.r0r{tuplerh)rNr/s r(_transient_countz3_anchor_candidate_surface.._transient_countsN ;< 2A   (e}=s5zD1Dr'c3.K|] }|ywNr&)rprs r(rz,_anchor_candidate_surface..s%<%K|]\}}|k(r|ywrr&)rr@rNr min_transients r(rz,_anchor_candidate_surface..s*4NE7 G $ 5 4sufno clean_background_expected=True overlay for this fp — W20F5 fallback widened candidate set to the z* BG(s) with the fewest transient markers (zF); LLM may still pick any of these but must declare the chosen anchor.T)rrrramped_fallback)rNDict[str, Any]returnr)rTr/rCminvaluesrh)rclean_candidatesrzrampedrrs @@r(_anchor_candidate_surfacers",224 HK 0%0%)   / !#%0%)  E %3%:%:%<M,224F  114V >))68@ @#!,!%  r'c |jdxsg}i}|D]+}t|trd|vr t|d} ||| <-t ||t|} | dstd|dt|D cgc]+} || jdtvrt| || -} } g} t|D]} || }|jdxsgD]d}t|ts t|jd} | j| | |jdd |jdd d fg}|s|jd ||jd |s|jd t|}t|| |d}t|}|Dcic]}|jds|d|}}t!|||}||t#|| ddgit%|t&ddgid||| | |||d S#tt f$rYwxYwcc} w#tt f$rY\wxYwcc}w)Nnumbered_elementsr_rbrRz\ has zero base_structural_unit markers; a dossier cannot be built without at least one unit.rQrr[r\)r@r_r[rQzino backgrounds reference this fp in background_master_plan; downstream BG consumers will produce nothing.ztfloor_plan_render did not yet emit a png_path for this fp; downstream consumers must wait for the render checkpoint.z{no floor_plan_overlay_payload entries reference this fp; the overlay step may be not_applicable upstream (v5 default path).rr+)rsr|r}r@)rLrrrzzQmaterials palette is an LLM-emit slot (W20B); W20A leaves this empty by contract.zQlighting identity is an LLM-emit slot (W20B); W20A leaves this empty by contract.)r| materialsfixed_elements_inventorystandard_of_living_bandlighting_identity) rL fp_image_pathrsdwelling_identityfp_geometry_candidatesfp_geometry_vlm_readbackbase_marker_inventoryoverlay_marker_inventoryper_bg_render_facts_by_bg_idanchor_selection_metadatarz)r.r0r1r TypeError ValueErrorrYrnr!rTrr`rCrr~rrr{rpr)rLfp_entry fp_png_pathrbg_specsrs raw_markersrUrWrVr|base_inventoryoverlay_inventoryr@rNr5rz vlm_readbackgeometryanchorrr per_bg_factss r(_build_one_dossierr.s,,239rK)+G"d#xr'9  bl#C UG, )I ^ $&UIA B  '?,"C 3<  1 26J J WS\2", /1' '[[!@AGRGEeT* %))H-.  $ $"!"YYw3+0995JB+O  H($ K  <  H   M 7KL,)H '~ 6F(04(0DHHW4EW t4(%+L$)_":  )B'(J'F: " "#+$0!/$5(4%+"7 O:&   ,z*  F4s5H 0H!1H&,H<H< HH&H98H9N)r9rsc t|}|s tdt|}t|}t |xsi}t |t rt|dk7rtd|td|Drtd|i} |jD]H\} } t| | |j| |j| i|j| g|| | <J| S)uBuild one dossier per fp_id present in the floor_plan_prompt cp. The iteration order matches the floor_plan_prompt cp insertion order. Each fp's dossier is independent — a per-fp failure raises ``BaseLocationDossierError`` and aborts the whole call; the step wrapper turns that into a ``failed_count=1`` checkpoint. Parameters ---------- fp_prompt_data: ``floor_plan_prompt`` checkpoint ``data`` block. master_plan_data: ``background_master_plan`` checkpoint ``data`` block. overlay_payload_data: ``floor_plan_overlay_payload`` checkpoint ``data`` block. Empty dict is acceptable (the per-fp dossier records a diagnostic). fp_render_data: Optional ``floor_plan_render`` checkpoint ``data`` block; used only to surface ``fp_image_path``. Missing render is non-fatal. grid_size: VLM readback grid size. Defaults to (10, 10) per W20 spec §2.4. z_floor_plan_prompt cp carries no floor_plans entries with status='ok'; cannot build any dossier.z!grid_size must be a 2-tuple; got c3LK|]}t|t xs|dkyw)rN)r0r)rvs r(rz!build_dossiers..s' ?Yz!S! ! +Q!V +Ys"$z0grid_size dimensions must be positive ints; got )rLrrrrrs) r6r!rIrOr;r0rrhanyr/rr.) r2rDrMr9rsfp_indexbg_specs_by_fpoverlays_by_fp fp_png_indexr:rLrs r(build_dossiersrs<".1H & 5  --=>N,-ABN+N,@bAL i '3y>Q+>&/ } =   ?Y ??&>ym L  &(C#>>+x'$((/)--eR8#''r2  E , Jr')r2rrDict[str, Dict[str, Any]])r9rrzDict[str, Optional[str]])rDrrzDict[str, List[Dict[str, Any]]])rMrrz$Dict[str, Dict[str, Dict[str, Any]]])rLrrUDict[int, Dict[str, Any]]rNone)rVrrWrrr)rUrrr)rUrrList[Dict[str, Any]])rsrr|rr}rrr)rsrrr)rrrrrz List[int])rLrrrrrrr)rrrr)rLrrrrz Optional[str]rrrrrsrrr) r2rrDrrMrr9zOptional[Dict[str, Any]]rsrrr)#r% __future__rtypingrrrrrr frozensetr__annotations__rrrrr Exceptionr!r6r;rIrOrYr`rnrpr~rrrrrrr&r'r(rstVn#>> (1(n+4 (+ ';=T&T^T%-?- +(<;6y6"&$$& ( )  2 4  &  F &2%%% %  %P8 '*(j j.j1 j  jZD-DDNs ss s . s # sssv04!2 ;";%;) ; - ;  ;;r'