jSrSSKJr SSKJrJrJrJrJr SSK J r J r SSK J r SSKJrJr SSKJr SSKJr SS KJrJrJr SS KJr S rSSS jjrSSjrSSjrSSjrg )u\Group the parse's floating text by the scan's block partition. A page like a stat-panel overview prints its content as short label and value lines. The parse captures every line, but as separate records — `$78B` is one record, `Investor capital` another, joined by nothing — and anything consuming the records one at a time hands a reader a label without its number. The scan's response partitions the same page into blocks: every line belongs to exactly one node, and the lines of one panel come back inside one node. This module carries that partition onto the parse's own elements. Each plain text item on a scanned page is assigned to the block whose rectangle holds its center, and the items of a block are re-parented under one group node in the document, in their reading order, at the first member's position in the flow. The items themselves are untouched: their text, boxes and references stay the parse's own, and nothing is transcribed from the scan. The scan contributes one fact the parse does not state — which lines sit together. The blocks come from the response's text nodes and from its table nodes. The scan reads a panel collage as a table as readily as it reads one as text, and a printed panel is the same panel either way. A table node contributes each of its cells as a block, so a value and its label bind at the cell, not across the whole collage; the node's own rectangle stands behind the cells for text that sits between them, and the smallest containing block always wins. The same division of authority the graft enforces for figures: the scan looked at the page, so it decides the partition; the parse's elements supply the content. Only plain text items parented to the document body take part. An item under a picture is the graft's business, a heading or a footnote or a caption has a role the partition must not swallow, and an item already inside a group is left alone, whether the parse made the group (a list, an inline group) or an earlier run did. That is why a rerun over an enriched parse changes nothing. A block holding fewer than two items gains nothing from a group and gets none. ) annotations)DictListOptionalSequenceTuple)DoclingDocumentTextItem)NodeItem) DocItemLabel GroupLabel)logger) PageDigest)NormBoxnorm_boxprov_box)PageScanzscan-text-blockNc [U=(d 05nURR5HIupVXT;dM URcMURRURR 4XE'MK SnUGHnUR UR5n U cM$SU R5V s/sH n U cMU PM n n U RHPn U RSU R55 [U R5n U cM?U RU 5 MR U (dMUR URS5up0n[XR5H@n[!UX5n [#X5nUcMUR%U/5RU5 MB ['U5HnUUn[)U5S:aMUR+US[,R.[0SS9nUHn[3UUU5 M US - n[4R6"S UR[)U5S R9S USS 555 M GM U$s sn f)zGroup each scanned page's plain text by the scan's block partition. Mutates `document`: the items of each block move under one group node at the first member's reading-order position. Returns how many groups were written. Nrc3n# UH+oRS:XdM[UR5v M- g7f)textN)kindrbox.0items %group_scanned_text..Ps)]~tV\I\***~s55c3^# UH#n[[U5H o"cMUv M M% g7fN)mapr)rrowrs rrrTs%k)9#SSVEWc##EW#)9s- -)g @g@F)siblinglabelnameafterz,Text blocks: page {} grouped {} item(s) ({})z / c3P# UHoR=(d SSSv M g7f)N)rrs rrrpsHidIIOSb1is$&)dictpagesitemssizewidthheightgetpagecontexttablesextend cell_boxesrrappend_floating_textr _owning_block setdefaultsortedlen insert_groupr UNSPECIFIED SCAN_BLOCK _reparentrinfojoin)documentscansdigests page_dimsdimspage_nor4writtenscandigestrblockstable table_boxr1r2membersrblock block_indexr/groups rgroup_scanned_textrU8s  R D!--/   499#8!YY__dii.>.>?DM0GTYY' > ^v~~] ] ]  ]]E MMk)9)9k k +I$ i( #  N; -/"8YY7D4/C!#.E ""5"-44T: 8 "'?KK(E5zA~))a (>(>ZW\*E(D%0 qLG KK> E  HeBQiHH  +1P NI s 9I$I$cURn/nURHnUR[R:wd(UR =(d SR 5(dMKUR(aURSRU:waM{UR(aURRU5OSnXRLaMURU5 M U$)a The page's plain text items parented to the body, in document order. Everything else keeps its place: an item under a picture belongs to the graft, a labelled item has a role, and an item already inside a group, the parse's own or an earlier run's, keeps its group. r*rN) bodytextsr%r TEXTrstripprovrJparentresolver9)rEr4rWfoundrr\s rr:r:us ==DE ::** *499?2I2I2K2K yyDIIaL00D8 26++$$X.4    T LcUcgUSUS-S- USUS-S- p2Sn[U5HFunupgpXbs=::aU::dMO MXss=::aU ::dM)O M-X- X- -n Ub XS:dMCX4nMH UbUS$S$)zrThe block whose rectangle holds the item's center; the smallest when blocks overlap; None when no block claims it.Nrr#r() enumerate) rrNcxcybestix1y1x2y2areas rr;r;s {!fs1vo "SVc!f_$9(,D(0 BB >r>>bn"nnG(D|t1g~y 1 &470D0r_cfUR(aURRU5OSnUb=URVs/sH oDRUR:wdMUPM" snUlURR UR 55 UR 5Ulgs snf)zMove one item under the group, keeping its identity and its order. Reference surgery rather than delete-and-re-add: the item keeps its self_ref, so nothing else in the document renumbers. N)r\r]childrencrefself_refr9get_ref)rErrT old_parentrefs rrBrBs 37++$$X.4J.8.A.A_.AsXXQUQ^Q^E^s.A_  NN$,,.)--/DK`s B.!B.r ) rEr rFzSequence[PageScan]rGzDict[int, PageDigest]rHz(Optional[Dict[int, Tuple[float, float]]]returnint)rEr r4rtrszList[TextItem])rzOptional[NormBox]rNz List[NormBox]rsz Optional[int])rEr rr rTr rsNone) __doc__ __future__rtypingrrrrrdocling_core.types.doc.documentr r !docling_core.types.doc.items.noder docling_core.types.doc.labelsr r logururquber.core.figures.dpt3.modelsrquber.core.figures.geometryrrrquber.core.figures.modelsrrArUr:r;rBr_rrs}"H#88E6B5CC.  ;? :: :#:8 :  :z* 1 "r_