W6j6SrSSKrSSKrSSKrSSKrSSKrSSKrSSKJr SSK r SSK J r \ Rr SSjrSSjrSSjrSS jrSS jrSS jr"S S \5rSrSrSrSrSrSSjrSSjrSrSrSrSSjr SSjr!Sr"SSjr#SSjr$Sr%Sr&Sr'S r(S!r)S"r*S#r+S$r,SS%jr-S&r.S'r/0q0\RbRd4S(jr3S)r4SS*jr5SS+jr6S,S-S.S/S0.r7S11r8S2r9S3r:S4r;SS5jrS8r?S9r@S:rAS;rBS<rCSS=jrDS>S?S@SA.rESBSCS@SA.rFSDrGSSEjrHSFrISGrJSHrKSSIjrLSJrMSKrNSLrOSSNjrPSMSO.SPjrQSQrRSSRjrSSSrTSSTjrUSUrVSSVjrWSWrXSSXjrYSYrZSZr["S[S\\5r\S]r]"S^S_\5r^S`r_Sar`SbraSScjrbSdrcSerdSfreSgrfSShjrgSiSjSkSl.rhSmSnSoSpSqSr.riSSsjrjStrkSurlSvrmSwrnSqroSxrpSyrqSzrrS{rsS|rtS}ruS~rvSSjrwSrxSrySrzSr{Sr|Sr}Sr~SrSrSrSrSrSrSSjrSSjrSrSrSSjrSrSrSrSSjrSrSr\"S\55rSSjrSSjrSrg)a Pushes a purchased car into DealerKit (Right Drive's DMS) as a real "Due In" stock record, with its photos and service history documents attached. Built against the real live app (rightdrive.app.dealerkit.uk), not guessed markup: see the 2026-08-21 CLAUDE.md entry for the reconnaissance this was built from. DealerKit is a Vue single page app whose form fields carry no name/id/placeholder attributes worth relying on (custom ui-textbox/ui-select components, several rendered as hidden duplicate instances at once), so every click here finds the one currently VISIBLE match rather than trusting a fixed selector index, the same defence used throughout this module. Scope, deliberately narrow for this first build (2026-08-21, Mark: "lets do a test to confirm everything works first then automatic"): - Creates the due in record from the registration alone (DealerKit's own DVLA lookup fills in make/model/spec). - Types BidBrain's own purchase price into the Purchase Details step's Purchase Price field before Save. DealerKit has its own live Motorway integration that can fill this in (and a full itemised Buyers Premium / Delivery Charge / Indemnity Fee breakdown, plus real documents) entirely on its own, proven live twice, but ONLY ever seen firing on a genuinely human driven session (Mark himself, mouse and keyboard, twice) and NEVER on an automated one (three separate attempts: two headless script runs and one slow, deliberate, click by click session driven by Claude through the Browser pane, the last one checked again 15+ minutes later and still blank). See the 2026-08-21 evening CLAUDE.md entries for the full pattern. So for an automated push, typing the figure in ourselves is the only way it does not sit blank forever, confirmed with Mark before re adding this (the same mechanism was built, proven to work, then removed earlier the same day on the mistaken theory that DealerKit's own integration made it redundant for every kind of session, not just a human one). Only the single combined Purchase Price is set (BidBrain's own already computed total), not an itemised Buyers Premium / Delivery Charge / Indemnity Fee breakdown, that needs a separate Add Expense flow not yet investigated. Nothing else on the Stocklisting step is touched (Bought From, invoice ref, VAT qualifier, finance settlement, funding, sales location, Retail Price): Retail Price is Mark's own market call, typed in by hand, and "Bought From" needs matching or creating a DealerKit contact, not attempted here. - Uploads up to 4 real exterior photos, and any service history document and V5C document photos already captured, via DealerKit's own Edit > Images / Edit > Documents upload widgets (a plain multi file input each, drag and drop or a file picker, unrelated to the two despite the shared UI chrome). Service history and V5 photos go up together as one Documents upload (2026-08-23, Mark: "add the v5 to the automation"). Still never the seller's driving licence, which stays uncaptured everywhere in this project (see motorway.read_v5_photos), and still never captured at all for a listing that has not actually been bought, the V5 reader only ever runs against a genuine purchase's own summary page in purchases_run.py. Never guesses: if a step's expected element is not found, this raises loudly rather than clicking something else and leaving a half filled record. N)urlparse)platesc[R"5U-n[R"5U:aUR5n[U5HTnUR U5R 5nU(dM+USS:dM6USS:dMAUR U5U4s $ [R "S5 [R"5U:aMg![a SnNf=f)aThe first match of a locator that is actually rendered on screen right now. DealerKit renders several hidden duplicate instances of the same modal/component at once, so `.first` or a fixed index is not reliable, only the real bounding box tells you which one a human would see.rwidthheight?NN)timecount Exceptionrangenth bounding_boxsleep)locatortimeoutdeadliner ibs 9/Users/stevendouglas/BidBrain/bidbrain/dealerkit_write.py_visiblerUs yy{W$H ))+  MMOEuA A++-AqQwZ!^( a{{1~q(( 4 ))+   E sC C'&C'c[X5up4U(dgURRUSUSS- -USUSS- -5 g)NFxryrT)rmouseclick)pagerrelrs r_click_visibler!hsM W &EB JJQsVaj1n,afq{Q.FG c[U[5(aURU5O U=(d Un[XR XS9U5$)awithin, if given, is a locator (or selector string) to scope the text search to, for example the currently open modal. Needed because _visible() only checks that an element has a real bounding box, not that it is actually inside the viewport: a same-worded heading lower down the page (found live, 2026-08-23: a "Documents" section further down a vehicle's own page, unrelated to the Edit Vehicle modal's own "Documents" tile) can still have width/height > 0 while sitting off screen, and DOM order alone decided which one won, sometimes the wrong one. Scoping to the open modal removes the ambiguity outright rather than trying to out-guess which "visible" match a human would pick.exact) isinstancestrrr! get_by_text)rtextr%rwithinroots r_click_visible_textr,ps?$.fc#:#:4<< 4D $ 0 0 0 CW MMr"c$[R"5U-n[R"5U:a_UbURX5OURU5nU(aU$[R"U5 [R"5U:aM_g)aLike page.evaluate, but retries until it returns a truthy value or timeout runs out. A fixed page.wait_for_timeout before a ONE SHOT evaluate (the pattern used throughout this module before this helper existed) was found live 2026-08-23 to be genuinely racy for set_expense_tax_date specifically: two back to back runs against the exact same vehicle, same code, same 1500ms wait, landed in two different real states (one correctly inside "Edit Expense #NNN", one still sitting on the outer "Choose an area to edit" menu), meaning DealerKit's own SPA transition time genuinely varies run to run, not something a single fixed wait can cover reliably.N)r evaluater)rjs_fnargrpollrresults r_evaluate_untilr3siyy{W$H ))+ .1ou*4==QVCW M 4 ))+ r"c[R"SSUR55n[R"5U-n[R"5U:aUR SU5nU(a:USnUR R USUSS- -USUS S- -5 g [R"S 5 [R"5U:aMg ) aClick the smallest visible element whose text contains needle (case/space insensitive). Used wherever an exact text match is not reliable, for example a search result card whose wrapping element has no useful class or attribute of its own, or the EDIT button whose visible label sits next to a material icon ligature in the same text node ("mode_edit\nEDIT"), so no element's own text is ever exactly "EDIT" and get_by_text(..., exact=True) never matches anything.z\s+a(target) => { const els = document.querySelectorAll('a, div, li, button, span'); const out = []; for (const el of els) { if (el.children.length > 20) continue; const t = (el.innerText || '').toUpperCase().replace(/\s+/g, ''); if (!t.includes(target)) continue; const r = el.getBoundingClientRect(); if (r.width > 0 && r.height > 0) { out.push({x: r.x, y: r.y, w: r.width, h: r.height, area: r.width * r.height}); } } out.sort((a, b) => a.area - b.area); return out.slice(0, 3); }rrwrrhT333333?F)resubupperr r.rrr)rneedlertargetrmatchesms r_click_containingr@sVVFB  /Fyy{W$H ))+ --  ! $  A JJ  QsVafqj0!C&1S6A:2E F 3/ ))+ 0 r"cUcg[U[5(a UR5(a [U5OUnSn[R"5U-n[R"5U:aUR SU[ U5/5nU(aURSUS35nUR5 UR[U55 UR5 URS5 UR5n [U5[U 5;agg[R"S5 [R"5U:aMg) u Type value into the Purchase Price (£'s) field on the Stocklisting step. That field carries no name/id/placeholder, only a label div naming it, so the real input is located by finding the label's own position on screen and then the geometrically nearest visible text input to it, the same proven pattern _add_expense_items' own "Unit Net" field already uses (2026-08-23), rather than walking up the DOM from the label to the nearest ancestor holding an input (the original approach, found live 2026-09-05 to grab the wrong field once DealerKit added a "Purchase Invoice Ref #" field just above this one). A second bug found the same day, once the position based rewrite above still failed: `el.children.length <= 3` is not enough on its own to tell a real, short label apart from a huge ancestor wrapper. This page's own outer content container (everything on screen, the whole vehicle list plus the open modal) has only 1 to 3 DIRECT children of its own, so it passed that check too, and its own innerText, which is the WHOLE PAGE's text, genuinely contains the words "Purchase Price" buried somewhere inside it, thousands of characters in. Since it comes earlier in raw DOM order than the real, specific 20 character label and both report a real, nonzero bounding box, the search's own `break` fired on that huge wrapper first. A short text length cap, the same guard already used for the Unit Net field (`.length > 20` there), rules this class of false match out for good: a real field label is a handful of words, never a whole page. A THIRD bug found the same day, once both of the above were fixed and the right input was finally being typed into correctly (input.value confirmed landing before Save): DealerKit's Save request itself still carried purchase_cost: null. Captured live via a full request body dump. This is a Vue component whose own reactive form state only syncs from the raw DOM value on a genuine blur, not just on keystroke events, so typing via raw mouse and keyboard events (which never blurs the field before Save is clicked) leaves the framework's own internal model unaware the value ever changed, even though the DOM input.value itself reads back correctly, which is exactly why the earlier "landed" confirmation kept reporting success. Fixed by switching to Playwright's own locator.fill() plus an explicit .blur(), the same proven, already working pattern set_retail_price uses for this identical class of problem on a different DealerKit panel. Finds the input by marking it with a temporary, unique attribute (rather than raw mouse coordinates) so Playwright's own Locator API, not manual mouse and keyboard events, drives the actual fill and blur. Reads the DealerKit-visible value back afterward, via the same marker, to confirm the figure genuinely landed before ever clicking Save; never raises, a field that cannot be confirmed is left for Mark to fill by hand rather than risk a wrong or silently empty figure.Fdata-bb-purchase-pricea([marker, labelKeys]) => { const els = document.querySelectorAll('div, span, label'); let label = null; for (const el of els) { if (el.children.length > 3) continue; const raw = el.innerText || ''; if (raw.length > 40) continue; const t = raw.toUpperCase().replace(/\s+/g, ''); if (!labelKeys.some(k => t.includes(k))) continue; const r = el.getBoundingClientRect(); if (r.width > 0 && r.height > 0) { label = {x: r.x + r.width/2, y: r.y + r.height/2}; break; } } if (!label) return false; const inputs = Array.from(document.querySelectorAll('input[type=text]')); let best = null, bestDist = Infinity; for (const inp of inputs) { const r = inp.getBoundingClientRect(); if (r.width === 0 || r.height === 0) continue; const cx = r.x + r.width/2, cy = r.y + r.height/2; const dist = Math.hypot(cx - label.x, cy - label.y); if (dist < bestDist) { bestDist = dist; best = inp; } } if (!best) return false; document.querySelectorAll('[' + marker + ']').forEach(e => e.removeAttribute(marker)); best.setAttribute(marker, '1'); return true; }[],Tr )r&float is_integerintr r.listrrfillr'blurwait_for_timeout input_valuer) rvaluer label_keysmarkernumberrmarkedfieldlandeds r_fill_purchase_pricerUs d }%eU338H8H8J8JSZPUF %Fyy{W$H ))+  6T*% &9 < LL1VHA/E KKM JJs6{ # JJL  ! !# &&&(F6{c&k) 4S ))+ T r"c\rSrSrSrSrg)DealerkitMissingMileageiaRaised by add_due_in when it is about to create a genuinely NEW Due In record with no known mileage to give it (2026-08-26, Mark: "never create a listing in DK unless we have the mileage as this causes DK to estimate the mileage which is dangerous"). Only ever raised on a first push, an already existing record is returned early before this check ever runs ("this is only for the first push if its already present in DK dont over write"), matching every other hard gate in this project, play safe rather than let a wrong, DealerKit guessed mileage sit on a real business record. There is NO bypass (Steven 2026-08-27, "NEVER ALLOW Dealerkit to manually guess the mileage"): the old allow_missing_mileage escape hatch is gone, the fix is always to enter the real mileage on the purchases page first.N__name__ __module__ __qualname____firstlineno____doc____static_attributes__rXr"rrWrWs 5r"rWcURS5=(d SR5nU(aU$URS5S:XaS$S$)aThe supplier to attribute a car's price to on DealerKit, when nothing more specific applies: the row's own recorded supplier where one was set (2026-08-31, a hand added car's own supplier wins so it is never wrongly filed against Motorway), else CarWow or Motorway by platform. Shared by add_due_in's own new Bought From selection (2026-09-05) and push_purchase's "already exists, correct a price" branch, which used to compute this same fallback inline, on its own.supplierr5platformCarwowCarWowMotorway)getstrip)purchaseras r_default_supplierri+sC Z(.B557H||J/8;8KKr"c ^^ [TS5nU(d [S5eTRRUSUS5 TR S5 U4SjnTR S5nU(d [S5eTR S 5nUR5 URU5 UR5 TR S 5 U"S 5nU(d [S 5eTRRUSUS5 TR S 5 [TS5 U"S5nU(d [S5eTRRUSUS5 TR S 5 [TS5 0m U 4SjnTRSU5 TR S5n U (dTRSU5 [S5eTRRU SU S5 TR S5 TRSU5 T RS5S;aQ[SU<ST RS5(a([R"T RS55SS OSS35eg)a<Creates a new DealerKit contact for a car's own supplier, via the real "+" button beside the Stocklisting step's "Bought From" field (2026-09-05, Steven: "the fix needs to be that it actually adds the supplier first ... so it works", so pressing Price on a genuinely new supplier never again needs someone to go and create the contact in DealerKit by hand first). The "+" button is found by its own visible text ("add", the material icon's own ligature) rather than a class, since a page wide class search for the same green button style landed on a COMPLETELY different, unrelated "+ Add Invoice Item" button once the Bought From field's own dropdown had already been opened (found live), and the two happened to share a screen position at that moment. Checked scoped by its own nearby text ("Bought From") instead, found reliable across several fresh attempts. Opening it reveals a real "New Contact" modal STACKED on top of the still-open Stocklisting one (both report the same z-index, so a naive "first visible dialog" check finds the WRONG one, the one underneath; scoped to `.contact-modal` specifically here). Company Name and Contact Type (set to "Supplier") are the two fields this actually needs; Referral Source LOOKS optional (Save stays enabled with it blank) but is not, DealerKit's own server refuses the save with a real 422 ("The referral source id field is required.") that the button's own enabled state gives no hint of, found by checking the real network response rather than trusting the click. "Other" is used there, a neutral choice for a contact recorded purely as a supplier, not a marketing lead. Verifies via the real save response (never a guess): raises loudly if DealerKit refuses it, for whatever reason, rather than reporting a false success and leaving the caller to type a price with nowhere real to attribute it.a() => { const btns = Array.from(document.querySelectorAll('button')); for (const b of btns) { const r = b.getBoundingClientRect(); if (r.width === 0 || r.height === 0) continue; const icon = b.querySelector('i, .material-icons'); const text = (b.innerText || icon?.textContent || '').trim(); if (text !== 'add') continue; const node = b.closest('.row') || b.parentElement?.parentElement; const ctx = node ? (node.innerText || '') : ''; if (!ctx.includes('Bought From')) continue; return {x: r.x + r.width / 2, y: r.y + r.height / 2}; } return null; }z@DealerKit: the Bought From + (add contact) button was not found.rrc(>TRSU5$)NaP(label) => { const modal = document.querySelector('.contact-modal'); if (!modal) return null; const labels = Array.from(modal.querySelectorAll('.ui-select__label-text')); for (const l of labels) { if (l.textContent.trim() !== label) continue; const container = l.closest('.ui-select'); const r = container.getBoundingClientRect(); return {x: r.x + r.width / 2, y: r.y + r.height * 0.75}; } return null; })r.)labelrs r _field_pos,_create_supplier_contact.._field_posrs}}    r"a() => { const modal = document.querySelector('.contact-modal'); if (!modal) return false; const labels = Array.from(modal.querySelectorAll('.ui-textbox__label-text')); for (const l of labels) { if (l.textContent.trim() !== 'Company Name') continue; const input = l.closest('label').querySelector('input'); input.setAttribute('data-bb-new-contact-name', '1'); return true; } return false; }zDDealerKit: the New Contact modal's Company Name field was not found.z[data-bb-new-contact-name]rEz Contact TypezDDealerKit: the New Contact modal's Contact Type field was not found.SupplierzReferral SourcezGDealerKit: the New Contact modal's Referral Source field was not found.Otherc>SUR;a?URRS:Xa$URTS'UR 5TS'ggg![ a gf=f)Nz /api/contactsPOSTstatusbody)urlrequestmethodrujsonr )respr2s r_on_save*_create_supplier_contact.._on_saves_ dhh &4<<+>+>&+H #';;x !%v,I &  s"A AAresponsea() => { const modal = document.querySelector('.contact-modal'); if (!modal) return null; const footer = modal.querySelector('.ui-modal__footer'); const btn = footer ? footer.querySelector('button') : null; if (!btn) return null; const r = btn.getBoundingClientRect(); return {x: r.x + r.width / 2, y: r.y + r.height / 2}; }z=DealerKit: the New Contact modal's Save button was not found.ruz DealerKit refused to create the z contact: rvNzno response captured.)r3 RuntimeErrorrrrLr.rrJrK_click_dropdown_optiononremove_listenerrfrzdumps) r company_nameposrnrRrStype_pos referral_posr|save_posr2s ` @r_create_supplier_contactr9sDD    C$ ]^^JJSXs3x($ "]]  F abb LL5 6E KKM JJ| JJL#.)H abbJJXc]HSM2#4,/0L deeJJ\#& S(9:#4) F GGJ!}}  H  Z2Z[[JJXc]HSM2$X. zz(:-.|.>j7=zz&7I7Itzz&**V,-ds3Oeffg i  .r"cB^^UU4SjnU"5nU(d>[TT5 TRS5 TRST5nU(agU"5nU(d[ST<S35eTRR USUS5 TRS5 g) a.Picks an EXISTING contact in the Stocklisting step's own "Bought From" field (2026-09-05, Steven, testing by hand: "you need to add a supplier when adding a car before the purchase price will save", found after three other real bugs in this same day's session had already been fixed and Purchase Price still would not persist). "Bought From" opens the same .ui-select component family as the Add to Funding modal's own fields (found live, so _click_ui_select_field, already proven for that modal, opens it correctly), but this is a live CONTACT SEARCH, not a plain static list: opening it reveals a real `.ui-select__search-input` (placeholder "Start typing to search") with NO options showing at all until something is typed, so the already existing _click_dropdown_option (built for the Funding modal's own plain, always-populated list) finds nothing here. Once a name is typed, each real result is a multi-line `li.ui-select-option` (contact name, then its postcode, then its own category, e.g. "Motorway EC1A 2BN Supplier"), never a bare single-line match, so the option is matched on its FIRST LINE (the contact's own name) rather than the whole cell's text. Never creates a new contact (the green + button beside this field is left alone, same as it always has been); raises if contact_name has no matching result after typing, so the caller can fall back to leaving Purchase Price for a person to enter rather than typing a figure that will only look saved, never genuinely persisting.c>[TS5 [TS5nU(d [S5eTRR USUSS- -USUSS- -5 TR R T5 [TS TS S 9$) Nz Bought Froma`() => { const els = document.querySelectorAll('.ui-select__search-input'); for (const el of els) { const r = el.getBoundingClientRect(); if (r.width > 0 && r.height > 0) return {x: r.x, y: r.y, w: r.width, h: r.height}; } return null; }z4DealerKit: the Bought From search box was not found.rr6rrr7a(name) => { const opts = Array.from(document.querySelectorAll('.ui-select-option')); for (const el of opts) { const norm = s => (s || '').toLowerCase().replace(/[^a-z0-9]/g, ''); const first = (el.innerText || '').split('\n')[0].trim(); // "Carwow" on DealerKit, "CarWow" here: the same // contact (2026-09-16, the reason a price could be // left untyped at creation). if (norm(first) !== norm(name)) continue; const r = el.getBoundingClientRect(); if (r.width > 0 && r.height > 0) return {x: r.x + r.width / 2, y: r.y + r.height / 2}; } return null; }@r)_click_ui_select_fieldr3rrrkeyboardtype) search_box contact_namers r _try_pick&_select_bought_from.._try_pickst]3$    UV V C:c?Q+>> 3R\]`RadeRe@ef <(   #  r"al(name) => { const labels = Array.from(document.querySelectorAll('.ui-select__label-text')); for (const l of labels) { if (l.textContent.trim() !== 'Bought From') continue; const container = l.closest('.ui-select'); const r = container.getBoundingClientRect(); if (r.width === 0) continue; const disp = container.querySelector('.ui-select__display-value'); return !!disp && disp.textContent.trim() === name; } return false; }Nz(DealerKit: no Bought From contact named z was found.rr)rrLr.rrr)rrrrselecteds`` r_select_bought_fromrs4# J +C  !|4 c"==     k ElEUU`abbJJSXs3x(#r"cV[UR5RnSR UR 5=(d SR 55SSnURSU3U(aSU3-$S-$![a URnNtf=f![a SnNLf=f)zxOne line on a DealerKit reply that said no: the status, the path and the first words of the body, for a failure message. r5Nz on : )rrwpathr joinr)splitru)r{rrvs r _reply_wordsr2s!&&xx*1134Tc:kk]$tf %2dV EE" EE xx s"A=9B=BB B('B(cXURS5=(d S$![a gf=f)zWhat the open DealerKit window is showing right now, in DealerKit's own words: its error or warning lines when it has any, otherwise the whole window's text. Never raises, an empty string when nothing can be read.aG() => { const opens = document.querySelectorAll('.ui-modal.is-open'); const m = opens.length ? opens[opens.length - 1] : document.body; const picks = m.querySelectorAll('[role=alert], .alert, .error, .invalid-feedback, .text-danger, .has-error, [class*=error], [class*=danger], [class*=warning]'); const parts = []; for (const el of picks) { const t = (el.innerText || '').trim(); if (t && !parts.includes(t)) parts.push(t); } if (parts.length) return parts.join(' '); return (m.innerText || '').trim(); }r5r.r rs r _modal_wordsr@s; }}   s  ))cSRU=(d SR55SUnSRSU=(d /55SUnSUS3nU(aUSU3- nU(aUS U3- nU(d U(dUS - nU$) zThe sentence a send raises when DealerKit neither created the car nor said it already had it (2026-09-06). Carries DealerKit's own words, from the window and from its server, so Dealer OS's timeline says why instead of "the page may have changed".rr5Nz; c36# UHo(dM Uv M g7fNrX.0rs r +_no_confirmation_message..Ys9NaqN  zDealerKit did not add z& after its registration was submitted.z The Add Vehicle window says: z DealerKit's server replied: z0 No reason was shown. The page may have changed.)rr)regsaidrefusalscaprepliesmsgs r_no_confirmation_messagerSs 88TZR&&( )$3 /Dii9HNN99$3?G "3%'M NC  /v66 .wi88  AA Jr"c ^^[X[U55nU(aSSSUS.$U(aURS5(d[US35eUR [ S3SSS 9 UR S 5 [XRS 55(d [S 5eUR S 5 [URS55upVU(d [S5eURRUSUSS- -USUSS- -5 URRU5 UR S5 SnU=(d 0RS5nU(a[URS5SS9upU (atURRU SU SS- -U SU SS- -5 URR[![#U555 UR S5 SnSSUS.m/mUU4Sjn UR%SU 5 ['USSS9(d[SUS 35e[(R("5S!-n [(R("5U :aTTS"(dJTS#(d@UR S$5 [(R("5U :aTS"(d TS#(dM@TS"(aUR S%5 UR+SU 5 TS"(dSTS#(dIS&n [X[U55n U (a STS#'U TS''O[[1U[3U5T55eTS"(a0U(a)TRS'5(a[5UTS'5TS('TS"(GaGS&nU(aDU(a=URS)5nUc)URS*5=(d URS+5nSnS&nUbv[7U5nS&n[9S5Hn[;UU5 Sn O U(dAS,U<S-3U(a.S.S/R=[!U5R?55S&S03OS1-nS&n[AX5nSnUb U(dS2nU(Ga[CU[D5(a)[EU5RG5(a [#U5OUnTRS'5n[9S5HnURS3S4S59nSn[9URI55H$nURKU5RS S9 Sn O U(d OKUR S65 U(a-[MU[ S7U35nURS85bSn OMSn O U(a U(dS9nU(d['US:SS;S<9 UR S=5 UTS>'UTS?'U(a[OS@USA35 T$STS>'S&TS?'T$!UR+SU 5 f=f![,a e[.a S&n GNf=f![.a S&TS('GNf=f![.anUnS&nAGMDS&nAff=f![.a GMQf=f)Ba Create the vehicle as a Due In stock record from its registration alone. Returns {"created": True} for a genuinely new record, or {"already_exists": True} when DealerKit already has this reg in its stocklist (real state, not an error, someone on the team may have added it by hand already). Raises if neither is confirmed. purchase is used two separate ways, deliberately never coupled: its mileage always gates whether a genuinely new record may be created at all (see DealerkitMissingMileage below), regardless of what is being pushed, since photos or documents alone still need a real vehicle_id to upload to and so still create the record; type_price (2026-08-26, a real bug caught building the mileage gate, a "photos only" push used to pass purchase=None specifically to suppress typing the price, which would have also suppressed the mileage check on a car that genuinely had one on file) separately controls whether purchase's own price actually gets typed into the Purchase Price field, matching push_purchase's own "only price gates the typed figure". Raises DealerkitMissingMileage before ever opening the Add Vehicle modal when this would be a genuinely new record and purchase carries no mileage. No bypass exists, see the class docstring above. When purchase is given and this is a genuine new record, types BidBrain's own already computed purchase price (winning_bid or price) into the Purchase Price field before Save. DealerKit has its own live Motorway integration that can fill this in on its own (proven live twice, see CLAUDE.md), but only ever seen firing on a session Mark drove himself by hand, never on an automated one (three separate attempts, one waited on for 15+ minutes, all blank), so an automated push types the figure in itself rather than leave it sitting empty forever. Retail Price is left untouched, that is Mark's own market call. Nothing else on the Stocklisting step is touched either (Bought From needs matching or creating a DealerKit contact, not attempted here). When purchase is not given, or its price could not be confirmed landing in the field, just clicks Skip, same as leaving it for Mark. Checks DealerKit's own stocklist for this reg FIRST, before ever attempting to create anything (2026-08-24, found live the hard way: a Carwow purchases batch created 13 genuine DUPLICATE Due In records for cars DealerKit already held as a completed, disposed sale, life_cycle_status 25). The bug was trusting DealerKit's own POST /api/stocklist response alone to detect "this reg already exists" (a 422): that check IS real and does fire for a reg still active in some other stage, but DealerKit does NOT return it for an already disposed one, it just creates a second record instead, no error at all. A reg lookup first (_find_vehicle_id_by_reg, already used elsewhere in this module) closes that gap regardless of the vehicle's own status, and is also simply cheaper: a genuine repeat call for an already pushed reg no longer even opens the Add Vehicle modal.FT)createdalready_existspurchase_price_filled vehicle_idmileagez has no mileage on file. Creating it on DealerKit now would let DealerKit estimate the mileage itself, which is never allowed. Enter the real mileage on the purchases page first. /vehiclesdomcontentloadedȯ wait_untilrrz a.pill-greenz^DealerKit: the + Vehicle button was not found on the vehicles list. The page may have changed.z e.g. EO67 GJVzdDealerKit: the registration field was not found in the Add Vehicle modal. The page may have changed.rrrrrXz-.ui-modal.is-open input[placeholder*="43000"]rrrp)rr mileage_typedc*>SUR;aURRS:XaURS:XaSTS'gURS;a[STS'UR 5nUR S5=(d) UR S 5=(d 0R S5TS 'gTR[U55 gS UR;a,URS :aTR[U55 ggg![ a gf=f) Nz/api/stocklistrtiTrrriddatarz/api/rp) rwrxryrurzrfr appendr)r{rvrr2s r _on_response add_due_in.._on_responses txx 'DLL,?,?6,I{{c!+/'( *$(y! 99;D+/88D>+_dhhv>N>TRT=Y=YZ^=_F<( T 23  T[[C%7 OOL. / &8 !sAD DDr~Nextr$z6DealerKit: the Next button was not found after typing rrrri Nrmileage_on_record car_price winning_bidpricezthe Bought From contact z could not be picked, rrr5z8the Purchase Price box was not found on DealerKit's formbuttonSavehas_text/api/stocklist/ purchase_costz2the figure was typed but DealerKit did not save itSkip@r%rrkr price_reasonz) Purchase price at creation: not saved, z&; it goes on the purchase expense next)(_find_vehicle_id_by_reg other_platerfrWgotoDEALERKIT_BASErLr!rrrget_by_placeholderrrrrr'rHrr,r rDealerkitSessionExpiredr rr_vehicle_mileagerirrrrrUr&rFrGr r _get_jsonprint)rrrh type_price existing_idreg_elreg_boxrmilesmil_elmil_boxrrlate_idrbought_from_setrralast_err_etypedfilledr=new_idbtnsclickedrrrr2s @@r add_due_inrdsBd*$[5JKK DSX)+ + i00%e< <   II *7ISXIY$ $ ^ < = = )   $t66GHOF  )   JJWS\GG$4q$88'#,QYIZ]^I^:^_MMs#M ^  +E "4<<0_#`jmn  JJ  WS\GG,tzz|,A! ))$)7"& -%%d+$Tn-=_VH+UVDxx0 Pricing opens a panel whose big central price display is genuinely a plain , found by searching for that exact placeholder text once a plain "RETAIL" text search only ever found the unrelated AutoTrader valuation figure sitting elsewhere on the same panel; the label only exists as an input placeholder here, never as real text content anywhere in the DOM. Typing into it fires a real, harmless GET to DealerKit's own pricing-insights endpoint (the live "AutoTrader price position" indicator recalculating), proof the field itself is genuinely wired up before Save is ever clicked. Saving needs Playwright's own real .click() with its full actionability checks, not this module's usual _click_visible_text (a raw mouse click at a bounding box centre): this panel renders SEVEN "Save Changes" buttons at once, and unlike every other duplicate-button case elsewhere in this file, more than one of the hidden ones still reports a real, non-zero bounding box, the one thing _click_visible_text's own visibility check relies on, so it silently reported success while actually clicking a dead duplicate (confirmed live: no PATCH request at all reached DealerKit's API that way). Tries each match in turn with a real, strict .click() until one genuinely succeeds. Verifies via the API afterward, retrying once, raising loudly rather than reporting a false success, the same lesson every other DealerKit write in this project has already learned the hard way.api /vehicles/ /overviewrrr EDIT2DealerKit: the Edit button was not found (vehicle ).rPricingT.ui-modal.is-openr%r*z3DealerKit: the Pricing tile was not found (vehicle u£RETAILz9DealerKit: the Retail Price field was not found (vehicle rc>TRSSS9n[UR55H"nURU5R SS9 g g![ a M4f=f)Nr Save ChangesrrrTFrrr rrr rrrs r _try_save#set_retail_price.._try_savesd||H~|>tzz|$A  !!$!/%   A A A rrr retail_pricescreenz:DealerKit: Retail Price did not actually save for vehicle r)set_retail_price_apirrrLr@rr,rrrHroundrrJr'rKrrrf) rrrrSboxr=r rrs ` rset_retail_pricersLD*44II :,i@M_inIo$ T6 * *OPZ|[]^__$ tYdCV WPQ[P\\^_``$$11*=>JE VWaVbbdeff u F KKM JJs6{ JJL#1X ;;  ! !$ 'Tn%5_ZL#QRDxx'61  ST^S__`a bbr"cU(dg[RRURSS55nU[R"SS9-R 5R 5U:H$![a gf=f)aDealerKit stores delivery_on as UTC midnight of the LOCAL (UK) day (confirmed live 2026-08-26: setting it to "2026-08-28" read back as "2026-08-27T23:00:00+00:00" during BST), so a plain stored_iso[:10] comparison is off by a day whenever the UK is on summer time. Adding a small fixed buffer before taking the date is safe regardless of which offset (GMT +0 or BST +1) applies, since the UK is never more than an hour ahead of UTC, unlike branching on the stored hour, which would need to know the exact DST boundary date to get right.FZz+00:00)hours)datetime fromisoformatreplace ValueError timedeltadate isoformat) stored_isodate_isodts r_delivery_date_matchesr su     , ,Z-?-?X-N O ##!, , 2 2 4 > > @H LL s/A// A<;A<c^[T[SU35n[URS5U5(agTR [SUS3SSS9 TR S 5 [ TS 5(d[S US 35eTR S 5 [TSSSS9(d[SUS 35eTR S 5 [TS5 TR S5 [TUS5 U4Sjn[S5HYnU"5(dMTR S5 [T[SU35n[URS5U5(dMY g [SUS35e)aDealerKit's own "Due In / In-stock On" date, on the vehicle's Miscellaneous edit panel (2026-08-26, Mark found it live: "on the car listing page go to edit then miscellaneous... the due in / in-stock date it there"). Its internal API field is delivery_on, the exact field an earlier session (see CLAUDE.md, 2026-08-23) concluded had NO editable UI path anywhere, checked at the time under Description, Additional Details, and this same Miscellaneous tab's Location and Delivery Details section, and even tried once as a raw API write, refused by Claude Code's own safety check as untested and invasive. That conclusion was wrong, or the field is newer, either way it is real and working now, found by Mark navigating the live UI himself rather than guessed at again. date_iso should be BidBrain's own collection_date (Motorway's real "Collected"/"Collection date" timeline read, or Carwow's own "Collection date" card, see motorway.read_collection_date and carwow.read_collection_date), the date the car is due at the dealership. No op, returns False without touching anything, if the stored date already matches (see _delivery_date_matches). Built the same way as every other DealerKit date write in this module: a real UI flow, never a raw API PATCH. Opens the field via _click_ui_select_field (the same .ui-datepicker__label-text component the Add to Funding modal's own date fields use) and picks the day via _pick_calendar_date. The Miscellaneous panel's own SAVE button is one of 24 elements on the page sharing that exact text, almost all hidden duplicates from other Edit Vehicle tiles rendered off screen (the same shape already seen and solved for Pricing's own "Save Changes", 7 of which report a live but dead bounding box), so every match is tried in turn with a real .click() until one genuinely succeeds, then verified via a fresh API read, retrying once, raising loudly rather than reporting a false success.r delivery_onFrrrrrrrrrr MiscellaneousTrrz9DealerKit: the Miscellaneous tile was not found (vehicle zDue In / In-stock On c>TRSSS9n[UR55H"nURU5R SS9 g g![ a M4f=f)NrSAVErrrTFrrs rr $set_delivery_date.._try_savesd||Hv|6tzz|$A  !!$!/%   r rrzBDealerKit: Due In / In-stock On did not actually save for vehicle r) rrr rfrrLr@rr,r_pick_calendar_dater)rrrexistingr rrs` rset_delivery_dater*siB.!1 MNHhll=98DDII :,i@M_inIo$ T6 * *OPZ|[]^__$ t_DI\ ]VWaVbbdeff$4!78#h(>?1X ;;  ! !$ 'Tn%5_ZL#QRD%dhh}&=xHH  [\f[gghi jjr"cf^[R"5U-nS=pV[R"5U:a/nU4U4SjjnURSU5 UR[STS3SSS9 UR S 5 UR SU5 UH6n U R 5n U RS 5nU RS 5n O UcUbXV4$[R"U5 [R"5U:aMXV4$!UR SU5 f=f![a Mf=f) aPoll DealerKit's own stocklist API (not the UI, which masks these figures behind a Show toggle) for up to timeout seconds, waiting for its own Motorway integration to fill in purchase_cost and retail_price after a fresh due in creation. Returns (purchase_cost, retail_price), either or both None if DealerKit had not filled them in by the deadline (a real, normal outcome, not every reg matches a source DealerKit recognises; the record itself is unaffected either way).Nc>ST3UR;a-URRS:XaURU5 ggg)NrGET)rwrxryr)r _capturedrs rr3wait_for_dealerkit_financials.._on_response6s>  -6199;K;Ku;T  # Expenses list and that specific expense's own "More" menu. match_text is real, distinguishing text from the expense's own row on the plain list (for example "CarWow" plus "Delivery Charge" for a standalone one item expense, never "Buyers Premium" in the same search so it does not also match a fuller combined expense), used to find and open the right one; raises if none or more than one row matches, never guesses which expense a human meant. Built 2026-08-24 to remove a real duplicate: WX16AUE had two separate CarWow expenses, one a standalone single Delivery Charge item left over from however these older records were first entered, found live by Mark ("if you find record with duplicate info WX16AUE delete it"). Real reconnaissance found two dead ends before this worked: the item's own kebab menu (Edit / Duplicate / Delete) DOES let a single line item be deleted, but only from the browser's own local form state, a "Save & Approve" afterward silently did not persist an expense emptied down to zero items (DealerKit likely refuses that, an expense needs at least one line); this function goes through the EXPENSE's own separate "More" button (bottom left of the Edit Expense modal, distinct from the item row's own more_vert kebab, matched on its real "more_horiz" icon text so it is never confused with the item one) instead, whose own Delete removes the whole record in one action and needs no separate Save, verified live: item 752 and expense 571 were both confirmed gone via the API immediately after, no second save step needed. Triggers a real native window.confirm ("Are you sure you want to delete this expense?"), handled the same way as delete_vehicle's own dialog. Verifies via the API afterward that the matched expense's own real numeric id (read from the modal's own "Edit Expense #NNN" heading before deleting) no longer has any items, the same lesson as every other write in this module, a click reporting success is not proof something actually happened.rrrrrrExpensesraR(needles) => { const els = Array.from(document.querySelectorAll('div')); const out = []; for (const el of els) { if (el.children.length > 4) continue; const t = (el.textContent || '').trim().toLowerCase(); if (!t) continue; if (needles.every(n => t.includes(n))) { const r = el.getBoundingClientRect(); if (r.width > 0 && r.height > 0) out.push({x: r.x, y: r.y, w: r.width, h: r.height}); } } return out; }rz5DealerKit: expected exactly one expense row matching (vehicle z ), found rraE(row) => { const el = document.elementFromPoint(row.x + row.w / 2, row.y + row.h / 2); let node = el; for (let i = 0; i < 5 && node; i++) { const btn = node.querySelector && node.querySelector('button, svg, i.material-icons'); if (btn) { const r = btn.getBoundingClientRect(); if (r.width > 0 && r.height > 0) return {x: r.x + r.width / 2, y: r.y + r.height / 2}; } node = node.parentElement; } return null; }zPDealerKit: could not find the edit control for the matched expense row (vehicle rrrz() => { const h = Array.from(document.querySelectorAll('*')) .find(e => e.children.length === 0 && (e.textContent || '').includes('Edit Expense #')); return h ? h.textContent.trim() : null; }zEdit Expense #(\d+)r5z[DealerKit: opening the matched expense did not land on its own Edit Expense modal (vehicle aw(expense_id) => { const modal = Array.from(document.querySelectorAll('[role=dialog]')) .find(d => (d.innerText || '').includes('Edit Expense #' + expense_id)); if (!modal) return null; const els = Array.from(modal.querySelectorAll('button')); for (const el of els) { if ((el.textContent || '').includes('more_horiz')) { const r = el.getBoundingClientRect(); if (r.width > 0 && r.height > 0) return {x: r.x + r.width / 2, y: r.y + r.height / 2}; } } return null; }z@DealerKit: the expense's own More button was not found (expense z , vehicle rkflagFc`>T(aSTS'UR5 gUR5 gNTr6acceptdismissdacceptedconfirms r _on_dialog"delete_expense.._on_dialog" #HV  HHJ IIKr"dialogDeleteTr$zBDealerKit: the expense's own Delete option was not found (expense z@DealerKit: the delete confirmation dialog was not seen (expense r0?with[]=expenseItems&with[]=expenseItems.expense Nrc3z># UH0oRS5=(d 0RS5T:Hv M2 g7f)expenserNrf)rit expense_ids rr!delete_expense..s/uSsBvvi(.B33D9ZGSss8; expense_itemszDealerKit: expense z> still has items after delete, it may not have saved (vehicle )rrrL_open_edit_tiler&rItuplerglowerr.lenrrrr9searchrHgrouprr,rcontextrxrfrzany)rr match_textr?tneedlesr>row pencil_postitler?more_posr@after still_therer>rKs ` @@rdelete_expenser_PsD II :,i@+U<$Dj1$FPQ[^bdi]jFkFk*B*Q  qwwy *B &&( ) mm  G" 7|qCJ>R"|9S\N! =   !*C  J mnxmyy{|}}JJZ_jo6$ MM  E (%+26A xzDyEEGHI IQWWQZJ}}  H ]^h]iist~s@BCD DJJXc]HSM2$H GGHj!3"4>!cdncooy{EzFFH IJ J d# Xz2  F ]^h]iist~s@BCD D LL $ $  /*5ef % tv uSXS\S\]lSmSsqsSsuuK0 /overview URL.rrrrrzinput[placeholder=Search]z(DealerKit: the search box was not found.rrrrrr DealerKit: z" did not appear in search results.z/vehicles/(\d+)/overviewzDealerKit: opening z. did not land on its overview page (landed on z instead).r)rrrLrrrrrrrr@r9rRrwrHrS)rrrrr?s ropen_vehicle_overviewrbs. II *7ISXIY$t||,GHIOJ EFFJJSXG q 00#c(S]Q=N2NOMMs$ T ' '[-OPQQ$ -txx8A !#'((: /   qwwqz?r"c^^UR[SUS3SSS9 URS5 URS5nU(d[ SUS 35eUR R US US 5 URS 5 URS 5nU(d[ SUS 35eSS0mUU4SjnURSU5 UR R US US 5 URS5 URSU5 T(dgTS(d[ SUS 35eURRR[SU3SS9nURS:a9UR5RS5U:HnU(a[ SUS35eg!URSU5 f=f![a SnN { const els = Array.from(document.querySelectorAll('button, a, i, span')); for (const el of els) { if ((el.textContent || '').trim() === 'more_vert') { const r = el.getBoundingClientRect(); if (r.width > 0 && r.height > 0 && r.y < 200) return {x: r.x + r.width / 2, y: r.y + r.height / 2}; } } return null; }zEDealerKit: the vehicle's own menu (top right) was not found (vehicle rrrrka() => { const els = Array.from(document.querySelectorAll('.ui-menu-option__text')); for (const el of els) { if ((el.textContent || '').trim().toLowerCase() === 'delete') { const r = el.getBoundingClientRect(); if (r.width > 0 && r.height > 0) return {x: r.x + r.width / 2, y: r.y + r.height / 2}; } } return null; }z9DealerKit: the Delete menu option was not found (vehicle r6Fc`>T(aSTS'UR5 gUR5 gr8r9r<s rr@"delete_vehicle.._on_dialog@rBr"rCz@DealerKit: the delete confirmation dialog was not seen (vehicle rrFrrprTzDealerKit: vehicle 4 still resolves after delete, it may not have saved.)rrrLr.rrrrrrTrxrfrurzr ) rrr? kebab_posdel_posr@r{r^r>s ` @rdelete_vehicleris4 II :,i@+U<$   I bcmbnnpqrrJJYs^Ys^4$mm  G VWaVbbdeffH GGHj!3 ws|4 d# Xz2  F ]^h]iiklmm <<   # #~&6oj\$R\a # bD {{S ))+//$/:=K !4ZL@tuv v  Xz2 K s3F+2"G+F? GGcz^^UR[SU3SSS9 URS5 URS5nU(d[ SUS35eS S 0mUU4S jnUR S U5 UR RUS US5 URS5 URS U5 T(dg TS (d[ SUS35eURRR[SU3SS9nURS:a[ SUS35eg!URS U5 f=f)aJPermanently deletes a DealerKit contact from its own real page (`/contacts/{id}`, found live 2026-09-05, Steven: "to delete a contact inside dealerway you go to contacts, click the contacts name then click delete"), a real, plain "DELETE" button, not a kebab menu the way a vehicle's own delete is. The button's own visible text shares one node with the material icon ligature ("delete DELETE"), the same "no element's text is ever the plain word alone" pattern already seen elsewhere in this file (the Edit button), so it is matched by a CONTAINS check, not an exact one. Triggers a genuine native window.confirm, worded plainly: "Are you sure you want to delete '' and their ALL past deals, invoices, bookings and other related items? This cannot be undone." Handled the same real way as delete_vehicle's own confirm, a real page.on("dialog", ...) listener, never the auto dismiss a plain click would otherwise get. confirm=False stops right before accepting the dialog, for a dry run; the dialog is always dismissed either way in that case, nothing deleted. Verifies via the real API afterward that the contact genuinely no longer resolves, the same discipline as every other delete in this module: a click reporting success is not proof something actually happened. Only ever use this on a contact known to be safe to remove (a test contact carrying no real deals, invoices or bookings); the confirm dialog's own wording is the one honest warning DealerKit itself gives that this is not always reversible.z /contacts/rrrra() => { const els = Array.from(document.querySelectorAll('button, a')); for (const el of els) { const t = (el.innerText || '').trim(); if (!t.toUpperCase().includes('DELETE')) continue; const r = el.getBoundingClientRect(); if (r.width > 0 && r.height > 0) return {x: r.x + r.width / 2, y: r.y + r.height / 2}; } return null; }z4DealerKit: the Delete button was not found (contact rr6Fc`>T(aSTS'UR5 gUR5 gr8r9r<s rr@"delete_contact.._on_dialogrBr"rCrrrz@DealerKit: the delete confirmation dialog was not seen (contact z/api/contacts/rFrrpzDealerKit: contact rfT) rrrLr.rrrrrrTrxrfru)r contact_idr?rhr@r{r>s ` @rdelete_contactrn^sK8 II :,7DV`eIf$mm  G QR\Q]]_`aaH GGHj!3 ws|4 d# Xz2  F ]^h]iiklmm <<   # #~&6nZL$Q[` # aD {{S0 { const b = Array.from(document.querySelectorAll('.ui-modal.is-open button')) .find(el => (el.textContent||'').includes('Save Changes')); return b ? b.disabled : false; }g?rrrr$z6DealerKit: Save Changes was not found after uploading rrraz# did not actually save for vehicle z (server still reports unknownrz, expected at least ), even after a retry.)rrrLrr@rr,rfirstr set_input_filesr r.rrQ) rrr local_pathsrrSbeforer=rstill_disabledexpectedattemptr]s rrr#sW" !X-H;EII :,i@+U<$ 4U 3F T6 * *QRVQWWYZ[[$ tdCV W_YK7deff$\\> ? E EF ||~0177 ;'yy{RH ))+     3 ))+  !s;//H"4tD!WX\W]]^_` ` d#d6  !2{# # a< %T622  ! !$ '&tdC  ! !$ '#$  dV>zlK!*/*; K1 '%hZ/E G HHr"c [XSU/S5$)zAttach one file already on disk (a generated PDF, for example) as a vehicle Document. Never deletes file_path, that is the caller's own file, unlike the url based uploads which clean up their own temp downloads. Documentsz a document)r)rr file_paths rupload_local_documentrs -T{YKYe ffr"c B[XS[U=(d /5S5$)anAttach several files already on disk in one go: Carwow's own service record and V5C logbook scans, downloaded and extracted locally by carwow.read_purchase_documents, and, since 2026-08-26, Motorway's own service history and V5 scans too (motorway.download_purchase_media, see purchases_run.py); push_purchase still routes those through the ordinary upload_documents call, which now detects a local path just as well as a url (see _upload_via_edit_tile), this dedicated entry point is for a caller that already knows every item is local, such as the Delivery Report PDF. Never deletes the files, they are the caller's own.rr)rrI)rr file_pathss rupload_local_documentsrs" -T{DQ[Qa_aLbdo ppr"c6[XSU=(d /SUS5$)Nrrr)rr photo_urls max_imagess r upload_imagesrs" 8j>NBP[Q[=\^f ggr"c0[XSU=(d /S5$)Nrrr)rr document_urlss rupload_documentsrs ; @SQSU` aar")rfee transportrphotosrretailcD[URS5=(d /5[URS5=(d /5-nX RS5=(d /Vs/sH+n[U[5(dMU"U5(dM)UPM- sn- nU$s snf)a-Every document on this Mac's disk that a send attaches to the car: Carwow's service records and V5 scans, and, since 2026-09-16, what Dealer OS holds for the car (the MotorCheck history check PDF, dealer_os_documents). A path no longer on disk is left out rather than failing the upload. Pure, tested.document_pathsv5_local_pathsdealer_os_documents)rIrfr&r')rhexistsrrs rlocal_documents_forrs ./52 6hllK[>\>b`b9c cE ,,'<=CC iCA STVYHZa_efg_haC iiE Ljs$B BBc"US:=(a X:$)zWhether a repeat send should leave DealerKit's documents alone: it already holds at least as many as this send would add, and there is something to add at all.rrX)havesendings rdocuments_already_therers Q; *4?*r"cV[US5(d[SUS35e[U5HXnURU5 [ XSSS9(a gUR S5R 5S:Xa[US5 MXMZ [S US US35e![a M{f=f) aOpen the vehicle's Edit menu and click one of its tiles (Expenses, Images, ...), trying again while the menu draws (2026-09-15, found live on HX67UMH: the buyer's fee step raised "the Expenses tile was not found" on a record whose expenses were plainly there, the menu had not finished drawing 1.5 seconds after Edit). Up to `attempts` looks, `pause_ms` apart; Edit is pressed again if the menu is not open. Raises with the same wording as before when it never appears.rrrTrrNrrz tile was not found (vehicle )r@rrrLr,rr r )rrtileattemptspause_msrs rrNrNs T6 * *OPZ|[]^__? h' t>Q R  ||/0668A=!$/> # .KJU 35n&U&R S?5n%U%b/[;[=U%5[=U$5- 5S@:a U%nSAU%3US;'ORU%bS;U;a SBU%SCU$SD3US;'O<[?X U$5n#U#(a$U$nU#SE:XaSFOSGn'U%cS+U$U'3O SHU%SIU$U'3US;'OSJUS;'Sn(SKU;aCUR SL5cSMUSK'O+[AX USL5n)Sn(U)(aS USL3OS%USL3USK'[5n*SU;aUbU*RCS5 SU;aU S:aU*RCS5 SU;aU S:aU*RCS5 S;U;aU#(aU*RCS;5 SKU;aU((aU*RCSK5 SU;aU(aU*RCS5 S.U;aU(aU*RCS.5 S1U;aU(aU*RCS15 UU US)US US*UUU#U(U U [EU5[EU*5USN.$![,a e[.aYn!S8R1[3U!5R555SS9n"UHnS:U"3US-:XaUOS$-U[(U'M Sn!A!GNSn!A!ff=f)OaPush one purchase (a row shaped like db.list_purchases()'s output) into DealerKit: due in record, then images, then service history documents. Returns a summary dict. Raises on the first step that fails outright; a step with nothing to do (no photos, no service history) is just skipped, not an error. Raises DealerkitMissingMileage (never caught here) when this would be a genuinely new record and purchase has no mileage; no bypass exists, see add_due_in. items (2026-08-26, Mark: "choose which data to push", the row's own kebab menu) is an optional subset of {"price","photos","documents", "retail","due_in_date"} plus, since 2026-09-05, "mileage" and "fee" (the two extra things a car sent whole carries, see SEND_ITEMS); None (the default) pushes all five, the original all-in-one behaviour, now including Retail Price (2026-08-26, "the Retail Est is the retail price that should be mapped to DK", reversing the earlier "always manual" decision) and Due In / In-stock On (2026-08-26, "lets now wire up the due in date in BB to DK"). The due in record itself still always gets created first regardless (photos and documents both need a real vehicle_id to upload to), only whether BidBrain's own purchase price gets typed into it is what "price" actually gates.rN>rrrrrXr)rhrrrrrrzalready on DealerKit, z photos there)rz uploadedzCnothing to send, DealerKit already has these or BidBrain holds nonerservice_history_photos v5_photosz documents thererrFrz5nothing to send, BidBrain has no mileage for this carrTz entered as ,z miles when the car was addedchangedzset to z milesoldz, DealerKit had r5zalready right on DealerKit, effective_fee transport_fee?E2Fhll[fNgNmkmInn )2 EKK[D\M$K@bc "9c)ns:.N O O&YHUO(.G$ %d  $aH[ !)5G% &$X. ) g!5&**^:T:TDVNE[D\\]^M )*4XwOF#LLNS4@h4O%(I%5#''):L:LMRT5V+C01&s+)#$'JM,,"G--&&$(M+&4LL!23  !`HX *O&$Tn-=_ZL+YZ"*,,~">*s53IER[L\3\/]`d/d. 'EoEV%W" ,1J(EoEVW++4+5k'm"$4Di#P ##,L0@E0I,OfCJYJaIi[*F1@@QQWXaWbcfbg/hX&*IHX&O <<) * 2&aH] #'(CT:UVG"ORY:K1L0M'N.J8TeKfJg,h ] #J$=4w4HqLx dvz{#4,x }%D[y! }ud}{# )$!"23!'(?!@&$,*#$t Z( % K'   cHHSV\\^,Tc2E7I%5QHKyHX}^`6b+C01 cs( YZ2AZ--Z2ChassiszBuyers PremiumzDelivery ChargezAssurance / Indemnity Fee)rrr indemnity_feer)zStock PurchasezCost of Purchasec[U=(d S5R5nU(aUSS:Xa[U5S:aUSSnOUS/:XagSRU5R 5$)zA DealerKit button's own label with the material icon glyph taken off: "add EXPENSE" is "expense", "add ADD" is "add", a bare "add" (the icon only + beside a contact picker) is "".r5rrrNr)r'rrQrrP)r)r s r_add_button_labelrsa  O ! ! #E qU"s5zA~ab  5'  88E? ""r"c ^[U=(d /5VVs/sH9upURS5(dMX[URS554PM; nnnU(dgUV^s/sH*mSTS;d[U4SjS55(aM(TPM, sn=(d UnUVs/sHoDS(dMUPM sn=(d UnUVs/sHoDSS:Xd SUS;dMUPM sn=(d UnUVs/sHoDSRS 5(dMUPM! sn=(d Un[ US S 9S $s snnfs snfs snfs snfs snf) aWhich of the open modal's buttons with "add" in their text is the expense item's own ADD (2026-09-16, seen live on Steven's account: the form offers "+ EXPENSE", "+ CREDIT", an icon only "+" beside the supplier picker and the item row's own "ADD" at once, and picking by size or by the word "expense" pressed the wrong one twice). cands is [{text, ui, area, ...}] as the page reports them. A button that starts another expense, a credit or a contact is never it; the icon only "+" is never it while anything else offers; a label reading "add" or naming the item wins; then the smallest. Returns the index into cands, None when nothing offers.r)Nrrc32># UH oTS;v M g7f)rNrX)rr6rs rr%_choose_add_button..s8xLwqadLws)rHcreditcontactinvoicerruicN[USRS5=(d S5$)Nrarear)rFrf)rs r$_choose_add_button.. sqtxx'7'<1!=r")r r)rxrfrrUr)candsrcrowsrrealnamedrs ` r_choose_add_buttonr&s"BK5;TVAW iAW[\[`[`ag[h 4Q$QUU6]3 4AWD i  yt!v1~S8xLw8x5xAt y B~BD $t!tAt $ ,D @1! 1Q4Q @ HDE -Udhhtn!U - 6B r= >q AA j z $ @ -s:E E)'EE* E ;E E*EE EcFURS5nSHenURSUU(aUOSSS9nUc g[U5S-S-U::dUS :XdMAS SKnUR U5R S 5s $ g![a SnNzf=f![a SnNkf=f) uA small, low quality picture of the topmost open DealerKit window, clipped to just that window so it stays small (2026-09-16, Steven: "is there any way at all that you will be able to make it so you can sign in to dealerkit from here?" — no browser session or password lives outside the Mac, but the Mac can send back a real picture of what it saw, not just words, every time a step fails or the survey runs). Returns base64 JPEG bytes (a plain str) or None. Never raises; quietly gives up rather than block or fail whatever called it.aD() => { const opens = document.querySelectorAll('.ui-modal.is-open'); const m = opens.length ? opens[opens.length - 1] : null; const r = (m || document.body).getBoundingClientRect(); const vw = window.innerWidth, vh = window.innerHeight; const x = Math.max(0, r.x), y = Math.max(0, r.y); const w = Math.max(50, Math.min(r.width || vw, vw - x)); const h = Math.max(50, Math.min(r.height || vh, vh - y)); return {x, y, width: w, height: h}; }N)# jpeg@)rqualitycliprrr*rascii)r.r screenshotrQbase64 b64encodedecode)r cap_bytesrr-rr2s r _capture_shotr6#smm     ??/2d#LD < t9q=A  *gm ##D)009 9     D s"A?B? B B B B c [U5nU(dg[R"USS9 [[RR X5SSS9nUR SUS[U5S 35 [S [U5S 5HnUR S XFUS --S -5 M! UR S5 SSS5 g!,(df  g=f![a gf=f)zBase64 JPEG lines appended to a log file, ASCII only so diagnostics.tail_lines (a plain text reader) carries it intact: a header the puller recognises ("image: