The context graph for one product in one market, for every go-to-market job beyond the buyer list: short-form video ads, creator seeding, marketplace listings and images, trends and social listening, buyers, pricing and the rules before sale (industrial goods: buyers, trade shows, search demand, pricing, rules). One read is meant to be enough to do the job correctly. How to read it: (1) shared_brief holds the lines that bind several jobs, said once (the sale gate, the trademark, wording to change, duty and proof of origin); each job lists their keys in shared. wait_for_gate on a job means nothing is published, listed, sent or pitched before the gate clears; gate.ban means no approval lifts it and the product as it is cannot be sold there. blocked_by on a job means that job is prohibited in this market (advertising alcohol other than beer in Poland, selling alcohol online in Taiwan): the brief says so instead of how to do it, and nothing is made. (2) A job's brief holds that job's own instructions; a rule or node id named in a line is a node whose text says what to do. needs_prior_consent means sales email there goes only to buyers who agreed to it. (3) uses lists the nodes the job reads. Rule nodes carry the gist (text), who must act, the status and start date, lead_time for approvals and the official source with the date it was checked; deadlines lists the rules that start within 13 weeks. (4) Calls: read_first_calls before creating (or reuse a fresh last_read), then next, then then. Every call passes its tool's input schema as given; fill lists the arguments to ask the user for ({arg, ask, example}); needs: account marks tools that exist only when signed in. Money: a call with paid: true charges credits itself (find_buyers). A job with paid: true costs est_cost_usd in total for cost_for, charged by charged_by (live_call = the paid data and generation steps that the job's next call lists); a job with paid: false has no cost fields. Show the cost and get approval before either. Evidence: reply figures name their yardstick. The market node's reply_vs_avg is measured on the product's industry (reply_scope industry, reply_basis says what 1.0 means, with the 90% interval and the buyers contacted); buyer-type nodes are ranked on all industries' emails (reply_scope all), because the data has no cell for an industry by buyer type. When the market's confidence is stronger than any buyer type's, the market node says why in buyer_evidence, and the buyers brief leads with the product's own expected replies (the same figure as gtm_plan's reply_target). Nodes come from the same sources as gtm_plan, video_studio, image_studio, channel_plan, price_ladder and blind_spots (same sale gate, claims_check and, signed in, the same company): the market (UN Comtrade imports, growth, Korea's share), buyer types and the channels buyers answer on (rank), the creator platforms in order with the first one's video format, the marketplaces in order with the first one's main-image rule, language, everyday scene, claims, disclosure and AI-content rules, product rules and easy-to-miss requirements, the price ladder multiple, and the live-data signals to read first (motion names the channel_plan call that returns their endpoints and prices). There are no edges: order is each node's rank. Each job reads the requirements that bind it, the easy-to-miss ones included: ad rules for video and seeding, rules for sending samples to creators and buyers (e.g. US Prior Notice covers food samples, alcohol cannot go by US mail), listing and label rules plus what you take on when you sell direct by parcel without an importer (e.g. Japan's 24 units per item, packaging EPR), prior consent for sales email, approvals a buyer asks about first, and extra duties or taxes for the price. Korea-only lines (Korean FTAs, KOTRA pavilions) are left out for products made outside Korea. rules_coverage says how much TradeGraph knows for that market (national, bloc, pointers or none: none is not a clearance); rules_page links the public page with the same rules when there is one. Signed in, it starts from the company's own records (learned): the last market reads saved with company_context (kind=signal) sit on the signal nodes with their age and whether they are stale, so a fresh read is reused instead of paid for again; seeding_list results per platform sit on the platform nodes and the seeding brief; hook tests (kind=result) sit on the video format and brief, so the next video starts from the angle that held best. Save what you learn the same way so the next graph starts from it. These are the company's own numbers only; nothing is shared with other companies. Call it after gtm_plan when the user wants to launch in a market ("what do we need for Japan", "make videos for TikTok in Vietnam", "list on Qoo10"), or before video_studio and image_studio so hooks and frames start from the market. Omit country to use the data's first market. job returns one job only, with the nodes and shared lines that job needs. Free, read-only.