, so the census took the heading from the

above it and its harvested label was a run-on of the whole refresh toolbar. Rendered by AIEO_DMM_Competitions::render_insights(), not by the competition_insights shim. B74: the LTV cell concatenated get_woocommerce_currency_symbol() onto number_format_i18n, which puts the symbol on the wrong side in any currency that leads with it.', ), 'ad_connectors.ad_spend_connectors' => array( 'label' => __( 'One Google account, three jobs', 'ai-eshop-optimizer' ), 'page' => 'ad_connectors', 'anchor' => 'ad-connectors-ad-spend-connectors', 'kind' => 'grid', 'phone' => 'list', 'tier' => 'inherit', 'scope' => 'local', 'source' => 'query', 'rows_est' => 7, 'caps' => array( 'sort' => 'none' ), 'adopted' => true, 'note' => 'MIGRATED to T0, session 18. TWO tables of one shape (the wiring card and the boost-governance card) that the census counts as one row; wiring_row() returns data now and one wiring_table() renders both. `sortable => false` on purpose — the order IS the story (spend, boost, campaigns; enabled, sandbox, kill, caps) and alphabetising a list of gates is worse than the list. First user of the badge `tone_key`: "off" is GREEN for paid boosting (off is the shipped, safe state) and grey for sandbox (off means live spend), so the tone is a per-row judgement the value => tone map cannot express. No row_key: there is no record behind these rows and keying them on the translated label would put a Greek sentence in a data-row-key.', ), 'form_insights.table' => array( 'label' => __( 'Answer distribution per field', 'ai-eshop-optimizer' ), 'page' => 'form_insights', 'kind' => 'grid', 'phone' => 'grid', 'tier' => 'inherit', 'scope' => 'local', 'source' => 'query', 'rows_est' => 4, 'caps' => array( 'sort' => 'none' ), 'adopted' => true, 'note' => 'MIGRATED to T0, session 18. NOT a B73 phantom: it is real markup, one table PER FORM FIELD inside its own card, so the census saw no card anchor of its own and took a run-on label from the field heading. The share column was a flex row wrapping two nested divs with a hardcoded #2271b1 — the `bar` type. Rows stay clickable through `row_data` (the drill needs field, stored value AND label, which is more than an identity). ⚠️ `tabindex` and `role` are NOT data-* so row_data cannot carry them; the screen re-applies those two attributes after render, which is the one thing left that touches a rendered row. Since 2026-08-30 it renders from assets/pages/dmm/form-insights-body.php, which the AJAX refresh path also uses — there is no longer a second hand-built table in the screen JS.', ), 'form_insights.section_scores' => array( 'label' => __( 'Average score per question', 'ai-eshop-optimizer' ), 'page' => 'form_insights', 'kind' => 'grid', 'phone' => 'list', 'tier' => 'inherit', 'scope' => 'local', 'source' => 'query', 'rows_est' => 7, 'caps' => array( 'sort' => 'none' ), 'adopted' => true, 'note' => 'Added 2026-08-30 with the field-shape classifier. One row per RATING question in a section, sorted worst-first — `sortable => false` because that order is the whole point of the table. The bar is scaled to the question\'s OWN top of scale via a `scale_max` row key, never a hardcoded 5, so a 1-10 question beside a 1-5 one does not share an axis it does not share on the form. `label => false` on the bar because the Average column already prints the number. Tone comes from a `tone_key` computed per row as a fraction of that scale, for the same reason.', ), 'form_insights.tick_answers' => array( 'label' => __( 'Answer distribution per field', 'ai-eshop-optimizer' ), 'page' => 'form_insights', 'kind' => 'grid', 'phone' => 'grid', 'tier' => 'inherit', 'scope' => 'local', 'source' => 'query', 'rows_est' => 4, 'caps' => array( 'sort' => 'none' ), 'adopted' => true, 'note' => 'Added 2026-08-30. Several standalone "tick if this applies" checkboxes under one section heading are not several questions — they are the answers to the heading ("how did you hear about us"), and one one-row table each is the shape that hid it. Share is out of ALL entries, not out of the people who ticked, which would score every one of them at 100%. Only forms when the section holds two or more; a lone tick keeps its own block.', ), 'form_insights.ranked_options' => array( 'label' => __( 'Ranked in order', 'ai-eshop-optimizer' ), 'page' => 'form_insights', 'kind' => 'grid', 'phone' => 'grid', 'tier' => 'inherit', 'scope' => 'local', 'source' => 'query', 'rows_est' => 9, 'caps' => array( 'sort' => 'none' ), 'adopted' => true, 'note' => 'Added 2026-08-30. A Gravity `rank` field stores ONE comma-separated row per entry and the POSITION is the answer, so the summary is a mean position and LOW WINS — the bar carries `invert => true` for exactly that, or it would grow as the answer got worse. Rows are deliberately NOT drillable: "who put shipping third" is a different question from "who chose shipping" and the row does not carry the answer that would make the drill honest.', ), 'form_insights.matrix_rows' => array( 'label' => __( 'Rating grid', 'ai-eshop-optimizer' ), 'page' => 'form_insights', 'kind' => 'grid', 'phone' => 'grid', 'tier' => 'inherit', 'scope' => 'local', 'source' => 'query', 'rows_est' => 5, 'caps' => array( 'sort' => 'none' ), 'adopted' => true, 'note' => 'Added 2026-08-30. One row per SUB-INPUT of a multi-row likert. ⚠️ The row ids are not guessable — the add-on skips .10 and .20, so a ten-row likert ends at 7.11 — and each sub-input stores ":", so only the half after the colon is the answer. Neither form on the reference install has one; the shape is built from the add-on source and exercised by the classifier, not by live data.', ), 'delivery_stats.fulfillment_delivery_time_metrics_by_postal_area' => array( 'label' => __( 'Delivery times by postcode area', 'ai-eshop-optimizer' ), 'page' => 'delivery_stats', 'anchor' => 'fulfillment-delivery-time-metrics-by-postal-area', 'kind' => 'grid', 'phone' => 'wide', 'tier' => 'inherit', 'scope' => 'local', 'source' => 'query', 'rows_est' => 8, 'caps' => array( 'sort' => 'client' ), 'adopted' => true, 'note' => 'MIGRATED to T0, session 18. ⚠️ Same table as `fulfillment.delivery_metrics`; both ids are frozen and both are marked adopted. Three things went out with it: B74 — $fmt_money HARDCODED "€" and the C locale (number_format($f, 2, ".", ",")), so a shop that writes 1.234,56 € everywhere else read €1,234.56 here with the separators the wrong way round; its own 48-line sorter with its own arrow spans and data-sort-type; and two