Sponsored-display is the banner format. It sits alongside sponsored-research (native, in-list) and reuses the same window._sponsoredResearch object under a .Display key.

The unit is a Criteo-shaped onsite-display entry: build-time PNG plate (see plates) plus DOM text overlay. The plate carries no legible copy because flux2-klein cannot spell, and the disclosure label must be selectable text for data-nosnippet and screen-reader compatibility anyway — two constraints, one architecture.

Provenance surfaced via a ? affordance on every rendered banner showing model, sha256, seed, reproducible, synthetic=true. This is what an audit surface looks like when the underwriter is a local flux2-klein instance, not a third-party ad server.

ORACLE: src/wal_sh/adtech/sponsored_display/browser.cljs (browser only — no pure core.cljc; the format is DOM-native).

SD1

Plate is a build-time PNG produced by wal-sh.adtech.plates.gen; the browser never generates plates.

SD2

Every word on the banner is DOM text; the plate carries no copy. The plate is the visual anchor; the text is the message.

SD3

Disclosure label is selectable text with data-nosnippet on its container. Screen readers get a proper text alternative.

SD4

A ? affordance on the banner exposes the creative's provenance JSON — model, sha256, seed, reproducible, synthetic true. Clicking it is a genuine audit surface, not boilerplate.

SD5

Load order: sponsored-research must be initialised first; this ns attaches Display onto the same window._sponsoredResearch object. Registry-enforced (see publish.el wal-sh/adtech-systems).

SD6

display-format returns nil when the manifest has no plate for the (dir, slot) pair. The renderer must handle nil by not emitting the slot; no fallback image.

SD7

The RenderingAttributes-shape input matches the Criteo onsite-display API contract; the product arg is a sponsored-research SKU with a RenderingAttributes JSON string we parse for the taxonomy_text key.

Browser (browser.cljs):

(load-manifest! [url])                            ; -> Promise, fetches /static/img/sponsored/manifest.json
(pick-plate dir slot)                             ; -> {:src :width :height :alt :sha256 ...} or nil
(display-format product slot)                     ; -> Criteo entry map or nil
(render entry)                                    ; -> DOM node
(after container product slot)                    ; -> mount after existing element

Public surface:

window._sponsoredResearch.Display = {
  Plates: {
    load:  (fn [] -> Promise)                     ; fetch manifest
    pick:  (fn [dir slot] -> plate-map or nil)
  }
  displayFormat: (fn [product slot] -> entry-map or nil)
  render:        (fn [entry] -> Node)
  after:         (fn [container product slot] -> nil)
}

Manifest shape (/static/img/sponsored/manifest.json):

{
  "generator": {"model": "flux2-klein", "version": "..."},
  "plates": [
    {"dir": "research", "slot": "leaderboard",
     "src": "/static/img/sponsored/research-leaderboard.png",
     "width": 728, "height": 90,
     "alt": "...", "sha256": "...", "seed": 12345,
     "reproducible": true, "synthetic": true},
    ...
  ]
}

• IAB Standard Ad Unit Portfolio (Interactive Advertising Bureau 2017) defines the dimensions (leaderboard 728×90, mrec 300×250, billboard 970×250, mobile 320×50) that plate generation targets.
• Zeng, Kohno & Roesner on bad ads (Zeng, Kohno, and Roesner 2020) characterises the creative-quality failure mode this spec's per-banner provenance addresses: an audit surface tied to the creative, not the container.
• Ali et al. on delivery discrimination (Ali et al. 2019) motivates the ? affordance — provenance on the creative is the smallest visible unit an outside audit can verify.
• ANA 2023 MFA transparency study (Association of National Advertisers 2023) on inventory quality — display's disclosure surface is exactly what the MFA studies argue is under-instrumented.
• Cloudflare Rocket Loader documentation (Cloudflare, n.d.) on script-deferral pattern; the same deferral we use for the banner mount to avoid CLS on the main-content paint.
• data-nosnippet is a Google-defined HTML attribute for suppressing snippet extraction; documented at https://developers.google.com/search/docs/appearance/snippet #prevent-snippets.

• sponsored-research/spec.org — this module attaches to its window._sponsoredResearch object.
• plates/spec.org — build-time plate generation via flux2-klein.
• sponsored-formats/spec.org — historical taxonomy of display in the sponsored-format catalogue.
• attribution-audit/spec.org — display impressions and clicks append touchpoints to the ledger.
• ORACLE: src/wal_sh/adtech/sponsored_display/browser.cljs
• web-history — the banner era on the invention of the 468×60, the leaderboard, and the <a><img></a> as a first-class ad unit.
• [BROKEN LINK: No match for fuzzy expression: *2013–2018: Mobile + programmatic era] on Criteo's shape and the onsite-display API convention.

• Multi-slot rendering per page. Currently one plate per (dir, slot) pair; a per-page A/B on plate variants is not wired. The plumbing exists (plate A/B in the manifest) but the renderer picks the first match.
• Retina/HiDPI. Plates are 1x; a 2x variant would double the storage cost and quadruple the byte budget. The IAB portfolio (Interactive Advertising Bureau 2017) specifies 1x + 2x pairs; wal.sh ships 1x only.
• Fallback strategy on missing plate. Current behaviour: don't render. Alternative: render the DOM text on a solid greyscale fill. The former is honest (no plate = no banner); the latter keeps the slot dimensions reserved but changes the visual contract.