For ad platforms

Better creative inputs, before the auction.

Context Catalog is a protocol for ads that arrive labeled, checked, and connected to real outcomes. Platforms spend less effort on redundant creative, learn faster from new creative, and give advertisers a clearer reason to grow their budgets.

U.S. Patent Pending
The pressure

The generative flood lands on the platform.

Creative volume is exploding

Generative tools let every advertiser submit far more variants, many of them near-copies. Each one has to be processed, stored, and ranked.

Every new ad starts cold

New creative needs exploration impressions before predictions can be trusted, even when it reuses footage that has already performed.

Inputs arrive without context

Platforms see files, not intent: which claims are approved, how the ad is built, and what the advertiser already knows works.

What platforms gain

Efficiency, signal, and growth.

Efficiency

Less redundant work

Near-duplicates are screened before indexing and ranking, and verified segments shorten exploration for new creative.

Signal

Data advertisers already trust

Approved claims, segment structure, and clean-room outcomes arrive with the ad, from the advertiser who knows them to be true.

Growth

A reason to invest more

Advertisers who can see which creative drives long-term value have a clearer case for increasing spend.

Technical breakdown

Five components, one loop.

Each component exchanges a documented JSON format. The demo implements all five, including real ES256 signing and verification, entirely in the browser.

  1. Advertiser → Pre-flight gate
    Signed ad record
  2. Pre-flight gate → Ranking + delivery
    Admitted ads
  3. Ranking + delivery → Data clean room
    Conversion events
  4. Data clean room → Pre-flight gate
    90-day outcomes (aggregate only)
  5. Pre-flight gate → Advertiser
    Pre-flight results, segment readout
Customer-level data never leaves the clean room; only aggregates flow back.
  1. 01

    Signed asset manifest

    A C2PA-style record attached to each ad: the catalog version it was built from (by hash), the claims it uses, and its temporal ingredients (hook, problem, demo, call to action), each with a content hash and optional lineage to a parent segment. Signed with ES256.

  2. 02

    Pre-flight gate

    Runs before an ad enters the auction: verifies the signature and every lineage claim, resolves the catalog version and blocks prohibited claims, compares precomputed embeddings with live ads to screen near-duplicates, and derives compact semantic tokens.

  3. 03

    Cold-start inheritance

    When a verified segment's lineage points to a stored winner, the new ad inherits that segment's prior and needs only a short verification run instead of full exploration.

  4. 04

    Segment-level readout

    After spend, attribution returns for each segment (retention, click-through, and net 90-day customer value from a clean-room cohort) with a recommended action: rework, lock, or hold.

  5. 05

    Human-approved write-back

    The advertiser's harness drafts a learning from the readout. It enters the catalog only on human approval, creating a new, hash-chained catalog version for the next ad.

The gate concept is inspired by publicly described approaches such as asynchronous embedding precomputation and semantic tokenization (for example, Meta's published SOLARIS and GenRecs research). It does not describe how any platform works today.

Compliance and trust

Built for review, not just speed.

  • Provenance: every ad is cryptographically tied to the brand record it was made from.
  • Tamper evidence: altered records and false lineage claims fail verification.
  • Policy before spend: prohibited claims are blocked at the gate, not after delivery.
  • Audit trail: catalog versions are hash-chained, and each learning records who approved it.
  • Privacy: customer-level data stays in the clean room; only aggregates cross the boundary.
Two execution paths

Advertiser pixels, or platform generation.

Path A works with creative made anywhere. Path B is a proposed interface where the platform's own generative tools build from a signed blueprint, after the same checks.

Path A

Bring your own pixels

  1. 1.Your team, agency, or AI tools make the finished ad.
  2. 2.It ships with a signed record of what's inside it and which brand record it was built from.
  3. 3.The platform checks it before any budget is spent.

Works with any creative tool

Path B

Guided native generation

  1. 1.You send a signed blueprint instead of a finished file: your brand rules plus the proven segments to keep.
  2. 2.The platform's own generative tools build the rest.
  3. 3.The same checks run before anything is generated or shown.

Proposed: no ad platform offers this today

See the protocol run.

Every stage of the demo includes the real JSON documents exchanged between advertiser and platform.

Conceptual simulation for discussion purposes. Not affiliated with or endorsed by Meta, Adobe, Canva, Google, AWS, Snowflake, or the C2PA. All brands and data are fictional and illustrative.