How Machine-Checkable Is IAB Privacy Signaling?
IAB privacy signaling is a transport: the checkable layer is string shape, not a person-fact
Open-source conformance tooling for digital advertising specifications.
Digital advertising fails without throwing: a tag that bills and renders nothing, identifiers in a pixel nobody reviewed, a payload that cannot be told from fraud, an agent that transacts with nothing checking it. IAB Tech Lab and AAO specifications cover some of that. The rest is outside the PDF. The work here is open-source tools and measurements for those failures, including the places the documents do not reach.
Every arrow is a boundary between two companies. RTBlint and VASTlint sit on the payload as it crosses. Pixellint sits on the measurement artifact before it fires. Each box links to the source that does it.
The failures
They never throw, which is why they persist.
The specifications exist and are public. What is missing is a check at the moment a payload crosses between two companies. Some of that check is in the IAB Tech Lab and AAO documents. A lot of it is not. That gap is where revenue leaks, where user data travels unreviewed, where fraud becomes indistinguishable from error, and where autonomous systems now transact with nothing verifying them. Every tool here is free, auditable, cites the document behind each finding, and is built to run inside that gap.
Revenue integrity
A tag that fails renders nothing and still bills. The rules that catch it are the unglamorous ones: a missing impression element, a duplicate one that double-counts, a media file on plain HTTP that secure inventory blocks, quartile tracking nobody wired so measurement receives no signal at all.
VAST-2.0-inline-impressionVAST-2.0-duplicate-impressionVAST-2.0-linear-tracking-quartiles
User data and privacy
Tracking URLs are where identifiers actually travel, and they are the least reviewed artifact in the chain. Pixellint flags raw email addresses sent where the vendor requires SHA-256 hashing, credentials embedded in a URL, macros fired unexpanded to a third party, and any tracker still riding plaintext HTTP.
vendor.meta.pii.unhashed_emailcore.url.userinfo_deprecatedcore.macro.unexpanded_in_fired_url
Verifiability and fraud
A malformed payload and a fraudulent one look identical to everything downstream, because both are unverifiable. These tools do not detect fraud. They remove the ambiguity fraud hides in, by checking that the verification resources meant to catch it are present, correctly declared, and reachable.
VAST-4.1-verification-vendorVAST-4.1-verification-no-resourceVAST-4.1-js-resource-https
Agentic accountability
Buying is moving to agents on both sides of the transaction, with no human reading a validator UI in between. Every tool ships an MCP server so an agent can check its own output deterministically, and VASTlint implements AdCP 3.0 governance so a pipeline can declare and enforce content standards rather than assume them.
get_adcp_capabilitiescalibrate_contentvalidate_content_delivery
Each chip is a real rule or tool id you can look up in the catalog. Nothing on this page describes something that is not shipping.
Research
The measurements behind the tools.
Five preprints. Four ask what the documents leave open: how much of OpenRTB any validator can enforce, what header-bidding traffic actually sends, whether structural VAST validation is cheap enough for a live bid path, and how much of IAB privacy signaling a receiver can check. The fifth measures the CTV verification architecture that makes fraud cheap to run. Datasets and code are published with them.
Why CTV Ad Fraud Keeps Working
91.5% of pricing-relevant claims a buyer receives about a CTV server-side impression are self-declared, against 70.0% for the same claims on web display; no scored OpenRTB field on either path is attested
Measuring OpenRTB Dialects in Client-Side Header Bidding
42.0% of bid requests on a random header-bidding sample are flagged on type, enumerated-value, or required-field checks, 95% CI [36.0, 47.2]; major publishers sit at 40.4% [35.8, 45.0]; the intervals overlap
How Machine-Checkable Is OpenRTB?
417 normative sentences extracted from OpenRTB 2.6-202606 and 3.0, screened to 288 conformance statements and hand-coded into four enforceability classes
VAST XML Validation at Bid-Time Scale
A 17 KB CTV tag validates in 350 microseconds, 0.35% of a 100 ms bid window; a 44 KB tag in 2.1 ms, inside a typical creative-selection budget
The tools
One check per failure. Watch what it stops.
Each panel is a different artifact. Bid JSON, a wrapper chain into a CTV slot, a network request that has not fired yet. Without a check the failure is silent. With the validator it has a rule id.
RTBlint · Bid
v0.5.0The auction is lost. Nobody is told.
A bid request with a protobuf boolean where the OpenRTB JSON snapshot wants 0 or 1 does not throw. The DSP drops it. Fill from that request is gone. RTBlint grades the payload against the spec revision the partner is on, before it enters the auction.
bid_request.json
auction
- Bids received
- 0not sent
- Time in auction
- 120 ms · timed outnever entered
- Winning CPM
- nonenone
- DSP log
- no error · no bidrequest not forwarded
VASTlint · Creative
v0.9.1The slot billed. Nothing played.
A wrapper chain that dies on an HTTP media file still counts an impression. The player reports nothing useful. VASTlint walks the chain at intake, names the hop, and the deterministic HTTPS fixes apply in one pass, before the tag is trafficked.
- 0IntakeVASTlintidleVAST-2.0-mediafile-https
- 1Wrapperdsp.example/vast.xmlresolvednot fetched
- 2Wrapperadn.example/redirresolvednot fetched
- 3InlineMediaFile http://cdn.example/ad.mp4HTTP · blocked by playernever reached
Impression billedNot trafficked
CTV slot · 1920×1080
Pixellint · Measurement
EarlyThe pixel returned 200. The conversion did not.
A misspelled event name or a raw email in a tracker looks healthy in the network panel. Pixellint checks the URL against the vendor's own docs before it fires, while the tag is still in ad ops, not after attribution has already failed.
Network
GETfacebook.com/trgif200blocked184 msnot sent
0200 ms
| Query | Value | Vendor contract |
|---|---|---|
| id | 14288931 | ok |
| ev | Purchase | ok |
| em | unhashed | sent in the clearmust be SHA-256 |
HTTP200 OKnever left ad ops
Conversions00 · not fired
Use it in the browser
VASTlint and RTBlint run in the browser.
No install, no account, no key. Paste a payload and it grades against the same engine the CLI and the MCP server run. Pixellint is CLI and local MCP while it is early.
RTBlint · OpenRTB 2.0 through 3.0
Adoption
Published, installed, and in use.
Every figure below was read from a public registry API and can be checked against the same endpoint today. None of it is estimated, projected, or rounded up.
38,971
package installs across every published channel, as of 2026-08-26
Packages published
18
Distribution channels
7
Cost to use
$0
Show the per-package numbers
| Channel | Package | Count |
|---|---|---|
| npm | vastlint | 13,220 |
| npm | rtblint-core | 1,857 |
| npm | pixellint | 479 |
| RubyGems | vastlint | 3,859 |
| crates.io | vastlint-core | 5,913 |
| crates.io | vastlint-cli | 1,392 |
| crates.io | rtblint-core | 548 |
| crates.io | rtblint | 285 |
| crates.io | rtblint-mcp | 282 |
| crates.io | pixellint-core | 401 |
| crates.io | pixellint | 241 |
| crates.io | pixellint-mcp | 245 |
| Docker Hub | aleksuix/vastlint | 3,435 |
| GitHub releases | vastlint | 2,474 |
| GitHub releases | pixellint | 1 |
| GitHub releases | vastlint-erlang | 285 |
| PyPI | vastlint | 3,538 |
| VS Code Marketplace | aleksuix.vastlint | 516 |
| Total | 38,971 | |
How these are counted. Read straight from each registry’s public API on 2026-08-26. Nothing is estimated or projected. crates.io and RubyGems report lifetime downloads. npm has no lifetime endpoint, so that figure sums the daily range API from each package’s first publish date. PyPI uses the without_mirrors series, which excludes mirror traffic and is the smaller of the two figures available. Registries count machines, not people, so treat this as reach, not headcount.
AI and agents
A check the agent can call in its own loop.
An agent that assembles a campaign needs a deterministic verdict, not a screenshot of a validator UI. Every tool here ships an MCP server for that. 20 callable tools, structured JSON, no account.
Findings come back as structured JSON: a stable rule id, a severity, the line it came from, and fix guidance. An agent can act on that. It cannot act on a screenshot of a validator UI. Each server links to its source.
{
"mcpServers": {
"vastlint": { "type": "http", "url": "https://vastlint.org/mcp" },
"rtblint": { "type": "http", "url": "https://rtblint.org/mcp" },
"pixellint": { "command": "pixellint-mcp" }
}
}Standards
What the bodies publish, and what they do not.
IAB Tech Lab and AAO specifications define part of the bid, the tag, and the pixel. Implementing those documents closely is one job. Measuring the failures they do not name is the other.
IAB Tech Lab specifications
The specs that define how a bid, a creative, and a measurement event are structured. Every rule in the catalog names the document it came from.
- OpenRTB 2.0 through 3.0, including eight dated 2.6 snapshots
- VAST 2.0 through 4.4, plus VMAP, DAAST, and SIMID
- OMID verification checks and VPAID deprecation detection
Agentic advertising and AdCP
The Ad Context Protocol defines how agents negotiate and transact advertising. Validation is the part that keeps an autonomous pipeline honest.
- AdCP 3.0 capability discovery over MCP
- Content standards governance with pre-flight calibration
- Idempotent tool calls with replay protection
Contributing back upstream.
Implementing a specification closely enough to validate it surfaces the places where the specification itself is wrong, ambiguous, or untested. Those findings go back to the people who maintain it, as patches to the specs and their reference implementations.
Sets the technical standards the open advertising supply chain runs on, including OpenRTB, VAST, VMAP, and the agentic bidding framework.
Participant in the Advanced TV Working Group, and an active contributor to the reference implementations and testing tools on GitHub.
See the contributions on GitHubMaintains the Ad Context Protocol, which defines how AI agents discover each other's capabilities and transact advertising.
Certified AdCP Practitioner and Governance Specialist, and an active contributor to the protocol specification on GitHub.
See the contributions on GitHubHow it is built
The same three commitments in every tool.
Every rule cites a source
A finding names the spec section, the XSD, the RFC, or the documented revenue impact behind it. If a rule cannot cite something, it does not ship. That is what makes a result arguable with a partner instead of just an opinion.
One engine, every runtime
A single Rust core compiles to a CLI, a WASM module, a native library, and language bindings. Your CI, your browser, your BEAM service, and your agent all get identical verdicts, because they run identical code.
Free, and staying that way
Apache 2.0, with a written commitment: no paid tier that hides rules, no cloud version that requires an account, no rate limits, no telemetry without opt-in. Validation infrastructure only works if everyone can run it.