RTBlint

RTBlint

OpenRTB linter. Validates OpenRTB 2.x bid requests and bid responses against versioned IAB Tech Lab spec snapshots, from 2.0 through the monthly 2.6 releases (currently up to 2.6-202606).

Website and playground: rtblint.org

What it checks

• Malformed JSON and wrong top-level shape
• Required fields, including required non-empty arrays
• Unknown objects and fields per version catalog (ext subtrees stay open)
• Type mismatches (string, integer, float, boolean, object, and array forms)
• Documented enum values, including AdCOM lists and vendor ranges (500+)
• Deprecated, moved, removed, and not-yet-available fields across versions
• Semantic rules: site/app/dooh exclusivity, imp media type presence, skippable video dependencies, duration exclusivity, seatbid/nbr presence on responses, and more
• Response markup coherence: bid.adm content vs the declared bid.mtype (native JSON encoding, VAST/DAAST roots, double-encoded payloads)
• Request/response cross-validation: with the originating request supplied, every bid's impid, mtype, adm markup, dealid, seat, and currency are checked against what the request actually offered
• JSON dialect: spec JSON types flag fields such as imp.secure and regs.coppa as integers, while the IAB OpenRTB protobuf schema declares 28 of them bool. Either encoding is correct on its own transport and wrong on the other, so the caller declares which one it meant
• Exchange profiles apply published request, response and paired constraints for the selected vendor and traffic direction. Optional extensions stay optional unless the source explicitly requires them. Profiles include Google Authorized Buyers, Prebid Server, Microsoft Monetize, Magnite DV+ xAPI and separate buyer, supplier and SDK contracts. See exchange profile evidence for sources, fixtures and limits
• Nested specs OpenRTB carries as strings or opaque ext: Native Ads 1.2 markup (imp.native.request and native bid.adm, including required-asset pairing), GPP header vs gpp_sid and TCF 2 shape, ${AUCTION_*} macros on billing and loss URLs, EID/SUA structure, SKAdNetwork ext.skadn
• Privacy signal contradictions: regs.coppa / device.lmt / device.dnt / regs.gdpr versus identifiers on the same payload, TCF Purpose 1 versus device-storage IDs, US Privacy Opt-Out Sale versus hashed EIDs, email-shaped user.id / site.page, DSA Transparency field presence. Findings describe the document. They do not state a legal conclusion.
• ARTF envelopes and mutation sets, including applying the mutations and revalidating what comes out

Every finding carries a stable rule id, a severity, a message, and a JSON path.

ARTF

ARTF, the IAB Tech Lab Agentic Real Time Framework, hands an agent an OpenRTB payload inside an RTBRequest envelope and takes back mutations: proposed changes the orchestrator may accept or reject one at a time. Nothing in the framework checks that the auction still validates once they are applied, and a mutation is only meaningful relative to the request it targets.

# The envelope, plus full OpenRTB validation of what it carries
rtblint validate --type artf-request rtb-request.json

# The mutation set against the auction it targets
rtblint validate --type artf-response --request rtb-request.json rtb-response.json

# Apply the mutations, revalidate, and report only what the mutations broke
rtblint validate --type artf-response --apply --request rtb-request.json rtb-response.json

Three passes:

• Envelope. Required members, lifecycle against the payloads actually carried, tmax plausibility for an in-auction call, originator and applicable_intents enum values, and the carried bid request and bid response validated as protobuf JSON.
• Mutations. The response id echoes the extension point request id (not the bid request id), each declared intent is in applicable_intents, the operation and payload oneof member match the intent, and every semantic path (/imp/{id}, /imp/{id}/pmp/deals/{id}, /user/data/segment, /seatbid/{seat}/bid/{id}) resolves to something the auction carries. ADJUST_DEAL_MARGIN is reported as having no OpenRTB field to write to, because it does not.
• Applied. The mutations are written in and the result revalidated, reporting the OpenRTB findings the mutations introduced with pre-existing findings filtered out. What the agent broke, not what arrived broken.

The ARTF v1.0 document and its .proto use different vocabularies for the same mutation (activateSegments and a value: {IDsPayload: ...} wrapper against ACTIVATE_SEGMENTS and top-level oneof members). Payloads written from the document are mapped and reported as artf.mutation.legacy_spec_encoding rather than dismissed as unknown.

Surfaces

OpenRTB 3.0 validates through its layered envelope: the transport objects (Openrtb, Request, Item, Deal, Source, Response, Seatbid, Bid) and the AdCOM 1.0 domain objects under item.spec (Placement), bid.media (Ad), and request.context. A 2.x payload sent to a 3.0 validator gets a migration diagnostic rather than a bare parse error. The 2.6-202204 snapshot has no extracted catalog and reports itself as unsupported instead of passing payloads silently. See ROADMAP.md for what's next and CHANGELOG.md for release history.

CLI

cargo install rtblint

rtblint validate request.json
rtblint validate --type response response.json
rtblint validate --type response --request request.json response.json
rtblint validate --version 2.5 --format json request.json
rtblint validate --dialect proto-json grpc-bid-request.json
rtblint validate --profile google-ab google-bid-request.json
rtblint validate --profile prebid-server pbs-auction.json
rtblint validate --profile xandr xandr-bid-request.json
rtblint validate --profile magnite magnite-bid-request.json
rtblint validate --resolve --cache ./supply-cache request.json
rtblint validate --summary bids.ndjson
rtblint validate --batch --summary bids.ndjson
cat request.json | rtblint validate --stdin

--request supplies the originating bid request for paired checks, including batch validation. --dialect proto-json selects the protobuf JSON encoding. --profile selects the vendor contract and traffic direction. Xandr checks bidder-facing Microsoft Monetize traffic; Magnite checks the DV+ xAPI extension model. Their optional seller, context and inventory identity fields are checked when supplied. Account-specific and runtime acceptance remain outside the payload. --resolve --cache <dir> checks SupplyChain hops against sellers.json and ads.txt / app-ads.txt from a local directory:

<dir>/sellers/<asi>/sellers.json
<dir>/ads/<site.domain>/ads.txt
<dir>/app-ads/<app.bundle>/app-ads.txt

Protobuf JSON also supports the IAB schema's native object alternatives: imp.native.request_native (requestNative) and bid.adm_native (admNative). Their native flags use booleans. The string alternatives request and adm keep their JSON-encoded Native Ads markup. Supply one alternative per oneof. Native assets remain validated and paired against required request assets. See the IAB protobuf schema.

Nothing is fetched; populate the cache yourself. --batch lints one JSON object per line from a file or stdin. --summary adds rule-frequency totals for a captured stream (--summary bids.ndjson for the histogram alone). See ARTF for --type artf-request and --type artf-response.

Exit codes: 0 valid, 1 validation errors, 2 usage or I/O error.

GitHub Action

The Action lives in this repo. Pin a release tag so CI downloads that CLI tarball:

- uses: aleksUIX/[email protected]
  with:
    path: fixtures/bid-request.json
    spec-version: 2.6-202505

version selects the CLI release (auto follows the action's own v* tag). spec-version is the OpenRTB snapshot. Linux and macOS runners, x86_64 and aarch64.

Node

import { validate, validateResponse, validateResponseAgainstRequest } from "rtblint-core";

const report = validate(JSON.stringify(bidRequest), "2.6-202505");
if (!report.valid) {
  for (const issue of report.issues) {
    console.log(`[${issue.severity}] ${issue.path}: ${issue.message} (${issue.id})`);
  }
}

// Cross-validate a response against the request it answers.
const paired = validateResponseAgainstRequest(
  JSON.stringify(bidResponse),
  JSON.stringify(bidRequest)
);

For gRPC bidstream payloads and ARTF:

import {
  validateDialect,
  validateProfile,
  validateArtfRequest,
  validateArtfResponseApplied,
  protoBoolDivergences,
} from "rtblint-core";

validateDialect(JSON.stringify(bidRequest), "proto-json");
validateProfile(JSON.stringify(bidRequest), "google-ab");
validateProfile(JSON.stringify(bidRequest), "prebid-server");
validateArtfRequest(JSON.stringify(rtbRequest));

// { result, application }: what the mutations broke, and the payloads they produced
const { result, application } = validateArtfResponseApplied(
  JSON.stringify(rtbResponse),
  JSON.stringify(rtbRequest)
);

protoBoolDivergences(); // the 28 fields the two schemas type differently

MCP server

Hosted Streamable HTTP (no install): https://rtblint.org/mcp. Smithery listing: aleksander/rtblint (same account as vastlint). Payloads sent to the hosted endpoint may be stored (identifiers stripped) so the rules can be improved; see rtblint.org/privacy. Local rtblint-mcp over stdio does not send payloads.

rtblint-mcp also speaks MCP over stdio. Tools: validate_bid_request, validate_bid_response (optional bid_request for cross-validation), validate_artf_request, validate_artf_response (apply writes the mutations and revalidates), list_openrtb_versions, get_adcp_capabilities. Validation tools take optional dialect and profile arguments.

The ARTF tools are the guardrail an agent calls around its own work: check the envelope it was handed, then check the mutation set it is about to propose, before the orchestrator sees it.

{
  "mcpServers": {
    "rtblint": { "url": "https://rtblint.org/mcp" }
  }
}

Local stdio instead of the hosted endpoint:

{
  "mcpServers": {
    "rtblint": { "command": "rtblint-mcp" }
  }
}

Rust library

use rtblint_core::{validate_bid_request_for_version, OpenRtbVersion};

let result = validate_bid_request_for_version(OpenRtbVersion::V2_6_202505, payload);
for issue in &result.issues {
    println!("{} {} {:?}", issue.severity, issue.id, issue.path);
}

validate_bid_request_with_profile applies an exchange profile (Profile::GoogleAuthorizedBuyers, Profile::PrebidServer) on top of the spec. validate_bid_request_with_dialect selects spec JSON vs protobuf JSON.

Published profile IDs: spec, google-ab, prebid-server, xandr, magnite, dv360, index-exchange, index-exchange-seller, unity, vungle, bidswitch, bidswitch-supplier, inmobi, inmobi-supplier, mobilefuse, mobilefuse-sdk, applovin-alx, commerce-grid, digital-turbine, sovrn, equativ, equativ-supplier, triplelift-supplier, adform-handler, yandex-sdk-bidding, pubmatic-openwrap, pubmatic-openwrap-ctv. Select the matching direction, such as bidswitch for buyer 5.7 and bidswitch-supplier for supplier 1.1.

See the profile scope and source evidence for checked fields, supported directions and explicit gaps. Paired response validation uses the request as reference context; run request validation separately for its complete report.

JSON Schemas

schemas/ holds a JSON Schema (draft 2020-12) per tracked version, for both payload types, generated from the same catalogs the validator uses. IAB Tech Lab publishes no JSON Schema for 2.6 or 3.0, so these are the machine-readable contract for each monthly snapshot:

https://rtblint.org/schemas/openrtb-2.6-202606-bid-request.schema.json
https://rtblint.org/schemas/openrtb-2.6-202606-bid-response.schema.json

They cover structure, types, required fields, documented value sets, and AdCOM enum lists. What they cannot express is what the linter adds: deprecated and moved paths, version-specific removals, and the semantic rules. Validating against a schema is not the same as linting.

Regenerate after any catalog change (CI fails if they drift):

cargo run -p rtblint-core --example export_json_schemas

Documentation

Beyond this README, rtblint.org hosts the reference material:

• Diagnostic code reference: every stable issue id, each with a page covering what it means, why it matters, and how to fix it
• Versioned rule catalog: what every OpenRTB release added, deprecated, moved, or removed
• OpenRTB versions: a page per tracked version, 2.0 through 3.0
• Common OpenRTB mistakes and validating in CI
• Bid request anatomy, bid response anatomy, and the OpenRTB FAQ

Spec data and provenance

The validator runs on structured catalogs extracted from the IAB Tech Lab OpenRTB specifications: object names, field names, type notations, enumerated value sets, and section citations. The catalogs carry no spec prose. The dialect table is derived the same way, by comparing those catalogs against the field types the IAB OpenRTB protobuf schema declares. RTBlint is not affiliated with or endorsed by IAB Tech Lab. See NOTICE for attribution.

Supply chain

OpenSSF Scorecard runs weekly and publishes a public score. Dependabot covers Cargo, npm, and GitHub Actions. CodeQL scans Rust and JavaScript on every push and PR.

Three cargo-fuzz targets (validate, validate_response, validate_artf) run for 30 seconds each on every CI push. The validator must not panic on arbitrary input.

cargo +nightly fuzz run validate -- -max_total_time=60

License

Apache-2.0. See LICENSE and NOTICE.

Research

Sekowski, A. (2026). How Machine-Checkable Is OpenRTB? Classifying the Normative Content of the Protocol That Clears Real-Time Advertising. Preprint. DOI: 10.13140/RG.2.2.27937.57448

Sekowski, A. (2026). Measuring OpenRTB Dialects in Client-Side Header Bidding. Preprint. DOI: 10.13140/RG.2.2.26572.78720

Sekowski, A. (2026). Why CTV Ad Fraud Keeps Working: A Verifiability Analysis of the Connected TV Supply Chain. Preprint. ResearchGate. Field inventory extracted from this catalog. Artifacts: github.com/aleksUIX/ctv-verification-gap.