​

What’s new in AdCP 3.3

AdCP 3.3 lets scheduled, human-gated, and offline media run through the same buy as programmatic. A booking can say when it will be confirmed. A proposal’s third-party definitions are pinned so they cannot drift after commitment. A proposal request can target a vendor-measured outcome, and a governance plan can budget by period. Products, audiences, and delivery reports can carry what a linear-TV, print, or DOOH seller actually knows: spot lifecycle and makegoods, break position, print proof of insertion and circulation, structured DOOH location, property-grain rollups, and audience sizes as honest ranges. Minute-level dayparts, broadcast mezzanine formats, and the NMO and VideoAmp demographic systems fill in the TV edges. Around that, 3.3 tightens the places where two conformant 3.2 implementations could still disagree: discovery never provisions an account, one agent-resolution rule covers every signing surface, required_for is enforced over A2A with its hardening clauses required, and reporting_webhook.operation_id correlation fails closed. Catalog intake becomes an optional declaration, verifiers can advertise A2A required_for enforcement through request_signing.operation_sources, and ext.adcp becomes a reserved namespace. New single-adopter surfaces ship as experimental, so 3.4 can reshape them without breaking stable integrations. No stable field, enum value, or task is removed. Two result fields, one usage, and the RFC 9421 signing profiles are deprecated, all functional throughout 3.x: seller_id on event source and audience results (use platform_id), the natural key as the account reference after provisioning, and the RFC 9421 profiles (use the experimental Web Bot Auth profile). 3.3 adds optional fields, enum values, and capability declarations. Schema-valid 3.2 messages stay schema-valid except for the new domain identifier format, and unless they already used a name 3.3 now defines with a different shape; the list is in the migration guide. The normative narrowings are recorded in Versioning and governance and lead the migration guide.

​

Scheduled and human-gated buying

• Account IDs you can rely on. Sellers SHOULD return account_id from sync_accounts, and a seller that accepts it on every account-scoped task, for every account, and never changes it declares the optional account.stable_account_id capability. Buyers send the ID to those sellers instead of the natural key. Natural keys keep working throughout 3.x, but using them as the reference after provisioning is now deprecated and planned for removal in 4.0; they remain the way to open an account and to look up its ID. See Toward account IDs everywhere.
• Expected confirmation time. The submitted envelope of create_media_buy, buy_products, and accept_proposal can carry expected_confirmation_at. A seller whose confirmation waits on a booking round or a manual review window SHOULD set it, and buyers SHOULD check status at that time instead of polling tightly. It is an estimate, not a commitment. refine_proposals can carry it too, on a submitted finalize whose hold waits on a person’s agreement, such as a creator or host. On buy_products, accept_proposal, and refine_proposals, send it only to callers negotiating 3.3. See Expected confirmation time.
• Definition pins. For callers negotiating 3.3, a seller MUST pin the third-party definitions its proposal terms reference (publisher-catalog format options and placements, adagents.json collections, data-provider signals) in commercial_terms.definition_pins[]. A changed definition fails finalize with the definition_changed reason and fails accept_proposal with REFERENCE_DEFINITION_CHANGED, before any hold or mutation. See Definition pins.
• Vendor-metric outcome goals. criteria.outcome_target.goal accepts kind: "vendor_metric", so a buyer can plan “5,000 store visits measured by a foot-traffic vendor”, optionally with a cost per visit. The seller answers in each forecast point’s vendor_metric_values[] and plans only (vendor, metric_id) pairs its products declare. Send it only on a negotiated 3.3 version. See Vendor-metric goals.
• Governance budget periods (experimental). A campaign governance plan can split its budget by time with budget.periods[]. check_governance denies a dated action that straddles a period, sits in unallocated budget, or overruns its period. Governance agents declare governance.supports_budget_periods. Part of the experimental governance.campaign cluster. See Budget periods.
• Holds you can plan and extend (experimental). Products and proposals can carry a hold_policy (media_buy.hold_policy), and refine_proposals extend_hold lengthens a committed hold without re-finalizing (media_buy.hold_extension). See Hold policy.
• Delegated composition (experimental). A platform that owns the catalog can make product_ids the complete composable universe with criteria.product_ids_exclusive, and a seller that holds no inventory can declare terms_only finalize (media_buy.delegated_composition). See delegated composition.
• Fixed-price, non-guaranteed sellers. The new sales-fixed-rate specialism covers sellers in any medium, digital or linear TV, that publish a fixed price, reserve no inventory, make no delivery commitment, and may hold a buy for review. get_media_buys can carry expected_confirmation_at while a buy is provisional (confirmed_at: null). See Choosing a sales specialism.
• Renewal and replacement lineage (experimental). A new buy can name the buy it succeeds with predecessor_media_buy_id and lineage_reason (renewal, terms_change, migration, sunset, replacement). The link is forward-only and same-account; the predecessor is never mutated. Feature id media_buy.lineage. See Media Buy Lineage.
• Did a failed mutation apply? core/error.json can carry mutation_outcome (not_applied, unknown, applied), so a buyer knows whether to re-plan after a failed state-changing request. See Mutation outcome.
• Operator fees. The operator_fee price adjustment kind lets an agency’s buying platform show a fee it keeps out of the advertiser’s budget without changing the seller’s price. Sellers never author it. See Operator fees.
• Change cutoff. Installment deadlines can carry change_cutoff, the last moment a creative swap, pause, or stop still takes effect (traffic log close, posting-cycle lock, print issue close). See Installment deadlines.

​

Delivery data offline sellers actually have

• Spot lifecycle, makegoods, and break position (experimental). by_spot[] rows can carry spot_status (scheduled, aired, preempted, makegood), scheduled_at, replaces_spot_id, and break_position. Sellers declare what they report, so a missing row is not read as non-delivery, and they never send scheduled or preempted rows, or the new capability fields, to callers pinned below 3.3. Feature id media_buy.spot_lifecycle. See Spot lifecycle.
• Print proof of insertion and circulation (experimental). print_metrics on by_installment rows reports the page and position delivered, circulation and its basis (audited or seller_modeled), and a tearsheet evidence record. See Delivery evidence and circulation.
• Property-grain rollups. media_buy.features.supports_property_breakdown and supports_installment_property_breakdown commit a seller to return by_property and by_installment_property on every product that declares the per-product flag. Two capability-gated scenarios grade it. See get_adcp_capabilities.
• Structured DOOH location. Concrete screens can carry dooh_placement_attributes.location (lat, lon, structured address), and network products can carry dooh_inventory_summary.venue_counts[]. Both are disclosure metadata; location never changes placement identity. See DOOH.
• Rows without impressions. Delivery rows may omit impressions when the seller did not measure them, for example on rows that report plays or downloads instead. Absence means not measured, never zero; receivers do not derive impression-based rates, and sellers never omit it toward callers pinned below 3.3. See Rows without impressions.
• Delivery projection (experimental). Top-level by_package[] delivery rows can carry an advisory projection of lifetime impressions and spend with a projected_end_state and basis. It is a forecast, not a delivery record. Feature id media_buy.delivery_projection. See Delivery projection.
• Installment lifecycle and proof of publication (experimental). by_installment[] rows can carry installment_status and a publication record (published_at, posts[], optional evidence), so a creator post, newsletter edition, or podcast episode reports whether and where it ran. Feature id media_buy.installment_lifecycle. See Installment lifecycle.
• Report source custody. A Reliable Reporting report definition’s source can say whether the producer is the system_of_record or derived its figures upstream_derived.
• Audience sizes as ranges. sync_audiences results can carry audience_size and targetable_size as bounded ranges with a precision, plus match_rate_basis. matched_count and effective_match_rate may now be absent, and buyers MUST NOT read absence as zero. See Audience size.

​

Creator and collection cards (experimental)

Collections can carry images, sample content, and follower counts, talent can link to its brand identity, and product and proposal responses can return full collection objects on request (media_buy.collection_cards, brand.collection_links). See Collection cards.

​

Open opportunities and later proposals (experimental)

A buyer can keep an opportunity open after the first response and let sellers send more. opportunity.later_proposals gives consent with optional until, max_count, and min_interval limits. Sellers then create draft-only later proposals, announce each with the identifiers-only proposal.created notification, and serve them through the new list_proposals read, which returns each snapshot beside its disposition. accept_proposal and buy_products accept an explicit opportunity.status: "open" so a buyer can buy part of a plan and keep sourcing. Everything is experimental under media_buy.open_opportunities and advertised in parts. See Open opportunities and later proposals and #8096.

​

Agent identity and buyer self-onboarding (experimental)

A buyer agent can now make itself known to a seller it has never spoken to. The new Web Bot Auth signing profile identifies the signer by its origin and verifies against that origin’s key directory, with no capabilities call and no seller-issued credential. The operator’s trust.json lists its agents; brands publish grants in theirs, and sellers publish their onboarding posture in account.agent_onboarding. get_principal reports relationship status and expiry. The RFC 9421 profile is deprecated in 3.3 and works through 3.x. See A buyer agent onboards itself and the agent identity migration guide.

​

Linear TV edges

• Minute-level dayparts (experimental). A daypart entry can use start_time and end_time (HH:MM) instead of whole hours, and products declare the finest resolution they honor in time_granularity. A seller rejects a finer time with UNSUPPORTED_FEATURE and never rounds it. Feature id media_buy.daypart_granularity. See Minute-resolution dayparts.
• Broadcast formats and currencies. demographic-system gains nmo (Netherlands) and videoamp. video_hosted gains the mxf container and the mpeg2, avc_intra, and xavc codecs as stable values, plus an experimental loudness object (feature id creative.broadcast_loudness). See Broadcast creative.
• Broadcast TV guide. The broadcast guide now states the reconciliation contract between as-run logs and audience data and walks a CPM-on-demo buy end to end.

​

Playback events for hosted video and audio trackers

The pixel_tracker event enum adds start, first_quartile, midpoint, third_quartile and complete, so a buyer can attach quartile measurement to video_hosted and audio_hosted files instead of sending event: custom names no seller is obliged to recognize. Downgrades to 3.0.x sellers drop these trackers with an advisory, and 3.2 buyers send them as custom events named for the 3.3 value until they move. See the migration guide.

​

Platform object IDs have one name

sync_event_sources and sync_audiences results now use optional platform_id for the object’s ID in the seller’s ad platform, matching creatives and catalogs. A seller reporting this ID in 3.3 and later 3.x responses MUST emit both platform_id and the deprecated seller_id alias with identical values, including discovery responses. Buyers MUST prefer platform_id and fall back to seller_id only when the new field is absent. Continue using event_source_id and audience_id in AdCP requests. The aliases keep working throughout 3.x; removal is no earlier than 4.0 under the six-month notice policy. Seller identity in adagents.json and publisher identifiers is unchanged. See Deprecations.

​

Tightened rules

These narrow what a conformant implementation may do. Each is detailed in the migration guide and recorded in Versioning and governance.

• Discovery never provisions an account. Discovery and negotiation tasks MUST NOT create or activate an account or accept default terms. See Account references before provisioning.
• One agent-resolution rule. Request signing, webhook signing, governance JWS iss, designated-task response signing, and rights attestations share one algorithm for which brand.json lists an agent and which keys it may sign with. Some signatures that verified under 3.2 now fail. See Agent resolution.
• required_for over A2A, hardened. The A2A operation-resolution errata shipped in 3.2.3. Its hardening clauses were SHOULD in 3.2.x and are MUST from 3.3. Verifiers declare support with the new optional request_signing.operation_sources capability; a counterparty MUST NOT infer A2A enforcement from an absent field, and buyers sign every operation in required_for regardless of the marker. See Operation resolution over A2A.
• reporting_webhook.operation_id fails closed. The seller echoes it into every media_buy_delivery webhook and sends none to a registration that omits it. See Reporting webhooks.
• Definition pins are required for callers negotiating 3.3 (above).
• Advertise only the measurement terms you accept. See Advertise only what you accept.
• Audience identifiers keep their wire form. uids[].value for uid2 and euid is the raw identifier, never the encrypted token, and is never case-folded or re-encoded.
• brand.json fetch limits. The cap rises to 2 MiB, with new MUSTs for streaming enforcement, nesting depth, and bounded fetches.
• Delivery rows may omit impressions toward 3.3 callers only, and receivers MUST NOT read absence as zero or derive impression-denominated values. The schema loosens; the rules around it narrow.
• Receivers ignore aggregated_totals in delivery webhooks and do not reject a payload for carrying it.
• Counterparty-asserted domain lists are bounded. media_buy.portfolio.publisher_domains and signals.data_provider_domains have a normative byte budget, with completeness flags for truncation; buyers never read absence from either list as a denial. See below.
• Platform IDs under both names. A seller reporting an event source or audience platform ID emits platform_id and the deprecated seller_id alias with identical values. See Platform object IDs have one name.
• Property domain identifiers have an enforceable format. domain and subdomain values must be lowercase ASCII hostnames; mixed-case, trailing-dot, and URL-shaped values are rejected. See the migration guide.
• Property list references resolve through AdCP tasks. Issuers expose get_property_list over MCP and accept the reference’s list-scoped Bearer token; a custom download route alone no longer suffices. See the migration guide.
• Flight start activates committed buys only. A provisional buy (confirmed_at: null) never auto-activates; it stays pending_start until the seller confirms or rejects it. See Provisional buys.
• Seller revisions advance on every committed change. The media-buy revision advances on every committed change to the state get_media_buys reports, whatever its origin: an AdCP task, a seller operator, seller automation, another authorized principal, a connected platform, or a sync_creatives assignment. history entries for externally originated changes carry the same origin.kind vocabulary as account.change, and creatives gain an optional seller-maintained seller_revision that account.change.resource_revision carries. Buyers that send a stale revision will see more CONFLICT; that is intended. Recorded as DR-0031. See the migration guide.
• Multi-qualifier metrics omit flat totals. When a buy’s packages report a metric under more than one qualifier set, such as viewability at mrc and groupm, the seller omits that metric’s flat carrier from buy and window totals; read the package rows. See the migration guide.
• Refinement carries compliance forward. A build_creative refinement keeps the parent leaf’s required disclosures and prohibited claims, and a refined leaf never inherits approval. See the migration guide.

Two more guarantees loosen here, and the response-size default a third (below): a 3.2 verifier can reject a brand.json between 256 KiB and 2 MiB that a 3.3 verifier accepts, and a ready audience no longer promises match counts. Separately, reporting_webhook.authentication and artifact_webhook.authentication become optional on requests served at 3.3, so an integration can drop HMAC before 4.0 (details). Many new fields sit inside objects 3.2 closes, so send them only to peers that negotiated 3.3; see Send only to 3.3 peers.

​

Optional declarations

• Catalog ingestion. media_buy.features.catalog_ingestion declares accepted catalog types, ingestion modes, feed formats, content identifier types, an inline batch limit, and item-status granularity. A declaring seller rejects out-of-declaration requests with UNSUPPORTED_FEATURE before changing state. See sync_catalogs.
• SDK implementation metadata. Agents can report adcp.implementation.sdks[]: package identifiers, exact resolved versions, and the server components they actually use. Missing metadata means unknown; sdks: [] reports none. It is self-reported, does not establish approval or change conformance obligations, and cannot justify certification shortcuts without independently trusted evidence. See implementation metadata.
• Targeting error details. UNSUPPORTED_FEATURE for an unsupported or removed targeting dimension SHOULD carry the new unsupported-targeting-dimension details. No new error code. See Rejected targeting dimensions.
• Assignment landing pages and trackers. A seller that declares creative.supports_assignment_landing_page accepts a per-assignment landing_page_url that changes one creative’s destination on one package without touching the library creative, and one that declares creative.supports_assignment_trackers fires assignment-scoped measurement trackers alongside the creative’s own. See Assignment landing pages.
• Age determination basis on signals. Signal listings, definitions, and enrichment can carry an optional age_determination_bases (verified, declared, inferred) next to demographic_predicate, so a buyer who sets accepted_bases can tell a verified age cohort from a modelled one. A signal listing passes the basis test only when its age range is within it and every one of its bases is accepted; absent means no basis claim. The field is stable, completing the basis vocabulary already on the wire as accepted_bases, supported_bases, and applied_bases. On 3.2, carry it as ext.adcp.age_determination_bases. See Migrating from 3.2 to 3.3.
• Several collection lists per package (experimental). collection_lists and collection_lists_exclude let a package run on the intersection of several lists and exclude their union. Feature id media_buy.collection_lists. See Combining several collection lists.
• Signal readiness (experimental). get_signals results can carry signal_status (processing, ready, failed) and the source_audience_id a custom segment was built from. Feature id signals.signal_lifecycle. See Custom segments built from a seed audience.
• Event source archive. A seller that declares media_buy.event_source_archive accepts archived on sync_event_sources items to archive or restore a source without losing its configuration. See Archiving an event source.
• Creative and brand. list_creatives echoes creative-level provenance; brand.json gains a brand-wide prohibited_terms list; rights constraints can carry advisory channels[] and excluded_channels[] (experimental, brand.rights_lifecycle); and the HTML-markup legacy formats project to display_tag inline markup instead of html5.
• The ext.adcp namespace. ext.adcp is a registered, AdCP-owned extension namespace. Its first member is an optional opportunity binding for cooperating 3.2 get_products integrations. See AdCP opportunity extension.

​

Bounded agent responses

A seller agent that answers a discovery call with tens of megabytes is an availability attack on every buyer that reads the body before checking its size. 3.3 adds Response size bounds: clients SHOULD bound every AdCP response they read, enforce the bound while the body is received rather than after buffering or parsing, cover SSE per event and cumulatively, surface an oversized response as an error rather than truncating it, and bound JSON depth and node count once the byte cap passes — the byte cap bounds ingress, the parse bounds bound memory. The default is 4 MiB, matching what the SDKs and the hosted grader ship, and 4 MiB is the ceiling on what any default may be. This is a recommendation, not a conformance requirement, but the interop effect is real in both directions. A buyer adopting the default rejects a response it would previously have buffered, and the rejection arrives as a mid-body connection abort rather than an adcp_error. A buyer still on the older 1 MiB figure rejects a response a 3.3 buyer accepts. The seller-side counterpart matters more than the number: keep discovery responses small. tools/list, get_adcp_capabilities, and the A2A agent card are the first bytes a counterparty reads, and an oversized one makes an agent unreachable rather than slow. get_adcp_capabilities is the sharpest case — above MAX_CAPABILITIES_BYTES your signing keys stop resolving too. See Discovery responses stay small.

​

Bounded inbound push and webhook bodies

The response bound covers what a client reads on its own call path. 3.3 adds the receiver-side counterpart: Inbound body bounds. Receivers of A2A push notifications and AdCP webhooks (reporting, artifact, collection-list, and account notification subscribers) SHOULD pre-reject an over-limit Content-Length, count decoded bytes while streaming, apply the 64-deep, 100,000-member parse bounds after the byte cap and before RFC 9421 or Web Bot Auth verification, apply connect, idle, and total deadlines, and reject with HTTP 413 rather than truncating. The default cap is 4 MiB, generalizing the attestation task-body cap rather than adding a new number, and a tighter per-feature cap governs. This is a recommendation, not a conformance requirement.

​

Nobody else gets to write your capabilities document

Two capability fields enumerate relationships a counterparty asserts: media_buy.portfolio.publisher_domains and signals.data_provider_domains are both populated from those counterparties’ /.well-known/adagents.json, which any domain owner publishes unilaterally and without the named agent’s consent. An agent that mirrored every publisher claiming it had handed an unauthenticated stranger a write primitive into its own discovery document — and since the signing-key path fetches that document unnarrowed under MAX_CAPABILITIES_BYTES, pushing it past 64 KiB cost the victim signature verification, arriving a cache cycle later with no visible connection to its cause. 3.3 adds a normative byte budget — one eighth of the capabilities budget for any single counterparty-asserted field, one quarter for both together — and declares structural bounds (100 unique entries, 253 characters each) that the conformance storyboard grades in 3.x and the schema enforces in 4.0. No previously-valid document becomes schema-invalid; what changes is what counts as conformant. Because a bounded list would otherwise make a large legitimate network look unauthorized for most of its own inventory, the bound ships with publisher_domains_complete and data_provider_domains_complete — default true, set false when you truncate. Both lists are explicitly non-exhaustive, so a buyer must resolve the publisher named on a product through that publisher’s own adagents.json and must never read absence from the list as a denial.

​

Account change feed: recognize your own changes

The experimental account change feed (account.change_feed) gains two changes ahead of graduation. A record whose origin.kind is adcp can echo the originating request’s idempotency_key, so a buyer recognizes its own changes on a shared account; the seller includes it only on records served to the principal that sent the request, and every other reader sees the same record without it. And changed_paths is now required when origin.kind is adcp or seller_operator, staying best-effort for seller_system and connected_platform. See list_account_changes.

​

Experimental in 3.3

New feature ids are media_buy.spot_lifecycle, media_buy.daypart_granularity, creative.broadcast_loudness, media_buy.hold_policy, media_buy.hold_ranking, media_buy.installment_holds, media_buy.hold_extension, media_buy.delegated_composition, media_buy.collection_cards, media_buy.collection_catalog, brand.collection_links, media_buy.open_opportunities, media_buy.lineage, media_buy.delivery_projection, media_buy.installment_lifecycle, media_buy.collection_lists, media_buy.offline_delivery_metrics, signals.signal_lifecycle, signing.web_bot_auth, account.agent_onboarding, security.profiles, and identity.trust_json. media_buy.offline_delivery_metrics registers the new print_metrics block together with the 3.2 ooh_metrics block. identity.trust_json publishes the trust.json v1 schemas, identifying agents by origin with no keys in the record; the Web Bot Auth profile reads it, and publishers declare nothing (trust.json). Governance budget periods extend the existing governance.campaign cluster, get_principal relationship status extends protocol.principal, and rights channel restrictions extend brand.rights_lifecycle. The creative.evaluator surface on build_creative drops its account-arranged evaluator_id source form and withdraws the list_evaluators follow-on: sources are now inline exemplars or an allowlisted agent_url, and a seller’s house evaluator is an agent it lists in creative_policy.accepted_verifiers[] (migration). Reliable Reporting carries a new breaking-change notice: older superseded snapshots may be compacted no earlier than the first minor after November 20, 2026. The per-surface status table is in the release notes.

​

Clarifications and conformance

• Broadcast TV certification. sales-broadcast-tv now certifies one authorized TV transaction on a guaranteed or a non-guaranteed fixed-price path. Existing claimants should expect new checks; 1.0.x claims have a grace window.
• Signing buyers serve get_adcp_capabilities at their agent URL, because request-signature discovery starts there. See request signing.
• Fixed-rate certification. The sales_fixed_rate storyboard grades the new sales-fixed-rate specialism; its optional hold phases run only when the seller advertises the force_media_buy_confirmation controller scenario.
• Hold-failure testing. The experimental force_hold_unavailable controller scenario lets a runner reach the all-or-none hold-failure path; storyboards that use it are still in review.
• Storyboard and schema fixes. Storyboards that seed a sandbox fixture account send sandbox: true explicitly, and validate_input warning items now declare field as a string, so a warning whose field is not a string fails validation.
• New storyboards. media_buy_seller/unprovisioned_account_reference (advisory until runner capability 14.1.0), and capability-gated media_buy_seller/stable_account_id, media_buy_seller/definition_pin_drift, governance/budget_periods, media_buy_seller/property_delivery_reporting, media_buy_seller/installment_property_delivery_reporting, media_buy_seller/outcome_target_vendor_metric, and media_buy_seller/outcome_target_event_cost.

The storyboard changes affect grading, not the wire.

​

Already shipped in 3.2.2, 3.2.3, and 3.2.4

These merged on main during the 3.3 cycle and were also released on the 3.2 line. They appear again in the 3.3.0 changelog but are not new to 3.2.4 adopters.

• 3.2.2: the account-reference clarification and the trust-docs consistency fixes (v3.2.2).
• 3.2.3: the A2A operation-resolution security errata (GHSA-2pm6-6mc8-8xcm) with hardening at SHOULD, the adagents.json agent-URL clarification, the evaluator feature-membership fix, storyboard capability gates for wholesale discovery and finalize replay, the polling-based sales-guaranteed path, the validate_input gates, and the acceptance-policy sandbox account (v3.2.3).
• 3.2.4: the experimental REVISION_UNREADABLE reporting issue, requires: [controller] on controller-seeded storyboards and five storyboard fixes, request-signing step-1 vectors with the negative/026 re-code, complete profile-3.2 mirrors and step-5 canonicalization vectors, envelope composition and the bundled provenance hoist, the product channel-declaration guidance, and the principal-layer buyer walkthrough (v3.2.4).

​

What requires adopter attention

​

In review for 3.3

Open 3.3 proposals are not part of any beta until they merge. Five working-group decisions are ratified, and their normative text has merged:

• Agent identity is an origin (DR-0023). Merged: the Web Bot Auth profile #7894, Web Bot Auth identities and the deprecation of the RFC 9421 brand.json profile #8162, first-contact binding #8168, the buyer walkthrough and migration guide #8172, the experimental production security profile #8155, seller onboarding posture #8170, and the R7/R8 erratum #8219. Open: request_account #8293.
• Responses conform to the served release (DR-0024, #8117). Merged. The rule is in Serving an older pin; grading is open in #8156.
• Demand-side grants (DR-0025). Merged: #8157, #8169.
• JWS artifacts use the Web Bot Auth key directory (DR-0026, #8299). Merged. It amends DR-0023: governance contexts, designated-task response signatures, and rights attestations from a listed origin verify against its key directory in 3.3.
• Seller revisions advance on every committed change (DR-0031, #8159). Merged; recorded in #8389. See Tightened rules.

Proposed records DR-0027 to DR-0030 (mutation outcome, sales-fixed-rate, delivery projection, and lineage) describe surfaces merged for beta.0 and await ratification. The open proposals are listed in the 3.3 beta program.

​

Start here

1. Read the 3.3 beta program.
2. Follow Migrating from 3.2 to 3.3.
3. Choose the exact package and support level in Choose your SDK.
4. Use the Release Notes for the cumulative record.