Accepted OpenProposal terms can drift before they reach the OpenRTB bid request

OpenProposal lives in planning. The buyer agent holds a brief covering channel, geography, budget band and format constraints. Seller agents return structured products, and the buyer agent scores fit ^[17]. Execution is a second document, written after that decision ^[6]. According to the dev.to post, nothing in OpenRTB names which OpenProposal response won ^[7]. There is no proposal_id, no pointer back to the brief version, and no guarantee that imp.video still carries the duration or placement semantics the comparison used ^[7].

Connected TV shows where that breaks. OpenRTB 2.6 added rqddurs, an array of exact acceptable durations in seconds, and made it mutually exclusive with minduration and maxduration ^[9]. The template in the post's example uses the range fields anyway ^[11]. Of the 56 whole-second lengths that range admits, 54 are lengths the proposal never scored ^[1]. The fix on the wire is one field, "rqddurs": [15, 30], with both range fields removed ^[12].

Validation does not catch the drift ^[16]. The range request is well-formed JSON ^[20]. A check against OpenRTB alone has no way to know the proposal asked for 15 and 30 ^[7]. The post recommends rtblint, a bid request and response validator it describes as independent of IAB Tech Lab and not required by the spec ^[13]. The post also says plainly that the tool validates the auction object, not the RFP story ^[16]. What rtblint does catch is the opposite mistake. Against the 2.6-202303 snapshot, a request carrying rqddurs alongside minduration returns an error: "rqddurs is mutually exclusive with minduration and maxduration" ^[15]. Leave out the optional version argument on its MCP validate_bid_request call and it checks against the latest tracked 2.6 release in the build, not necessarily the snapshot the SSP pinned ^[14].

Other agent protocols carry agreement inside the message. In AdCP, buyer and seller settle the version message by message, and a mismatch between them comes back as a typed error. AAMP 2.3 added trust verification on price-moving paths ^[8]. OpenRTB records its dated snapshot in onboarding documents, not in the JSON ^[21].

OpenProposal's encoding layer is good work on a real problem. Software cannot compare thousands of RFP responses unless every seller describes its products the same way ^[18]. I think the missing piece is small. A proposal handle and a brief version in the bid request would let a buyer or an exchange diff the emitted imp.video against the scored product before the request goes out.