Skip to content

v2.1.0 — Verified end-to-end transactions

Choose a tag to compare

View all tags
@atc964 atc964 released this 20 Jul 23:17
v2.1.0
b452c78

v2.1.0 — Verified end-to-end transactions, spec-conformant wire, restored capabilities

The buyer agent now completes verified end-to-end deal bookings against the companion Seller Agent — proven on live runs with real LLM reasoning over real HTTP (latest verification: seller-issued deal DEMO-0B90FBE27BC7, matched against scenario expectations).

Breaking change — OpenDirect wire field names

All OpenDirect wire fields now use the published spec's lowercase names: productId→productid, startDate→startdate, endDate→enddate, accountId→accountid, quantity→qty, and enum spellings (Guaranteed→guaranteed, Cancelled→Canceled), among others. Clients using the previous camelCase dialect must update. Python-side field names are unchanged (populate_by_name=True).

New and restored capabilities

  • Meta Ads integration — campaign/adset creation (PAUSED) with Meta-issued campaign IDs, reporting endpoints (GET /reports/{job_id}, /meta/campaigns, /meta/report), ported into the v2 service architecture.
  • Mixpeek contextual enrichment — IAB taxonomy classification client, research tool, and MCP surface (now 43 MCP tools).
  • Hardened booking pipeline — research-crew orchestration fixes so recommendations reliably reach the approval gate; execution errors surface in the job record; per-line CPM ceiling enforcement now engages on briefs carrying kpis.max_cpm_usd; budget-aware approval semantics.
  • Robust persistence — JSON-safe serialization across all booking state.

Conformance

An executable OpenDirect 2.1 conformance program now measures the wire against the published spec: field-name conformance is at 100% for implemented objects; structural convergence (avails array/envelope forms, additional required endpoints) is scoped and sequenced. Current honest position: built on the OpenDirect 2.1 transaction model with a measured, improving conformance posture.

Documentation

Docs now match the code: pages for removed subsystems pruned, storage/identity/deal-library pages rewritten to current architecture, AI-client setup guides corrected (X-API-Key auth, /mcp/ endpoint, real tool names), all counts aligned to auto-generated inventories (43 tools / 14 endpoints / 38 event types).

Known gaps (tracked, sequenced)

Negotiation is implemented but not yet invoked by the real-mode booking path before walking; avails responses are honest but stateless (no inventory ledger/holds yet); structural spec convergence continues. Test suite: 3319 passed / 0 failed.

Assets 2
Loading