CONNECTOR GAPS — r1.4 cross-domain sweep

Date: 2026-08-26 Scope: the 15 sources named in whitepaper amendment r1.4 §14.1 ("Unified Connector Gateway"). This register exists because that section's draft claimed all 15 "normalize through one gateway into canonical envelopes" — code verification found that claim true for less than half of them. Method: read-only sweep of src/shared/virtuoso_models/transforms/mappers.py, miz-oki-adk-agents/kg-canonical-ingest/mappers/, services/service-marketing-connectors/, services/measurement-rails/, services/intent-*-extender/, miz-oki-adk-agents/boss/connector_gateway/tools.py, services/service-canonical-ingestion/.

There is no single named gateway

Three separate, non-unified mechanisms exist:

service-canonical-ingestion (Cloud Run) — POST /api/v1/ingest wraps payloads in the live CanonicalEventEnvelope (schema v3.5.1, contracts/mizoki_contracts/envelope.py). Callers must pre-normalize before calling it; it has no MAPPERS[source] dispatch of its own.
MAPPERS[source](raw) → ingest_gate(event, upsert_fn) — an in-process library dispatch (src/shared/virtuoso_models/journey_event.py:198; dict at src/shared/virtuoso_models/transforms/mappers.py:434-440), the "no side doors" path named in .claude/rules/03-canonical-architecture.md. This produces a different envelope shape (src/shared/virtuoso_models/schemas/journey-event.json) than service-canonical-ingestion.
miz-oki-adk-agents/boss/connector_gateway/tools.py — a Boss-agent outbound action tool registry (gmail.search, drive.fetch, github.create_issue, bigquery.query, neo4j.cypher, calendar.create_event). This is not an ingestion path; it does not touch either envelope.

Connector-by-connector status

Source	Status	Path	Evidence
Google Ads	LIVE (code)	`service-marketing-connectors` adapter + `MAPPERS`	`services/service-marketing-connectors/direct_connectors.py:511` (`GoogleAdsAdapter`); `mappers.py:434-436`
Meta Ads	LIVE (code)	`service-marketing-connectors` adapter + `MAPPERS`	`direct_connectors.py:466` (`MetaAdsAdapter`); `mappers.py:435`
SendGrid	LIVE (code)	`MAPPERS` + separate cell mapper	`mappers.py:436` (`map_sendgrid`); `miz-oki-adk-agents/kg-canonical-ingest/mappers/email.py`
Shopify	PARTIAL, fail-closed	`MAPPERS` + extender	`mappers.py:437` (`map_shopify`); `services/intent-shopify-extender/main.py` (`SHOPIFY_WEBHOOK_SECRET` gate)
GA4	PARTIAL, fail-closed	`service-marketing-connectors` adapter + extender	`direct_connectors.py:549` (`GA4Adapter`); `services/intent-ga4-extender/main.py:39-44` (fails closed on unset `GA4_EXPORT_DATASET`)
Klaviyo	PARTIAL, aggregate-only by design	`MAPPERS` + connector	`mappers.py:437` (`map_klaviyo`); `services/service-marketing-connectors/klaviyo_connector.py:145-200`; excluded from bucket path (`bucket_connectors.py:79`, "no-per-person-nodes decision")
The Trade Desk (TTD)	PARTIAL, fail-closed	`service-marketing-connectors` adapter + extender	`direct_connectors.py:693` (`TradeDeskAdapter`); `services/intent-pullers-extender/main.py` (`TTD_REDS_BUCKET` gate)
OpenRTB	IN BUILD, narrow	`MAPPERS` (journey lane only) + separate cell mapper	`mappers.py:436` (`map_openrtb`); `miz-oki-adk-agents/kg-canonical-ingest/mappers/openrtb.py`. No HTTP adapter in `direct_connectors.py` — not reachable via `service-canonical-ingestion`
BigQuery	not a source — destination	landing table for canonical events	`CANONICAL_EVENTS_BQ_TABLE` in erasure/ingestion code; never an inbound connector
Vertex AI	not a source — inference backend	LLM backend behind `virtuoso_call`	`src/shared/virtuoso_models/vendor_adapters.py:132-162`; never feeds the envelope as a data source
Graph storage	not a source — projection target	Firestore-backed KG	`services/service-canonical-ingestion/projector_kg.py` projects envelopes → KG; no reverse ingestion path found
Gmail	not wired — Boss outbound tool only	`connector_gateway/tools.py:18-27`	`gmail.search`/`gmail.read` — no wiring to `MAPPERS`/`ingest_gate` or `service-canonical-ingestion`
Drive	not wired — Boss outbound tool only	`connector_gateway/tools.py:29-33`	`drive.fetch`
GitHub	not wired — Boss outbound tool only	`connector_gateway/tools.py:34-38`	`github.create_issue`
Calendar	not wired — Boss outbound tool only	`connector_gateway/tools.py:48-52`	`calendar.create_event`

Summary

3 LIVE, 4 PARTIAL/fail-closed, 1 IN BUILD/narrow — 8 of 15 are genuine, wired ingestion connectors.
7 of 15 are not ingestion sources at all: 4 are Boss-agent outbound action tools never connected to ingestion (Gmail, Drive, GitHub, Calendar), 3 are ingestion destinations/backends misdescribed as sources (BigQuery, Vertex AI, graph storage).
No component in the repo is named "Unified Connector Gateway." The nearest real things are service-canonical-ingestion and the MAPPERS[source] → ingest_gate dispatch — two separate mechanisms with two different envelope shapes, not one gateway.

Register items

These are added to the Feature Coverage Matrix Gap Register (§10, rows G-25 through G-28):

G-25: CLOSED 2026-08-26 — content_qa didn't scope repo-root docs/marketing//docs/product/; scripts/check_canon_docs.py now gates them in CI (content-truth-gate job). (build)
G-26: Three coexisting envelope schemas (v1.0.0 superseded, v3.5.1 live via mizoki_contracts, journey-event.json via MAPPERS) have no convergence date beyond "tracked at DATA-001." (build)
G-27: service-canonical-ingestion and MAPPERS[source] → ingest_gate are two parallel, non-unified ingestion paths with different envelope shapes; no single gateway exists despite marketing prose using that phrase. (build)
G-28: OpenRTB has no HTTP adapter and is unreachable via service-canonical-ingestion — journey-lane mapper only. (build)
G-29: GitHub is not a connector at all. There is no catalog entry, no adapter, and no mapper — the only github strings in service code are a Boss-agent outbound action (github.create_issue, §3 above) and an unrelated token check in coding-moa. Raised 2026-08-31 when "bolt on proper GitHub and Google properties" was scoped: the Google half was a real, blocked gap (closed the same day — see TENANT_ONBOARDING_CHECKLIST.md §5.6), the GitHub half had nothing to bolt onto. Deferred by owner decision rather than designed speculatively; a future build needs the ingest question answered first (what GitHub activity is a marketing/intent signal, and for whom) before a catalog entry can enumerate honest fields. (build)