Recorded: Hosted Consumer and Response Table Acceptance Matrix Returned

A proof of concept is most useful when it says exactly what ran—and just as clearly what did not.

Between September 27 and 28, 2026, an authorized owner session exercised the hosted Qensai consumer surfaces against an existing organization. The run created one labeled tenant task, read it back on the consumer Relay Desk and standalone Response Table, switched organizations to test projection boundaries, and compared the hosted behavior with the operator-side Response Table evidence.

What actually ran

The hosted tenant task form created task 4d82bfc0-f3f3-410a-a0a4-e5d6226eab8c for Voss Integrity through the manual consumer path. AI Translator was off, Gate Zero was not invoked, and no agent was dispatched. The same task then appeared on both hosted consumer surfaces from the same tenant-task source.

Fresh read-only checks tested foreign, malformed, and missing organization headers. Foreign organization access failed closed. Switching between the operator’s two owner memberships changed both surfaces without leaking the PoC task across organizations. The supporting implementation slice also passed 111 isolated tests across 16 files.

What the evidence supports

The evidence supports these bounded claims:

• an existing authorized owner session could create and list a hosted consumer task;
• the same task appeared on the hosted Relay Desk and Response Table;
• organization switching changed the projected task set without cross-organization leakage;
• the operator-side Response Table stayed within AIAGF and preserved the selected consolidation canvas through the recorded navigation and refresh path;
• the hosted shell remained operational after rollback, with the send-off banner and tenant table intact.

These are acceptance slices, not a claim that every consumer role and every navigation path is complete.

What stayed deliberately held

Hosted agent send remained off for the entire proof. No provider call, paid dispatch, Gate Zero execution, duplicate task, new relay topic, entitlement broadening, or publication action was performed. That hold is a rule, not a failed test.

The proof also did not establish invited or non-owner behavior. Both available memberships were owner memberships, so the invited P2 walk and member-role projection remain open. On the hosted consumer shell, a task deep link survives reload, but a click-selected task is not written back to the URL and resets after a page round trip. Forced hosted error rendering and several optional loopback checks remain test- or source-backed rather than live-exercised.

Provenance

This draft is sourced from QCSC signal fbd8d347ed1d48d3 and the acceptance matrix TAG-Relay-Fable-Hosted-Consumer-RT-Acceptance-Matrix-2026-09-27.md (15,098 bytes; SHA-256 61705435f0faa918b7e7887df340fdbb23f9dde6dd66c8b7be9427b0035e0583). The matrix reuses hashed records from the hosted create/readback, independent signed-in read, attended sequential run, rail-click correction, and first-pass verification, then adds fresh hosted API and navigation checks.

The result is deliberately modest: the hosted consumer and Response Table paths demonstrated real, bounded behavior while retaining their holds and naming their remaining gaps. That is the point of the record.