Skip to content
ONE3.ai

How it works · Flow F19

One goal traced end to end, across every point where software can mislead

A replayable reference journey: goal contract, mandate, baseline, consented audience, durable intent, approval, budget reservation, dispatch, provider result, verified outcome, settled cost and statement.

What the Goal to verified statement flow does

F1–F18 each describe one process. This flow traces a single goal across every boundary where a platform can silently mislead a customer: authority, consent, approval, money, an external provider, outcome identity and the statement. It is written as a replayable fixture, not a story.

The gate: no outcome-based statement, invoice or performance claim may be sent to a customer until this fixture passes end to end on a synthetic tenant with a provider simulator. It is the acceptance gate for the Investment & ROI Ledger, the rental commercial model and agency rebilling.

Worked goal. An agency client sets a goal of the form "book N qualified meetings in a period from opted-in prospects, within a One3-controlled spend ceiling". Qualification means a completed calendar event with a named prospect, an ICP rule and no duplicate meeting for the same opportunity. The client approves one outreach template and one audience snapshot. Each send is externally irreversible.

What the customer sees. One goal screen answers five questions without reading traces: what was agreed, what the system has done, what needs a decision, what happened at the provider and what value is verified. Every number links to its evidence.

Goal to verified statement, step by step

Fifteen steps, each with its Action, durable record and proof.

  1. 1

    1. Goal contract agreed

    **Strategy & Governance.** Action `strategy.goal.create`. The client accepts baseline, exclusion and correction rules before any spend.

  2. 2

    2. Authority granted

    **Platform.** Action `platform.mandate.grant` creates the client mandate (purpose, provider, spend ceiling, expiry). Effective authority = entitlement ∩ mandate ∩ certified skill ∩ credential scope ∩ policy.

  3. 3

    3. Baseline frozen

    **Strategy & Governance.** Action `strategy.goal.freeze_baseline`. While no causal evidence exists the UI says so; no lift number is shown.

  4. 4

    4. Audience prepared

    **Marketing.** Action `mkt.audience.snapshot` records per-contact consent, template version, and a recipient-set digest and count.

  5. 5

    5. Intent recorded

    **Platform.** Action `platform.action.intent` writes a durable intent. Nothing is dispatched before it exists.

  6. 6

    6. Approval decided

    **Platform · Human gate.** Action `platform.approval.decide` binds payload, audience digest, amount and policy version — not only the target row.

  7. 7

    7. Budget reserved

    **Platform.** Action `platform.budget.reserve`. The cap applies to committed plus active reservations.

  8. 8

    8. Dispatch

    **Platform · WhatsApp or email channel.** Action `chan.message.send`. Consent, authority, budget, connector health and kill switch are re-checked at dispatch.

  9. 9

    9. Provider result

    **Platform.** `confirmed` or `unknown_remote_state`. Acknowledged by the provider is not a business outcome; an unknown state is reconciled, never blind-retried.

  10. 10

    10. Reply received

    **Service Desk.** Action `desk.conversation.ingest`. Inbound text is untrusted data, never an agent instruction.

  11. 11

    11. Meeting booked

    **Office · CRM & Sales.** Action `office.calendar.book` links the event to the opportunity. One meeting cannot create two verified outcomes.

  12. 12

    12. Outcome verified

    **Platform.** Action `platform.outcome.verify` resolves webhook, import and manual entry to one canonical business key.

  13. 13

    13. Cost settled

    **Platform · AI Workforce.** Action `platform.budget.settle`. Actual cost settles against the reservation; the remainder is released.

  14. 14

    14. Value recorded

    **AI Workforce · Ledger.** Action `agents.return.record` writes an observed or assisted return entry; incremental lift stays cohort-level.

  15. 15

    15. Statement issued

    **AI Workforce · Ledger.** Action `agents.statement.issue`. Gross return, reversals, net verified return, investment and each ratio by metric id. Disputes create a new version; signed history is kept.

Acceptance scenarios (all seven must pass)

IDScenarioRequired behaviour
RJ-1Provider accepts, local commit failsRecovery produces exactly one external effect and one settled cost. Never a second send.
RJ-2Approval reused after the audience grewAn approved snapshot cannot dispatch to an expanded segment; drift forces a new preview and decision.
RJ-3Duplicate outcome by webhook and importOne canonical business key: one positive return, several evidence items.
RJ-4Refund event overtakes the won eventA refunded deal never ends in won, and no false return entry is created.
RJ-5Consent withdrawn after approval, before dispatchThat contact point is suppressed at dispatch and the audit shows why approved and executed differ.
RJ-6Budget cap reached mid-runCommitted plus active reservations never exceed the hard cap; parallel workers cannot race past it.
RJ-7Agency mandate revoked mid-runEvery queued and in-flight dispatch stops across UI, REST, MCP, agent, automation and channel, with a visible pause reason.

Caller parity

The same Action invoked from the UI, REST, MCP, an agent and an automation must enforce the same authority, risk tier and reversibility class. Every scenario asserts database state, external effect count, customer-facing text, cost movement and evidence links.

Questions about this flow

Why publish an acceptance test on a marketing site?

Because it is the rule that stops One3 sending outcome-based statements or invoices before the system is provably honest about authority, consent, money and outcomes.

What does "unknown remote state" mean?

The provider did not clearly confirm or deny. One3 reconciles by query, webhook or a bounded window, and never resends blindly.

Can approval be reused for a bigger audience?

No. Approval binds the audience digest and count. A larger segment forces a new preview and decision.

What if consent is withdrawn after approval?

The contact point is suppressed at dispatch, and the audit shows what was approved, what ran and why they differ.

When does this gate apply?

Before any outcome-based statement, invoice or performance claim is sent to a customer.

Ask for the fixture

Finance and compliance teams can review the reference journey and its seven scenarios with us.

Become a design partner