L3 · Output
Commission desk and output studio
What the platform produces. A named human lodges a specification at the commission desk, or a standing trigger fires, and the agents return a finished artifact: a position, a launch kit, a filing, a risk register, a stakeholder briefing. Each specimen is shown in full, with its agents, gates, Canon reads and writes, and what it changed.
Specimens
8
Output classes
5
Brief types
6
Triggers
9
Five things this function ships
Communications and Corporate Affairs is an output business
What leaves the department is language, campaigns, positions of record, risk judgments and relationships managed on the record. Those five classes are what the agents produce and what the outcome ledger measures.
Statements, releases, Q&A packs, executive copy, owned-channel language. Publishable text in the house voice, with every claim traced to an evidence record.
Launch kits, message houses, market-by-market activation plans, creative briefs and the claim substantiation that has to survive a regulator reading it.
Consultation responses, comment letters, testimony preparation and positions of record — checked for contradiction against every live position in every other jurisdiction.
Exposure registers, pre-mortems, hostile-panel simulations, materiality scores and the control set attached to each. The work that is supposed to be boring.
Stakeholder engagement plans, legislator and regulator briefings, franchise-system cascades, coalition mapping and the record of who was told what, when.
Humans in the loop · origination
The commission desk
A human decides what the company works on, in a structured form. The desk accepts six kinds of brief, each with required fields, explicit exclusions, a deadline and a stated condition for refusing the work.
Position request
We need a defensible corporate position on a live topic, written down, with an explicit block of what we will not say.
Scope is the difference between a position and a press release. Markets outside scope inherit nothing.
The simulation panel is built from this list. Omit an audience and you will not hear their objection until it is public.
Written into the Narrative Canon as a hard block. Every downstream draft is checked against it.
An unsourced claim cannot clear G2. Naming the evidence up front avoids a gate failure at hour six.
Sets the gate clock and tells the Chief of Staff whether to run full simulation or the three-class screen.
- The position is versioned with an effective date and an expiry.
- A market variant may soften emphasis. It may not contradict the claim.
- Nothing publishes until the position exists as a Canon record.
Refused if the "what we will not say" block is empty. A position without a boundary is a talking point, and talking points drift.
- 01Trigger raised
position.requested
- 02Workflow entered
WF1 · Value perception dispute
- 03Autonomy ceiling appliedA3
The highest level any step in this workflow may reach. Set in data by the accountable executive, not in code.
- 04Gates that will fireG1G3
Named humans hold these. Work stops at each one until it is cleared or the deadline escalates it.
- 05Artifacts produced
Message house · Simulation result · Evidence pack · Canon position record
A brief is a contract, not a prompt. The desk will not accept free text — the required fields exist because the agents cannot infer scope, exclusions or evidence standing, and a missing field is how a confident wrong answer gets made.
Someone accountable lodged a spec at the desk. The trigger carries their name, their deadline and their exclusions.
A condition the platform watches for continuously. No one has to be awake for it. A human still holds every gate downstream.
Live production
Lodge a brief, run it, edit it, clear it, release it
Lodge a specification and the named agents run it. Each section carries the agent that wrote it, its confidence and its evidence. Edits are scored against the machine original. Clearing at the gate freezes the artifact; releasing it starts the market cascade.
Commission a real artifact
This is not a preview. What you write here goes to the production agents, comes back as finished text, and is held against your name in the record until somebody clears it.
- The position is versioned with an effective date and an expiry.
- A market variant may soften emphasis. It may not contradict the claim.
- Nothing publishes until the position exists as a Canon record.
Output studio
What came back
Eight finished specimens across all five output classes and all five workflows. Each opens as the artifact as delivered, followed by its provenance and the outcome it was measured against.
Value perception: the message house for franchisee pricing discretion
Scope: US system only. Must survive: franchisee association, value-seeking customer, equity analyst, business press. Will not say: no attribution of pricing outcomes to individual operators, no forward price commitment.
The claim
Our value platform is designed to be affordable everywhere it is offered. Where that has not been the customer experience, the design let us down, and we have changed the design.
Support
- The national value construct is now specified with a price ceiling rather than a suggested price, removing the discretion that produced dispersion.
- The removed digital deal is reinstated as a permanent feature of the app rather than a promotional cycle.
- Operator economics on the value platform are being addressed directly, so affordability does not depend on an operator absorbing the cost.
Proof
| Claim element | Evidence record | Confidence |
|---|---|---|
| Value-menu design permitted upward pricing | Q2 earnings acknowledgement, company disclosure | Certain |
| Two-thirds of the US traffic shortfall attributable to this and the digital-deal removal | Company statement, same disclosure | Certain |
| Dispersion narrows under a ceiling construct | Menu-price index, four-market pilot, eight weeks | Likely |
What we do not say
- We do not name or characterize the pricing behavior of individual operators.
- We do not commit to a future price point or a duration.
- We do not describe the franchise system as the cause. The design was the cause.
This block is enforced in generation. Downstream copy that violates it fails validation before a human sees it.
Adversarial readings tested
- “They are blaming their franchisees.” — Mitigated by moving causation onto design in every variant. Residual risk: medium, trade press.
- “They admitted the value platform does not work.” — Mitigated by the reinstatement proof point carrying the sentence. Residual risk: low.
Rule R2: no output leaves the platform without a Canon write-back. The write is what makes the next situation cheaper than this one.
Four weeks after issue, three highest-dispersion markets
No market issued language contradicting the position
Question asked matched a simulated reading almost verbatim
Against a 72-hour cascade window
The position held because the exclusion block was written before the copy, not after the first draft came back wrong.