Newsroom

Appendix E — The Newsroom Principle

E.1 Purpose of the analogy

The newsroom model is useful because this architecture is fundamentally concerned with epistemic separation: who may observe, who may commission work, who may author a solution, who may verify it, who may judge it, who may integrate it, and who may later correct or withdraw it.

A single controller that observes the host, diagnoses the deficit, selects a topology, generates parameters, evaluates itself, approves deployment and decides retention is equivalent to a newsroom in which one person assigns the story, rewrites the notes, authors the conclusion, fact-checks it, publishes it and destroys the corrections file. It may be fast, but neither evidence nor failure attribution remains trustworthy.

The newsroom analogy therefore explains why the architecture contains more roles than a minimal implementation apparently requires. It is not bureaucracy for its own sake; it is a defence against authority collapse.

E.2 Role mapping

System role	Newsroom analogue	Legitimate question
Tamiyo	Editor-in-chief / managing editor	Which desks, beats and investigations receive resources?
Narset	Assignments editor	Is there a story here, which region owns it, by when, and under what scope and budget?
Nissa	Reporting, photography and data desk	What was actually observed?
Momir	Writer / investigative journalist	What coherent candidate follows from the evidence and assignment?
Elesh	Copy, standards and house-form desk	Is the submission structurally legitimate, coherent and conformant?
Tezzeret	Production, layout and publishing pipeline	Can canonical copy become an executable edition without changing meaning?
Tolaria	Newsroom production environment, CMS and test presses	Where are ordinary and experimental editions produced and replayed?
Urabrask	Fact-checking and QA	Are the claims and executable behaviour supported by evidence?
Augustin	Publishing editor	Does this run, get returned, deferred, retested, rejected, or spiked?
Kasmina	Live-edition integrator	How is accepted material placed into the running edition safely?
Emrakul	Corrections, withdrawal and retirement	What published material should be sedated, corrected, deprecated or removed?
Sarpadia	Morgue and archive	What did we report, try, reject, retract and learn?
Oona	Front page, broadcast desk and presentation	What does the operator or reader see?
Leyline	Stylebook and editorial constitution	What language, records and procedures govern the newsroom?

E.3 Assignment brief versus editorial angle

The most important review test is:

Does this field specify the assignment, or does it impose an editorial angle?

Legitimate assignment fields include:

text insertion region resource class urgency class tactical deadline maturity mode assurance class

System-resolved constraints include:

text input/output tensor contract grammar profile exact budget derived from the authorised class compatible compiler and verifier versions permitted evaluation horizons

Prohibited editorial-angle fields include:

text preferred topology family deficit diagnosis suggested ancestor expected mechanism rank or width hint attention-like / convolution-like / norm-like recommendation free-form message to the designer

Narset may see Nissa's evidence because it must decide whether to commission work. It must not become the channel through which Momir sees that evidence. The assignments desk is allowed to say “investigate Region A under this scope.” It is not allowed to say “write a story proving rank collapse and conclude that attention is the answer.”

E.4 The source-routing rule

```text Nissa observes host state S ├──→ TelemetryEnvelope O ──→ Narset └──→ TelemetryEnvelope O ──→ Momir

Narset emits GrowthIntent I Leyline resolves GrowthRequest Q Sarpadia optionally emits ancestry A

Momir designs from O + Q + optional A ```

The following are constitutional failures:

• Narset forwards a rewritten observation to Momir;
• Nissa emits a recommended solution;
• Momir reads Narset hidden state;
• the request resolver infers a diagnosis;
• an ancestor is selected by Narset rather than the curriculum or retrieval policy;
• or the observation referenced by Momir differs from the one that triggered Narset's commission.

E.5 The right to spike the story

A commissioned story does not have to run. Likewise, a GrowthIntent does not imply that any candidate must be admitted.

The chain contains multiple legitimate stopping points:

• Narset may choose WAIT and issue no assignment.
• Momir may produce no structurally viable proposal.
• Elesh may reject malformed designs.
• Tezzeret may fail to compile faithfully.
• Urabrask may find runtime defects or inadequate evidence.
• Augustin may select no-op, reject, defer or require retest.
• Narset may abort stale growth before integration.
• Emrakul may later withdraw committed growth under a maintenance warrant.

This is the editorial equivalent of spiking a story, returning copy, publishing a correction, or withdrawing an obsolete article. The architecture is intentionally not a content mill that must publish something every time an assignment is raised.

E.6 Archive discipline

A credible newsroom archive contains accepted stories, abandoned drafts, corrections, retractions, failed investigations and source notes. Sarpadia must likewise retain:

• stock reference seeds and their outcomes;
• Momir children and parents;
• structural rejects;
• compilation failures;
• QA defects;
• Augustin rejections and no-op decisions;
• stale integrations;
• maintenance withdrawals;
• and scaffold-withdrawal status.

A winners-only Sarpadia is not institutional memory. It is mythology.

E.7 Where the analogy stops

The newsroom model is not a literal organisational chart and should not determine implementation details by itself.

• Nissa may be automated and highly mathematical; it is not a human reporter.
• Elesh's structural proofs are stronger than ordinary copy editing.
• Tolaria is both the ordinary training substrate and the counterfactual production environment.
• Augustin's utility policy is formal and versioned, not editorial taste.
• Kasmina and Emrakul operate on a neural host, not a publication.

The analogy is used to explain and lint authority boundaries. Leyline schemas, dependency rules, tests, blinding and deterministic provenance remain the source of architectural truth.