Editorial Register Classes

Three editorial register classes were validated during the triangulate Phase 0 experiment. Each register has different conventions for lede shape, attribution density, conditional vs declarative reasoning, and the structural treatment of "settled vs open" framings. The House-Style Editor agent in Phase 2 should select register from topic class; producing a single house style across all contexts would warp output toward a register that doesn't fit.

The classes

Political-news

Exercised by: Topic 1 (AB 2624 status), Topic 2 (MN childcare freeze 4-month status), Topic 4 (PA Medicaid fraud).

Conventions:

• Short attribution sentences.
• AP-style hedging on contested claims.
• "Documentary vs punitive" framing distinction for federal-pressure stories (e.g., "Congressional pressure remained documentary; no enforcement action was taken").
• Careful R/D / partisan bias-flagging on source materials.
• Inline direct quotes from public officials in their official capacity.
• Methods sentence + audit-log link at the end of the article body (one line, not a section).
• No ## Background, ## Status, ## Analysis subheads — these are wiki-style; news-style integrates background and analysis into the body.

Source-class profile: state-AG press releases, OIG reports, court filings, mainstream local/national journalism, R-aligned and L-aligned outlets named explicitly when used.

Financial-thesis

Exercised by: Topic 3 (TSMC → Intel re-rate mechanism).

Conventions:

• Thesis-statement lede (not event lede).
• Explicit conditional reasoning preserved — "if X, then Y" rather than "X causes Y" — because investment theses are inherently probabilistic.
• Falsification-triggers section in-body, not buried in the audit log.
• Verb-strength control particularly tight on stage-of-commitment claims ("preliminary" never "chose"; see transferable-f3-misinterpretation-patterns).
• Bear case at structural weight, not as a closing asterisk.
• Asymmetric-resilience notes (e.g., "picks-and-shovels less foundry-sensitive than the Intel-specific re-rate") made explicit.
• Position-disclosure as publication-blocker for any external publication. Internal v1 articles omit; external publication requires positions held / no positions / disclaimer.

Source-class profile: corporate earnings calls and IR, industry press (often Korean industry coverage in translation), Western financial press, vault autoresearch outputs inheriting their underlying bias-load.

Serious-history feature

Exercised by: Topic 5 (Bronze Age Collapse).

Conventions:

• Lede frames the scholarly state of the field, not the dramatic event itself.
• Explicit "settled vs open" closing structure — what consensus has reached, what remains under active scholarly debate.
• Popular-vs-scholarly framing flagged when the popular version diverges from the scholarly one ("the popular version of the story is no longer the scholarly version").
• Primary-source naming with dates (e.g., "Pharaoh Merneptah's fifth-regnal-year inscription, dated about 1207 BCE").
• Time-distant primary sourcing — Egyptian inscriptions, archaeological reports, paleoclimate proxies. No "the administration said today" sourcing; everything is mediated through scholarship.
• Two parallel storylines treated at equal structural weight when both are load-bearing for the current scholarly state (e.g., the multi-stressor causal framing AND Millek's destruction-scale revision in the Bronze Age article).
• Avoid the "civilizations crumbled" / "mysterious collapse" infotainment register; treat as news about scholarly progress.

Source-class profile: peer-reviewed papers, academic primary articles (ASOR, BMCR), serious-popularizer synthesis (ACOUP, Wikipedia for canonical synthesis), scholar interviews.

Phase 0 evidence

Topic 1 needed a v2 House-Style pass to convert wiki-style output to news-style; Topics 2–5 produced single-pass news-coded output by baking the discipline into the render rather than treating it as a sequential gate. The Topic 1 → Topic 2 transition was where the architectural lesson landed: House-Style is a render-time constraint, not a post-render gate.

From 2026-05-13-phase-0-retrospective: "The House-Style Editor will need exemplar corpora for each register class. Register selection should be driven by topic class, not produced as a single style across all contexts."

User-confirmed input on Topic 3: "Editorial style can probably still use some tweaking and this adds a new context to tune on (financial vs political)." Topic 5 added the third register.

What's not yet validated

These register classes are not exhaustive. Phase 0 did not exercise:

• Scientific-research feature — articles about specific peer-reviewed papers (genetics, climate science, medicine). Likely some convergence with serious-history feature but with stricter primary-paper citation conventions.
• Business / corporate feature — articles about company strategy / industry structure that aren't investment theses. Tonal middle ground between political-news and financial-thesis.
• Sports / cultural feature — patia, the user's other project, would test this; outside-domain test on Topic 5 chose history instead.
• Briefing / explainer — short-form explanatory articles synthesizing a complex topic; different conventions than feature-length.

See what-register-classes-emerge-next (placeholder; create when the question becomes load-bearing).

Implications for Phase 2

The House-Style Editor agent should:

• Accept a register class as input parameter (selected from topic class by the orchestrator).
• Have separate exemplar corpora per register — ~30 articles each per the Plan.
• Treat register conventions as render-time constraints encoded in the system prompt, not as post-render rewriting passes.

The orchestrator's topic-class → register-class mapping is itself a tunable decision; some topics may span two registers (e.g., a corporate-strategy story could be either business feature or financial thesis depending on whether it makes a price-direction call).

Evidence

• From 2026-05-13-phase-0-retrospective: "Three editorial register classes were validated during the triangulate Phase 0 experiment."
• From 2026-05-13-phase-0-retrospective on Topic 1 architectural lesson: "House-Style discipline baked into render, not run as a post-render gate."
• From 2026-05-13-phase-0-retrospective on Topic 3 user input: per-register tuning surface confirmed as load-bearing.

Related

• transferable-f3-misinterpretation-patterns — verb-strength patterns vary by register
• triangulate-4-phase-application-protocol — where the House-Style Editor sits in the pipeline
• vault-two-functions-substrate-and-methodology — register conventions are part of the methodology layer