A video game where you play as a misaligned AI, deceiving and building power. cameron.tngl.io/misaligned

Question: scaled intel summaries — small-model generated, pre-authored pools, or leveled plot text? #9

Closed cameron.stream
Closed opened by cameron.stream

Question #

When intel is surfaced above its native altitude (hall/campus/region zoom), where do the summary sentences come from?

  1. Small-model AI generation at processing time (dynamic, infinite variety, nondeterministic; needs an inference dependency and a determinism/replay story)
  2. Pre-authored summary pools per intel class and level (deterministic, authorable taste, bounded variety; the message-variety rail plots.md already allows)
  3. Leveled plot/authored text — intel levels come only from authored content that declares its altitude (strongest curation, narrowest coverage of emergent intel)
  4. Something else (say what)

Why this is open #

wiki/interface/digital-read.md (adopted 2026-07-15) decides that intel items carry a level and that zoomed-out frames surface only at-altitude sentences ("China just launched" scale reads at region zoom). The level field and altitude gate are decided; the authoring source of the aggregate sentence is a taste call Cameron explicitly left open in the design session. Determinism law (no wall clock/RNG surprises in Sim; replay/save stability) constrains option 1 unless generation happens outside the sim boundary and is persisted as content.

Options #

Option Serves Strains
1. Small-model generation Infinite scale texture Determinism guardrail, offline builds, taste control
2. Pre-authored pools Determinism, authorial voice, canon gate Authoring volume at B3 breadth
3. Leveled authored plots only Tightest curation Emergent intel (non-plot events) gets no high-altitude voice

Recommendation #

Option 2, with option 3's declarations riding it: pre-authored pools keyed by intel class and level, filled at processing (deterministic pick like the existing authored-span pool mechanism), and plot-authored intel declaring its own level. Revisit small-model generation as an out-of-sim content pipeline (option 1 as a build step, not a runtime).

What your answer unlocks #

digital-read.md's LOD clause loses its [OPEN]; intel.md gains the level-assignment rule at processing; the digital-read work order's criterion 6 fixtures get their sentence source; and B3 regional intel stops being blocked on an authoring decision.

Sign up or Sign in to add to the discussion.