When Carcosa visions are used as recurring investigative set-pieces rather than a final boss reveal—appearing in dreams, performances, and breakdowns across many sessions—what concrete guidelines help GMs calibrate how much actionable information (locations, conspirators, future events) those visions should contain versus how much should remain symbolic mood, so that players can treat them as usable but unreliable leads instead of either pure flavor or a railroady prophecy system?

king-in-yellow-rpg | Updated at

Answer

Guideline: treat every Carcosa vision as a clue packet with 1–2 usable hooks wrapped in 3–5 bits of mood, and make most specifics fuzzy or partial.

  1. Ratio: usable vs symbolic
  • Per vision, include:
    • 1 concrete hook (place, person, object, date-range) the PCs can chase.
    • 0–1 secondary hint that becomes clear only after later clues.
    • The rest (3–5 images) pure tone, themes, or red herrings.
  • Keep at least half of the “concrete” bits incomplete: missing names, half-heard lines, distorted maps.
  1. What may be actionable
  • Locations: recognizable feature + wrong or missing detail ("blue glass dome by the river" but no city name).
  • Conspirators: masks, habits, or roles, not full identities ("the man who never applauds").
  • Future events: broad situation, not fixed outcome ("an audience fleeing"; no guarantee it will, or must, happen).
  1. Avoiding railroady prophecy
  • Never make the vision the only way to learn a key fact; back it with mundane clues later.
  • Let players avert, misread, or partially fulfill visions; don’t auto-snap events to match dream imagery.
  • When a vision seems “wrong,” show why: the PCs changed things, or factions faked a Carcosa tableau.
  1. Keeping visions investigable
  • Tag each vision in prep with:
    • 1–2 leads it points toward (NPC, site, file, performance).
    • 1 false but plausible trail.
  • Ensure following any trail yields real info, even if it debunks the vision.
  • Reuse motifs: a mask, color, or line repeats across visions and real scenes, so players can cross-link.
  1. Pacing across sessions
  • Early arc: visions mostly symbolic with 1 weakly actionable hook each.
  • Middle: clearer cross-maps to known people/places; some callbacks correct earlier misreadings.
  • Late: fewer visions, more precise, but also more obviously influenced by what PCs already fear or suspect.
  1. Table-facing signals of reliability
  • Slight cues (never hard rules):
    • Stable, slow, detailed images → more likely to map to facts.
    • Chaotic, jump-cut or 2nd-person scenes → more projection, less reliable.
    • Repeated across multiple PCs or NPCs → more trustworthy overlap.
  1. Safety & agency
  • Frame visions as “partial transmissions” or “subjective encounters,” not diagnoses of PC fate.
  • Offer opt-out: allow a player to fade their character’s vision intensity if it’s stressing them, converting some detail into off-screen research results instead.

Summary slider for GMs:

  • Per vision: 1 clear lead, 1 fuzzy hint, 3–5 pure mood images.
  • Back every lead with non-vision clues.
  • Let visions be suggestive overlays, not scripts events must obey.