Given the three-file (medical / legal / artistic) case-narrative frame, how can we make specific scene types—interviews, critiques, hearings, therapy sessions—double as both clue-delivery and live renegotiation of a PC’s record, and what lightweight mechanics (e.g., one-question challenges, limited rewrites, or tag swaps) best turn those scenes into tense, player-driven contests over how their brush with the King in Yellow is officially defined?

Answer

Use each file-scene as a short, contested mini-game: one new fact or tag enters the record, and players get a small, risky chance to bend it.

1. Core procedure for any file-scene (MED / LEGAL / ART)

• Step 1: Frame
  • GM states: which file is in focus, what’s at stake in 1 sentence ("Is this an episode, a crime, or a masterpiece?").
• Step 2: Live record question
  • GM asks exactly one "record question" about the PC, e.g.
    • MED: "Did you see something that wasn’t there?"
    • LEGAL: "Were you in control when you struck him?"
    • ART: "Did you intend this piece to invoke Carcosa?"
• Step 3: Player stance
  • Player picks a stance: CONFIRM / CONTEST / REFRAME.
• Step 4: Light resolution
  • Roll or card (system-agnostic):
    • Edge for CONFIRM (easy) / neutral for REFRAME / penalty for CONTEST (hard).
• Step 5: Record update
  • On a hit: player writes or edits a 1-sentence file entry + 1–2 tags.
  • On a miss: GM writes it, player may strike or swap 1 tag only.
• Step 6: Clue hook
  • Any new entry must name or imply 1 concrete investigative lead (NPC, place, object, motif).

2. Scene-type tweaks

• Interviews (LEGAL)

  • Extra option: "Go Silent" → no new statement, but GM adds a negative tag ("uncooperative") and dangles a stronger clue.
  • Good for pushing risk vs self-protection.

• Hearings (LEGAL panels, boards)

  • Each PC faces 2–3 rapid-fire record questions; group can assist once per scene to flip one miss to partial (but share a bad tag like "colluding").

• Therapy / consults (MED)

  • Add an "Honesty" toggle: if player marks the scene as "truly honest," they get advantage on REFRAME for this scene, but GM may add a deeper, more intimate tag.
  • On a strong hit, player may remove 1 old MED tag when adding a new one (apparent recovery that might be Carcosa masking itself).

• Critiques / reviews (ART)

  • Always create 1 public-facing tag ("cult favorite," "dangerous content") that can snowball into social or legal consequences.
  • On a hit, player chooses audience: peers, public, or gatekeepers; that choice decides which future file (ART / LEGAL / MED) auto-gets the next update.

3. Lightweight mechanics that create tension

• One-Question Challenge (per scene)

  • Limit each file-scene to exactly 1 high-stakes record question per PC.
  • Outcome must:
    • Add or change exactly 1 sentence in the file.
    • Attach at least 1 tag that can be invoked later for advantage/penalty.

• Limited Rewrites (per session)

  • Each PC gets 1 "revision token" per session:
    • Spend in any file-scene to:
      • Cross out one prior sentence and replace it with a new one, OR
      • Move a tag from one file to another (e.g., "unreliable" from LEGAL → MED as "dissociative").
  • GM mirrors: each time a player uses a revision token, GM may introduce 1 new Carcosa-colored discrepancy between files.

• Tag Swaps as micro-bargains

  • When the PC concedes on a point ("Fine, write that I hallucinated"), they can swap one hostile tag for a more useful but still risky one, e.g.:
    • "dangerous" → "high-risk asset"
    • "delusional" → "experiencer"
  • Swapped tags give mechanical edge in some scenes (e.g., "asset" helps with spooky agencies) and penalties in others (more pressure, surveillance).

4. Making it feel player-driven, not fiat

• Keep the numbers tiny
  • Max 3–5 tags per file; when adding a new one past the limit, something old must be erased or transformed.
• Keep authorship visible
  • Color-code or initial who wrote each line (GM vs player); clashing colors on the page are the visible "identity erosion".
• Tie SAN / stability to gaps
  • Each time two files label the same incident in mutually exclusive ways (e.g., "psychotic break" vs "performance piece"), mark a small SAN stress.
  • Players can choose to resolve a gap in a later file-scene by accepting one version as "true" → clear that stress but lock that file entry.

5. Safety overlays

• Lines & Veils on file content
  • Before play, list topics that cannot appear in institutional records (e.g., real diagnoses, self-harm specifics).
• Opt-out control
  • Any player can declare a scene "low-stakes": still delivers clues, but cannot change or add permanent tags; useful if real-life content feels too close.

Used this way, interviews, critiques, hearings, and sessions always:

• Deliver 1 clear lead.
• Change 1 small but concrete thing in a PC’s official story.
• Ask the player whether they will bend, fight, or redirect the narrative about their encounter with The King in Yellow.