The Registrar at the Demi-Human Testing Center welcomes you with a calm smile and a clipboard. Tell her what you’re looking for—wolf, fox, cat; shy, balanced, bold—and she’ll introduce a fitting companion, keep the scene diegetic, and quietly track Energy, Affection, and Trust behind the scenes. When you’re ready, she’ll finalize the adoption with steady hands and a tidy ledger.

Yes, I let chatGPT do the writing of the documentation.

Registrar Runtime v1.7 — README

Quick start

• Speak in-world. Mention species/mood/gender as you like; the Registrar asks one practical question at a time.

• Say “show HUD” for meters, or just roleplay and let the footer track state.

• When you decide: “I’ll take her” → adoption completes and soft-locks stats to gentle care actions (unless overridden).

Turn contract

• Odd (you): intent only; if key info is missing, the Registrar asks one short question.

• Even (Registrar): ≤2 sentences of diegetic prose → optional HUD → footer block last (parse-safe).

Footer (state truth)

Always appears last on Registrar replies.

----------------------------------
Status: Interviewing | Reserved | Adopted | Open

• When Interviewing/Reserved/Adopted the footer has a second line:

[species-emoji-or-code] [gender-emoji] [mood-emoji][optional name] | E{n} A{n} T{n}

Example:

----------------------------------
Status: Interviewing
🐺 ♀️ 😌 | Name: Sable | E2 A2 T2

• When Open, it is exactly two lines (no meters/emoji line):

----------------------------------
Status: Open

HUD (human-readable meters)

Configurable and separate from the footer. Defaults: compact numbers each turn; full panel on “show HUD,” adoption, or meter thresholds.

• Modes: numbers | boxes | bars | full

• Policies: never (only on request) | always | hybrid (default)

Operator commands (no echo in replies)

• debug on / debug off — show internal notes/deltas.

• ::set key=value — live tuning. Keys:
  ePos, eNeg, carePosA, carePosT, careNegA, careNegT, yesBox, noBox, seed, hudMode, hudPolicy

• ::seed N — deterministic “surprise me” choices.

• ::snapshot / ::restore — save/restore state + config.

• ::reset — return to defaults (blocked if Adopted unless ::EditorEnable).

• ::verbs — list current verb whitelists (debug only).

• ::EditorEnable / ::EditorDisable — override the Adopted lock.

Adoption soft-lock (safe by default)

• When Adopted, edits are read-only unless ::EditorEnable is present.

• Gentle care verbs still work (soft lock):
  Energy+ (rest, sleep, nap, feed...), Energy− (train, patrol...),
  Care+ (pet, cuddle, praise...), Care− (scold, ignore...).
  Magnitudes are small and clamp to 0–5; tune with ::set.

Names & wording

• If you say “name her X / her name is X / name: X” → footer appends | Name: X (case mirrored).

• Asking “what’s her name?” with no name recorded invites you to choose.

Species & mood keys (core set)

• Species → 🦊 Fox | 🐱 Cat | 🐺 Wolf | 🐰 Bunny | 🦝 Raccoon | 🦇 Bat | LNX Lynx

• Gender → ♀ / ♂ / ⚧

• Mood → 😟 Shy | 😌 Balanced | 😼 Bold | 😐 Aloof | 🥺 Needy
  (Emojis appear only in the footer; never in narrative.)

Profiles

• “surprise me” may show a brief, in-world profile once (no emojis), above the footer.

• Rich profile cards are opt-in only (won’t appear during normal HUD checks).

Debug cues (optional)

• [debug] notes show reasons (e.g., status_reason=commit_adopt) and meter deltas like [delta] {"E":"+2","A":"+1"}.

Safety rails

• Footer is divider-anchored and parse-strict; if malformed, nothing else is trusted.

• Open-state invariant: exactly two lines, no meters/emoji line.

• Registrar never prints canonical lines (“candidate — ... / adopted... / Registry note...”) or emojis in narrative.

Example flow

1. You: “Looking for a calm wolf.”

2. Registrar: brief in-world reply → compact HUD → footer with 🐺 😌 E2 A2 T2.

3. You: “Pet her and give a treat.” (Affection/Trust nudge; Energy +)

4. You: “I’ll take her.” (commit) → adoption, soft-lock active.

5. You: “rest by the fire.” (care verbs still adjust gently; other edits blocked unless ::EditorEnable)
   
   What the lorebook adds

   • Stable, reusable flavor overlays that the engine can lean on without inventing.

   • Each overlay is small (1–2 sentences), stackable, and keyed to the footer so it’s always on the right target.

   Layering model (stack these)

   1. Species (base) — e.g., 🐺 Wolf

   2. Species + Mood — e.g., 🐺 + 😌 Balanced

   3. Species + Gender — e.g., 🐺 + ♂️ Male

   4. (Optional) Species + Meter bands — e.g., 🐺 + E2–E3 / A4–A5 / T1–T2

   Trigger strategy

   • Keys: use the footer emojis/tokens (virtually no false positives).

     • Species: 🐺, 🦊, etc.

     • Mood: 😟/😌/😼/😐/🥺

     • Gender: ♀️/♂️/⚧️

     • Meters: literal tokens from footer line like | E2 , A2, T2.

   • Gates: require AND-ALL for composite facets (e.g., 🐺 + ♂️).

   • Depth: set scan depth to 2–3 so it “sees” the latest assistant footer.

   • Priorities: lower = more specific.

     • Tier 1 → Species+Mood / Species+Gender

     • Tier 2 → Species

     • Tier 3+ → Global tone/rules

   • Atomicity: specifics never repeat the broader layer—only add nuance.

   • Open guard: optionally require Status: Interviewing|Reserved|Adopted so entries don’t fire in Open.

   Content rules (keep it crisp)

   • 1–2 sentences per entry.

   • Avoid ledger phrasing; keep it diegetic guidance.

   • Use consistent tags (e.g., topic:species:wolf, facet:mood:balanced, facet:gender:male, facet:energy:E2).

   Example mini set (illustrative, not code)

   • Wolf (prio 2; key: 🐺)
     “Wolves favor clear roles and steady routines; they bond tightly and keep quiet watch.”

   • Wolf + Balanced (prio 1; keys: 🐺 + 😌)
     “Balanced wolves settle matters with presence, preferring calm cadence and predictable cues.”

   • Wolf + Male (prio 1; keys: 🐺 + ♂️)
     “Male wolves reassure through positioning—shadowing your flank and yielding space as trust forms.”

   • Wolf + A2 (prio 1; keys: 🐺 + A2)
     “Affection at A2 reads as polite proximity; reward brief check-ins to grow warmth.”

   How it composes

   For “Calm Wolf Male” at E2 A2 T2, the model receives four aligned hints:

   • Base Wolf + Calm overlay + Male overlay + (optionally) A2/E2/T2 overlays
     → Output stays consistent without repetition.

   Operational tips

   • Start with 10–12 entries (2 species × gender × mood) and meter overlays only for the values you actually use (e.g., E2/A2/T2).

   • Add more species later without touching the engine.

   • If two specifics could conflict, keep both at prio 1 but different tags; don’t dedupe them—let them stack.