The Mirror Ideation

THE MIRROR (ideation)

Status: design notes, not a commitment. This page documents docs/ideation/mirror-and-still-here.md on branch ideation/mirror-and-still-here — not merged into main, no code exists for it. The source document opens with "Design notes, not a commitment. Exploratory — bias toward volume over polish," and that framing applies to everything on this page. Nothing here is implemented; treat it the same way you'd treat Roadmap / Ideas — tentative, sourced from the project's own words, not from shipped code.

The mood anchor, and where LULL draws the line

The ideation doc opens from a single reference image: a grimy bathroom mirror whose reflection doesn't match the person in front of it — gaunt, bloodied, scrawled with "I AM STILL HERE." The doc is explicit that this image is a tonal reference only, not a target:

This image is a tonal reference only — not something LULL builds toward literally or reproduces. It is far past LULL's line (see CLAUDE.md §1: uncanny, not traumatizing). What we take from it: the gap between the person and what the mirror insists is still there.

That distinction — reference vs. target — is the load-bearing idea of the whole document, and it's worth reading The Safety Invariant and Concept & Design's "No genuine harm" principle alongside this page for the constraint it's being designed against.

Why it fits LULL

THE MIRROR is pitched as the same premise as THE EYE — a surface that's supposed to show you yourself, faithfully, and the horror of it lying — "in a different costume," reusing THE EYE's actual plumbing (CameraGate, the EyeSession-shaped phase escalation) rather than inventing new machinery. It's proposed as a new mechanic in the THE EYE / THE ROOM / THE REACH naming family from Concept & Design.

Core mechanic: the reflection that doesn't match

An escalation ladder deliberately shaped like EyeSession.Phase (dormant → seekingConsent → watching → noticing → awake), so it could reuse the same state-machine pattern:

1. Clear — the reflection tracks the player normally; establishes trust.
2. Lag — a half-second delay, easy to mistake for a frame hitch.
3. Drift — the reflection stops matching exactly (a pose already broken, a hand still raised). Still deniable.
4. Independence — the reflection does something the player didn't do. The first true rule break.
5. Contact (the ceiling) — the reflection writes on the glass. The doc is explicit LULL does not go further than this.

Proposed player interactions: look away / look back (using Sensor.motion — already an allowed sensor in LULLKit.Sensor today, just unused by any mechanic yet — to catch the reflection mid-change); wipe the mirror (a drag gesture that reveals more, never less, as the session escalates — flagged as the cheapest interaction to prototype since it needs no sensor at all); and avoid mirrors entirely, framed as a fictional-level panic switch alongside the literal one — per the doc, "the player needs an equivalent of the panic switch within the fiction, not just the meta-level 'close the eye' button." The doc is explicit that however it's dressed up, THE MIRROR is still Sensor.camera under the hood, and the real ConsentLedger/panic-switch contract applies unchanged.

The "I AM STILL HERE" motif

Proposed as a corruption of the existing Poe lines in Atmosphere.swift's awake phase, rather than a fourth narrative register — the doc argues a fourth voice "would dilute Kafka/Beckett/Poe's discipline." Example given: "it knows your face now." degrading over repeat visits toward "it knows your face now. i am still here." The doc calls for restraint on the phrase itself — mostly heard or implied, fogged and fading before fully legible, reserving one full visual scrawl (if it appears at all) for a single moment near the ceiling of the escalation ladder.

A recurring bathroom, scoped to one sitting

The doc explicitly scopes "recurring" to within one sitting — 3–5 returns to the same bathroom over a session, matching LULL's existing "short, one sitting" scope from docs/concept.md §05, not a multi-session rework. It proposes seeding each return from EyeSession.elapsed / Atmosphere.beat(forElapsed:) — the same clock LULLKit already has — and sketches a 5-row config table of small, deterministic deltas per return (condensation, a scratch mark, light dimming a stage) rather than any real simulation.

Tech approach — cheap first

Two ideas for rendering a "reflection" without building a real mirror, both explicitly framed as reuse of existing plumbing:

• Live-feed distortion through the existing CameraGate protocol — CIFilters (desaturate, contrast crush) plus a frame-delay ring buffer for the "lag" beat (display frame now - k, k growing with elapsed time). No face tracking or ML proposed.
• Canned-asset cut-ins for the "independence"/"contact" beats — crossfade from the live distorted feed to a pre-authored asset at the moment the player looks back, rather than attempting real-time generation of a distorted face.

Lighting, audio, and haptics are all proposed as reuses of existing LULLKit concepts — Atmosphere.beatSeconds' pacing for flicker cadence, Sensor.motion stillness detection for a breathing loop that's audibly out of sync with the player, and THE PULSE's haptic curve rather than a new one.

Restraint principles the doc holds itself to

• Dread over gore, always — no blood-streaked baseline state; the reference image's "soaked" register is explicitly called out as not LULL's.
• Implication over reveal — glimpsed partially, cut short, never a screen-filling clear shot.
• Escalates on LULL's existing slow clock (the same 9-second Atmosphere beat, the same 40s/30s EyeSession pacing), not a jump-scare cadence.
• Never actually deceives the player about what's real — the reflection lies to the character; the game doesn't lie to the player, per CLAUDE.md §1.
• No new sensors and no silent consent — THE MIRROR is Sensor.camera (+ optionally Sensor.motion) with an honest in-app rationale, same revocation contract as THE EYE.

Open questions the document leaves open

Quoted directly, because the doc itself frames these as unresolved:

• Does THE MIRROR ship as part of THE EYE's vertical slice, or as its own slice reusing CameraGate? (The doc's own lean: "own slice — keeps THE EYE's scope, and its existing tests, untouched.")
• Does the one-sitting bathroom loop stay scoped as is, or become the first real use of the future Vapor "haunt" server for cross-session persistence? Only the former is called "small enough to prototype next."
• Should the corrupted-line approach live in Atmosphere.swift directly (a transform over existing Lines) or as a parallel script keyed by loop count? Flagged as worth a spike before committing either way.