Future plans: designs considered and deferred
A parking lot for mechanics that were designed but deliberately not built, so the thinking is not lost. Nothing here is committed. Each entry says what it is, what it would cost, and, crucially, whether it clears or breaks the game's core principles:
- resolution is server truth:
correct = (choice == "back") == has_anomaly; - confidence never enters correctness;
- the skill is remembering the meaning of a place, not its wording;
- the reset is hard, with no checkpoint;
- the look stays minimal (no HUD scores, no leaderboards, no tracking).
An idea that breaks one of these is not disqualified, but it becomes an explicit owner decision, not a default.
1. Recall: connecting the two verbs (deferred)
Status: deferred. The owner chose to keep the Act 2 / Act 1 wall as-is; the touch stays lore/atmosphere, not play. Kept here in full in case that changes.
Problem it answers: gameplay review #4 ("the second verb never touches the first"). Today touching the room (Act 2) is walled off from the go-on/turn-back call. The collection pays you to touch, so you touch, get a flashback, and make the same untouched decision. Two verbs sharing a screen, not a system.
The mechanic: add one new touch outcome, Recall. You touch a detail and the place tells you how that property has always been (its baseline), in-world: "The light here has always been a steady fluorescent." You have already read this loop's version, so for that one property you can now tell whether it changed.
Why it is play, not content:
- It rewards partial memory: the anomaly is one of ~5 shown details, so a blind recall is a coin-flip; a recall guided by a hunch ("something's off about the light") converts the hunch into knowledge. Fuzzy memory picks the target, the recall confirms it.
- It is a real trade (see cost), so the choice to recall sits inside the loop, next to the actual call. That is the two verbs in tension.
- Experts never need it; strugglers buy certainty at a price. That is a mastery gradient, so it also feeds review #7/#8.
The important nuance: Recall does not change the correctness formula. The server still owns the truth; the player still does the comparison. What changes is the information available before choosing, not the result of a choice. So it honors the letter of the Act 2 axiom ("the action never changes the progress result") while bending its spirit ("a touch now influences which choice you make"). That spirit-bend is the whole decision.
Cost models (it must not be a free oracle):
| Model | Feel | Notes |
|---|---|---|
| Fold risk (recommended) | Recalling risks the place folding you to level 0 | Reuses the existing fold, and redeems review #6 by turning the fold into the chosen price of prying. Self-balancing: recalling all ~5 details is >50% chance of a fold, so no brute-forcing. |
| Hard step cost | Each recall drops a level | Deterministic and legible, but blunt and not thematic. |
| Scarcity | N recalls per run, then none | Simple and gentle, but adds a counter to a minimalist screen. |
In every model, recall also forfeits the "Cold" record (a touch means it was not a clean escape), a soft cost on top of the hard one.
Scope: one arc only, behind a per-arc flag (like axis: vertical). Keep the
other two arcs pure-memory. This preserves the "trust only yourself" experience,
contains the risk, and fits the design where arcs already differ in mechanics.
Fairness: the recall returns a property's baseline value + a localized "as it
always is" sentence. It never returns _has_anomaly or _anomaly_prop, so the
payload still cannot be inspected to cheat.
Build outline (small): a per-arc recall block + a recall outcome in
interact() (return baseline sentence, roll the fold cost); one touch affordance
and reveal line reusing the touch-line plumbing; a Records tweak (recall forfeits
"Cold", maybe a new "trusted no recall" record); localize the recall sentences +
one or two UI strings; verify the payload still hides the answer.
2. Doubt as a wager (deferred)
Status: deferred; breaks a core principle ("confidence never enters correctness").
Problem it answers: gameplay review #2 ("the certainty prompt is not a mechanic"). Today rating yourself guessing just buys one more look and certain just commits; the rating never touches the outcome.
The mechanic: tie certainty to stakes. A confident + correct call advances (and is eligible for a "certain" feat); a confident + wrong call falls harder; guessing still buys the second look but marks the run assisted. Calibration becomes a skill you master and can be measured on, bridging review #7 and #8 for almost no new content.
Why deferred: it reverses the documented invariant that confidence is cosmetic. The partial, non-breaking version already shipped: the Steady record rewards winning without ever admitting a guess, surfacing calibration without changing the resolution. Promoting it to a real wager is the owner's call.
3. Mastery ranks / New Game+ (deferred)
Status: deferred; large scope, does not break core principles.
Problem it answers: gameplay review #8 (no mastery arc) and the standing retention gap (no renewable reason to return).
The mechanic: after a clean escape, a harder variant of that arc unlocks: more of the depth-gated anomaly types active from the start, tighter prose, fewer details shown, shorter look windows. A diegetic frame ("the place remembers you, and gets harder") fits the loop-aware theme. Expert players climb ranks; the reset stays hard at every rank.
Why deferred: it is a whole progression layer, best built after there is more
mechanical depth (see the anomaly-type work in REVIEW_GAMEPLAY.md) for the
ranks to escalate.
Notes
Shipped since (no longer deferred): escalating anomalies in kind landed as hard mode (cross-loop persistence/revert, gated by depth and difficulty), and the coach false-exit trail (the passenger's fixed utterance arming the exit) landed for Coach 8. Both cleared every core principle: the resolution formula and the memory skill are untouched. See CLAUDE.md ("Difficulty & hard mode") and gameplay review #3 / #5.
Tried and removed: the double change (two simultaneous anomalies deep in hard mode) was built and then pulled: turning back on either change is correct and ends the loop, so the "keep scanning" skill it wanted to teach can never form (gameplay review #5). It would only be fair with a post-hoc "two things moved" tell; until that is designed, hard mode stays one-change-per-loop.
Still open here: the legal decoy (a property that legally drifts every loop, never the anomaly) is a natural next hard-mode kind but needs per-arc decoy content, so it is not built yet.