Concept: Action Pivot — from tactical stealth to readable action-extraction¶

Pitch¶

Shadow Raiders becomes a fast, readable, one-thumb-plus-one-button action-extraction game — "loot fast, fight loud, get out alive" — reusing every current visual.

The raid stays: drop in, grab loot, reach the portal before it's too late, lose your haul if you fall. What changes is the moment-to-moment: instead of reading vision cones and planning stealth routes, the player steers straight into short, juicy fights, dodges telegraphed hits, crits enemies from behind by moving behind them, and hoovers up a celebration of loot. Sessions shrink to a few minutes; death leads straight into "go again," never into a menu of consequences.

Problem / opportunity¶

The released build proved the look sells but the play doesn't hold — and the July 16–19 publisher test put numbers on it (250 installs; see docs/research/playtest-2026-07-telemetry.md): CPI $1.52 is within benchmark, but D1 retention is 22.5%, only 75.5% finish the tutorial, and 74% of those who reach the real game churn inside the first biome. Players play exactly one raid per session (~9.4 min per session against the 10-minute raid timer), about twice on day 0, and don't come back. The fail-rate curve spikes wherever new content enters — the pack-hunting hound on floor 1-2, the boss wall on 1-5 (fail rate 43.9), the rivals-plus-outranging-Warden stack on 2-1 — a pressure cliff, not a uniform difficulty problem. Notably, the tested build was the tap-to-move stealth build with its full verb set (sprint, dodge, tap-hunt); the stealth-tactical loop failed with its tools intact, which closes off "just restore the verbs" as a fix. The returners tell the other half of the story: the ~20% who do come back play 25+ minutes over 3+ sessions on day 1 — the extraction kernel works for the players it lands with; the design in front of it is what loses everyone else. The current design asks a casual player to read a tactical stealth simulation (vision cones, alert states, noise propagation) while all pressure systems (round timer, storm, procedural sprawl) switch on at once. Fights are long and passive, rewards are deliberately scarce, a raid consumes a whole session, and dying sets the player back.

The publisher's reference — MISFITZ (Antihero Studios) — demonstrates the opposite recipe working in the same genre and on the same platform: top-down, auto-aim, decision-making over aiming, short raids, loot-at-risk as the only tension, and death designed to feed an immediate retry. The publisher's directive after the test: build a new extraction game from the current assets and part of the mechanics — not an iteration of the current game — using the good CPI as the window to test hypotheses quickly. That is this pivot: keep Shadow Raiders' proven visuals, extraction skeleton, and forgiving-death twist, and rebuild the minute-to-minute experience around action.

The concept¶

One thumb steers, one button acts. The left thumb drives the raider directly. Shooting and melee stay fully automatic — the player never aims. The right thumb gets a single big button: a hero action on a short cooldown (at launch, a dodge-dash with a moment of invulnerability; later, alternative actions as loadout/character identity). All skill lives in movement: closing distance, strafing out of a telegraphed hit, dashing through danger, choosing which fight to take.

Stealth becomes positioning, not a system. The signature backstab survives as a simple, always-on rule the player can feel in one encounter: hit an enemy from outside its facing — where it isn't looking — and the strike crits, with the freeze-frame beat already in the game. No detection meters, no suspicion states, no hiding in bushes. Circling behind a wound-up brute is the game's stealth now.

Short raids, one new pressure at a time. A raid lasts a few minutes — two or three raids fit in one sitting. The first real floor is designed to be won: dense easy enemies, generous loot, a visible exit. Each later floor introduces exactly one new pressure — the closing-extraction clock, rival raiders, the locked vault, the boss — continuing the tutorial's one-lesson-per-stage rhythm into the real game instead of dropping every system on the player at once.

Loot is a celebration, not an accounting entry. Kills and chests burst into visible, recognizable pickups that fly to the player. Beyond gear and upgrade currency, raids yield collectible set pieces themed to each biome — a collection album in the lobby gives every raid a "one more piece" pull, the strongest casual meta hook in the reference set.

Other raiders from the first raid. Bot rivals — already the liveliest content in the game (they loot, race for the key, extract with their haul) — appear from raid one as believable, beatable "other players." The extraction fantasy the ads sell is on screen from the first minute.

Death feeds the next run. Nothing banked is ever lost, no progress is reset — dying costs exactly the loot you were carrying, and the "go again" button is two taps away. The target feeling: dying makes you want the next raid more than extracting does.

References¶

Grounding note: the project's gd-knowledge corpus was not reachable in this session; references below come from the dedicated MISFITZ web research (2026-07-22, primary sources verified) and the project's own research docs.

MISFITZ (Antihero Studios — ex-Supercell/King team, Barcelona; pre-alpha Feb 2026, ~83.5K testers; sources: antiherostudios.com/games, the studio's "Alpha Test Results" blog post, mobilegamer.biz coverage of the $4.5M a16z-co-led seed round). Top-down, auto-aim, touch-first; self-described "ARC Raiders meets Brawl Stars"; 5–10 minute raids on a 16-player PvPvE map; hero roster with unique abilities; loot as seasonal collectible "Relics" feeding a collection meta; death forfeits carried loot only. Studio-reported (unaudited) pre-alpha metrics: ~50 minutes average daily playtime over ~3 sessions, and players who died retried at a higher rate (80.5%) than players who extracted (74.5%) — death designed as an appetite, not a punishment. Reported weakness: hero meta collapsed to two dominant characters — a balance caution for any roster we add.
Naavik's mobile-extraction analysis (2022) and the NetEase cautionary tales (Badlanders, Lost Light): casual mobile extraction works when the core ruleset (spawn → loot → extract → lose carried loot on death) is kept and the friction is cut (small maps, short "snackable" rounds, easy believable bots in onboarding). It fails when the wrong things are casualized: excessive movement speed and plentiful auto-collected loot made loot loss meaningless in Badlanders. Our pivot keeps the greed-vs-extract decision sacred.
Archero / Survivor.io — the one-thumb steer + auto-fire control standard the joystick already adopted; Survivor.io's magnet-vacuum loot feel is already in the game.
Brawl Stars — readability bar for top-down action and the character-with-one-ability model the action button points toward.
Arc Raiders (existing primary reference, see docs/research/) — extraction tension, rival-encounter drama, loot-at-risk; still the tone anchor.

Fit with Shadow Raiders¶

The pivot keeps: the raid → extract → bank loop, the no-gear-loss twist (PROJECT_CONTEXT.md §1 — now aligned with the genre's best-practice rather than fighting it), the biome/floor ladder, zone-colored maps, the vault/key race, rival-raider AI, the loot spill + magnet pickup, the hitstop/wobble/lean feedback layer, and the entire cute-fantasy visual set — the asset that earned the good CPI.

The pivot demotes: the stealth simulation that docs/design/stealth-visibility-analysis.md spent Phases A–B making legible (gradual detection, suspect states, echo-sense, bush concealment). Rather than continuing Phases C–D, the pivot concludes that the loop those phases serve is the loop players are leaving. Per the project's convention, these systems are parked behind flags, not deleted — consistent with how sprint/dodge/inventory were parked on this branch. Noise stays, quietly, as the thing that makes looting risky (loud actions attract nearby enemies).

The pivot completes what the virtual-joystick change (docs/superpowers/specs/2026-07-21-virtual-joystick-design.md) started: that spec removed the stealth verbs but left the stealth pressures; this concept removes the pressures and gives the freed thumb a new verb.

It also builds on docs/design/loot-inventory-redesign.md's diagnosis ("loot is a deferred, invisible number dump") — the ground-spill fix shipped; this concept adds the missing contents worth celebrating (collection pieces, more frequent recognizable drops).

Risks & open questions¶

Losing differentiation. Stealth was the stated signature. Post-pivot, the differentiators are the cute-fantasy setting, the flank-crit beat, and the forgiving extraction — is that enough identity vs. MISFITZ itself? (Mitigant: the fantasy skin and PvE-first solo structure are genuinely different — but note MISFITZ entered Global Open Alpha in June 2026, so "it's unreleased" is no longer a moat.)
The Badlanders trap. Faster pace + generous loot risks making loot loss meaningless. The greed decision ("extract now or push one more zone?") must stay sharp — loot density can rise, but banking must still require reaching the portal.
Hero roster scope. Does the action button stay one dash for everyone, or grow into characters/abilities (Brawl Stars model)? Roster = strongest long-term meta, but MISFITZ's own meta-collapse shows the balance cost. Decision needed before meta work starts.
How much tension survives? Removing the storm hard-death and the detection dread could make raids feel stakes-free. The closing-extraction pressure and rival ambushes must carry the tension — needs playtest validation on the first two floors.
FTUE debt. The 7-stage tutorial teaches the old game. Shipping the pivot without rewriting the FTUE would tutorialize systems that no longer exist — the rewrite is part of the pivot, not a follow-up.
Collection meta scope. Album/sets are the recommended casual hook but are net-new UI + economy. Ship the action pivot without it first, or together? (Recommended: action first, collection fast-follow.)