# VOID SENTINEL: LASTLIGHT

Design direction · 8–9 September 2026 · Proposed build target, not implemented behavior

**Wake the city. Wake what hunts it.**

A tactical roguelite about restoring a dead orbital outpost while defending its last reactor. The player chooses which districts to illuminate. Every activation gives the defense something it needs—and opens another approach for the Void.

The signature moment is a district coming alive: lights ripple down its conduits, machinery unfolds, an abandoned transit line hums, and a new enemy silhouette appears at its far end. The player caused both the opportunity and the danger.

This is the new design target requested by Tom. It replaces the old idle/prestige/pass-first direction for future work. Tom subsequently requested substantial WoW-inspired attunement and skill trees; [Attunement and talents](attunement-and-talents.md) defines that required part of the game. Existing code and older design documents remain historical material; this document does not claim that their behavior has changed. See [the engineering review](code-review.md), [the build plan](build-plan.md), and the interactive concept board at `/design-room.html`.

## The promise

Three feelings should survive every scope decision:

1. **I can see my plan working.** Distinct enemies, visible routes, readable range and honest forecasts. A chain reaction is spectacular because its cause is understandable.
2. **I brought this place back.** Progress physically transforms the battlefield. The reward is a living outpost, not just a number in a menu.
3. **That was my tower.** Individual defenses earn names and specializations during a run. The player remembers the unlikely veteran that held a junction.

The game should support deliberate thought and dramatic execution. Pause is free, preparation has no countdown, and 2× is an acceleration of the same rules. We are designing tactical decisions rather than a clicks-per-minute test.

## The first ten minutes

The opening is playable. No account wall, loot screen, or paragraph of lore.

| Moment | What the player does | What the game communicates |
|---|---|---|
| 0:00 | Wakes a small cyan reactor in a mostly dark outpost. | A clipped transmission: “One light left. Make it count.” |
| 0:15 | Places an autocannon on a visibly powered socket. | Its barrel tracks before the first drone appears; range and route are already visible. |
| 0:45 | Stops a short, readable wave. | Metal hits, a controlled energy flash, salvage snapping toward the core. |
| 1:30 | Places a second weapon to answer a scout flank. | Different silhouettes imply different jobs. |
| 2:30 | Chooses the Foundry or the Observatory. | Preview the new route, power, sockets and enemy package before committing. |
| 3:00 | Ignites the chosen district. | A wave of light restores the architecture. The soundscape gains a new musical layer. |
| 4:00 | Chooses a specialization for a contributing tower. | Its shape or firing effect changes. A named veteran begins to emerge. |
| 5:30 | Accepts or declines a clearly forecast Signal challenge. | More salvage has a specific, visible cost in enemies. |
| 6:30 | Restores the second relay and links the defense across the outpost. | A larger battlefield now reflects the player's choices. |
| 8:00–12:00 | Survives the Warden encounter and sends the first transmission. | The formerly dark district answers with hundreds of tiny lights. A restrained victory, then a replay invitation. |

These are pacing targets for the first playable sector, not measured results or rigid real-time deadlines. Thoughtful players can take longer without penalty.

## The decision that makes this game different

The core starts with **6 power capacity** and one active approach. A sector contains two mandatory relay decisions. Each decision offers two authored district alternatives; activating one closes its counterpart for that run. Each relay adds **4 capacity**, powered build sockets, a district benefit, and one new enemy approach. Capacity is a placement limit, not a consumable timer.

All consequences are shown before confirmation. Once confirmed, topology stays fixed until sector completion. The player cannot flicker relays on and off to trap enemies or farm rewards. Routes are authored graphs with verified entrances and exits; towers occupy sockets away from the path and cannot block it.

Example first choice:

| District | Gains | New pressure | Encouraged play |
|---|---|---|---|
| **The Foundry** | +4 capacity, three close-range sockets, +40 salvage on each subsequent cleared wave | A short western approach; an additional armored group appears in its announced waves | Flame, pulse and a concentrated choke point |
| **The Observatory** | +4 capacity, three long-sightline sockets, a mark ability that exposes one priority enemy | A long eastern approach with fast scouts and split timing | Gauss, sniper and priority targeting |

Starting numbers are tunable data. The contrast—income and a short armored route versus targeting and a long fast route—is the committed design. Benefits must remain useful with more than one tower composition.

The second decision offers **Habitat Ring** or **Capacitor Yard**, independently of the first. Both add 4 capacity and three sockets. The Habitat brings windows and civilian signals back to life, repairs 2 core integrity after each subsequent clear (capped at maximum), and opens an outer service route with staggered drones and a Relay Leech. The Capacitor Yard grants one 8-second local overdrive per encounter (+25% fire rate, non-stacking with another overdrive), and opens a wider outer approach where armored escorts protect a Leech. Its new path joins the existing graph at a different junction from the first district. The four combinations are authored and validated individually.

Every sector has seven encounters: two on the initial approach, two after relay one, two after relay two, then its boss. Relay activation is required to continue after encounters two and four. There is no passive preparation income, so waiting cannot solve the sector. Enemies emerge from the dark along forecast corridors; route geometry remains visible before lighting to keep the choice informed.

The opening encounter teaches drones; the second introduces scouts; the third presents the selected district's enemy package; the fourth mixes fronts and offers the first Signal; the fifth introduces the second district's Leech; the sixth tests staggered pressure across all active approaches; the seventh is the Warden. Exact counts and health values belong to the encounter tuning pass.

**Signal challenges** preserve the original early-wave risk/reward idea without requiring players to hurry. Before selected encounters, “Send a signal” adds a specified elite package for a specified reward. Declining remains a complete, viable game. Rewards are committed once at wave clear. No hidden adaptive punishment for playing well.

## A small arsenal with large consequences

Keep the six tower identities. Each has one clear job, a weakness, and two run-specific specializations. Avoid generic menus of +2% bonuses. These names and numbers are design targets, not claims about the generated art or current implementation.

| Tower | Role / power | Specialization A | Specialization B | Natural limitation |
|---|---|---|---|---|
| **Autocannon** | Reliable single-target fire / 2 | **Deadeye:** sustained fire marks one target for allied weapons | **Barrage:** short bursts split across nearby light enemies | Poor armor penetration |
| **Extractor** | Economic investment / 1 | **Recycler:** income on cleared waves increases with that wave's armored kills, with a shown cap | **Capacitor:** gives up bonus income for one local overdrive charge per wave | Uses a valuable powered socket and contributes no direct damage |
| **Gauss Array** | Slow piercing line shots / 3 | **Railbreaker:** pierce armor and fracture it briefly | **Conductor:** the beam arcs once from a marked enemy | Weak against spread-out fast targets |
| **Pulse Halo** | Area control / 2 | **Undertow:** periodic short pull, with elite resistance | **Resonance:** marked enemies transmit a weaker secondary pulse | Modest single-target damage |
| **Incinerator** | Short-range persistent heat / 2 | **Afterburn:** a small ground fire survives the first target | **Flashpoint:** burning fractured armor detonates, consuming the fracture | Limited coverage; burn never stacks without a cap |
| **Longwatch** | Long-range threat removal / 3 | **Executioner:** a large finishing hit against low-health elites | **Spotter:** less personal damage, longer and stronger marks | Slow cadence and poor swarm handling |

Three memorable build examples:

- **The Furnace:** Undertow gathers the wave; Railbreaker opens armor; Flashpoint turns the choke point into a controlled chain reaction.
- **The Long Watch:** Deadeye marks the vanguard; Conductor carries damage into the column; Executioner deletes the dangerous survivor.
- **The Last Stand (Observatory):** Capacitor stores a local overdrive; Barrage clears the swarm; the Observatory's mark lets Resonance turn the boss's escort into an opening. This build depends on that district choice.

All interactions use a short shared vocabulary: **mark, fracture, burn, pull, overdrive**. Each has one canonical rule. Damage recursion has explicit limits: secondary effects cannot recursively trigger themselves, and an enemy receives one event's damage once. Tooltips derive from the same definitions used by combat.

Towers receive contribution-based experience, including effective control and support, instead of rewarding only the killing blow. Miners progress from completed economic contributions. A tier unlock presents compatible choices at preparation time, once per earned tier. The first slice uses one specialization tier; a second signature tier belongs to the full run. Choices persist through reload and cannot be rerolled by reopening a modal.

## Veterans, attachment and replay

The game awards a restrained call sign after a meaningful feat: a Gauss that stops a dangerous armored column might become **Hinge**; a turret that protects a nearly lost core might become **Last Word**. Names come from authored pools and actual event conditions, with no runtime AI service. The player can rename favorites.

The first complete release is a three-sector expedition, targeting roughly 25–35 minutes excluding pauses. At each transit, the player may retain two veteran tower cores. Those cores keep their chosen traits and contribution history; they do not bring placed structures or free firepower onto the next map. Deployment costs the tower's normal build price. All placed structures are automatically salvaged at the displayed 50% rate and unspent salvage carries forward; a fixed, visible transit grant is awarded once. Sector balance must guarantee a viable defense with that grant and the default arsenal, even after a poor prior sector.

The permanent layer is substantial: **30 Commander ranks, an 18-node common talent tree, three 21-node specialization trees, named loadouts and four-step attunement chains**. Players grow stronger and earn new ways to play. Attunements unlock deep powers and advanced content through discoveries and authored challenges. Individual tower choices remain the changing layer inside each run. [The progression specification](attunement-and-talents.md) is authoritative for this system; it replaces the initial cosmetic-only proposal. The archive additionally preserves call signs and discoveries.

Three authored sectors provide different spatial problems:

1. **Cinder Dock:** compact industrial lanes; teaches district ignition and combined fire.
2. **The Glass Meridian:** separated observation platforms and long crossfire; tests coverage and timing.
3. **The Choir Engine:** concentric infrastructure around the final transmitter; tests relay resilience and the complete build.

Replay comes from relay alternatives, authored wave packages, specialization offers and optional Signal contracts. Randomness chooses among validated configurations, not arbitrary layouts. Shared seed codes are a later convenience; they do not imply a secure online leaderboard.

## Enemies that ask questions

| Enemy | Readable behavior | Question it asks |
|---|---|---|
| Drone | Compact ember core, arrives in groups | Can the defense sustain damage across a crowd? |
| Rift Skimmer | Swept silhouette, fast with an obvious direction | Did the player cover the flank and target the urgent threat? |
| Siege Carapace | Broad layered shell, slow armored advance | Is there an armor answer rather than only more basic turrets? |
| Relay Leech | A clearly marked elite drone variant that briefly disrupts a relay | Can the player prioritize a support threat before it interferes? |
| Warden | First-sector boss using the new large crystalline silhouette | Can the player adapt to a telegraphed disruption while maintaining coverage? |
| The Sovereign | Final encounter developed from the boss art direction with its own future animation set | Does the whole chosen defense cooperate under pressure? |

The Leech and Warden are encounter roles; the current art pack has four enemy images, not separate finished assets for every future role. Early elite variants use intentional overlays and distinct markings. No color-only identification.

**Warden encounter:** it advances on a visible route, deploys an escort, then winds up a relay suppression. A clear beam traces the target for at least three simulation seconds. The player can interrupt the exposed node by prioritizing fire; if it lands, the district benefit and its local towers are suppressed briefly. Do not lower global capacity or silently shut down unrelated towers. A temporary loss always recovers; it never deletes a veteran. Telegraph duration scales with simulation speed, with free pause available. Boss victory requires zero boss health, resolved escorts and a living core, not merely finished spawning.

**The signature final beat:** the score drops almost to silence, the Sovereign eclipses the restored city, and successive conduits go dark. The defense's surviving light becomes the visual map of the fight. The successful transmission brings distant stations online, one after another. This is an authored final encounter, not a promise of a cinematic campaign's worth of cutscenes.

## Presentation and control

Use the new Image 2 collection as visual direction and source art. Near-overhead machinery, titanium friendly surfaces, cyan conduits, obsidian enemies and violet/red threat energy. Emissive light belongs to function: a weapon charged, a district powered, a threat winding up. Constant full-screen neon would erase those signals.

The battlefield owns most of the screen. A slim top rail shows core, salvage, power and encounter. The bottom arsenal has six readable silhouettes. Selection opens one compact detail panel; range, route, expected damage and targeting appear close to their objects. The next-wave forecast is available without covering the current fight. Specialization choices use one panel after clear, not nested interruption stacks.

Mouse and keyboard on desktop; tap-to-select and tap-to-place with explicit confirmation on touch. Escape pauses or closes the current layer in a defined order. Space toggles pause unless an input has focus. Number keys choose towers; no hover-only instruction or required right-click. Tablet landscape is a release target. Phone portrait gets a deliberately rearranged HUD and a pan/zoom battlefield only after the first sector is fun and technically sound.

Reduced motion disables camera shake, large zoom beats and flashing; brightness/flash intensity is adjustable. Enemy classes use shape and icons as well as color. UI text must remain legible at 1280×720 and touch targets at least 44 CSS pixels on touch layouts. These are acceptance targets, not current certifications.

Audio is essential and currently absent from the checkout. A reactor drone establishes tempo; percussion and harmony enter as districts awaken; danger subtracts layers before a boss impact. Each weapon needs an identifiable attack/transient, impact and cooldown cue. Spatial audio should clarify location without making headphones mandatory. Respect browser audio activation; no sound until user interaction and no silent fake-loaded files. Sound generation/licensing is a separate future task, not something this image pack supplies.

## The first playable commitment

**Build one remarkable 8–12 minute sector before building three.**

Include Cinder Dock; two relay decisions with two alternatives each; seven encounters; four ordinary/elite enemy roles; one Warden encounter; all six basic tower roles; one specialization tier with two choices per tower; three demonstrable combinations; one Signal contract; readable controls; dependable save/retry; and a satisfying light-restoration ending.

The six tower roles can be introduced gradually within the slice. The engineering proof starts with only Autocannon, Gauss, Drone and Carapace. The finished first slice includes ranks 1–8, working early Commander/specialization choices, and the beginning of attunement. It does not require transit, online services or the final Sovereign encounter.

A playtest should answer: Can players explain a relay tradeoff without prompting? Can they identify why they lost? Do they try a different composition voluntarily? Do they recognize and remember a veteran? Do they want another run after the ending? These questions outrank the size of the roadmap.

## Commercial direction and scope decisions

Retain web-first distribution. Keep the existing no-forced-ads principle. During development there is no pass screen, paid damage multiplier, ad-powered continue, login system or entitlement mock. A full release can evaluate a supporter/cosmetic purchase and platform-permitted optional support ads after the game stands on its own. Platform capabilities and terms must be checked at that point; this plan makes no revenue, approval or traffic forecast. The original $100/month aspiration remains a hypothesis.

Do not begin with multiplayer, procedural terrain, a custom engine, an ECS framework migration, a live-service calendar or cloud saves. Build the talent/attunement framework and a working early path alongside the first sector, then fill the complete trees after their rules are reliable. The spectacle comes from causality, timing, composition and sound.

The strongest simpler alternative is a polished fixed-route tower defense with the existing progression. It is cheaper, but gives the player fewer decisions that belong specifically to Void Sentinel. If the relay choices fail playtests—players cannot foresee consequences, or one branch is consistently trivial—simplify the district mechanic before expanding content. Keep the tower combinations and visual restoration payoff either way.