We are rebooting the classic Reaper Bot from the original source and fixing the community-reported issues first. The goal is a stable, fair, and readable baseline that we can extend in small, testable steps.

All active, working code lives in mre/ only: https://github.com/saworbit/mre/tree/master/mre Everything under archive/ is legacy reference material. Do not edit or build from archive/.

28+ community issues addressed + expanded navigation features - see CHANGELOG.md for details.

• Crashes: SP edict overflow, MP route recursion, scoreboard overflow
• Movement: Jumpy strafing, water flashing, stuck in place, lift/door handling
• Hazards: Lava/slime avoidance, explosive self-damage, LG water discharge
• Fairness: Target selection, vacuum pickup, camping, skill-based aim jitter, observer targeting
• Powerups: Line-of-sight required for pickups (prevents through-wall grabs)
• Physics: Explosion knockback, backpack spawning
• Compatibility: sv_aim warning spam
• Aiming: Predictive lead capping

• High skill levels unlocked (skill 0-10, was 0-3) - skill 4+ = "god mode" bots
• Impulse 100 quick-add bot (standard convention from Frogbot)
• Velocity-based 3D swimming (oxygen-aware surfacing and pitch steering)
• Feeler steering + breadcrumbs (escape scans and dropped BotPath waypoints)
• Navigation learning (link types, usage weighting, danger/decay, rocket jumps)
• Silent Specters (stealth unstuck rollouts; jump noise minimized, enemy-aware penalty)
• Cursed Nodes (adaptive stuck-learning mesh with decay, biases rollouts + routing)
• Phantom Apprenticeship (spectral episodes with rollout validation, soft A* bias, decay; no teleport shortcuts or golden locks)
• Vortex Navmesh (incremental dynamic navmesh seeded from spawns/items, phantom-validated edges, cursed/glory-biased costs; no pre-bake)
• Vortex Telechains (teleporter warp detection and one-way quantum edges; chain-safe, hazard-aware, low-cost path fusion)
• Ripple Oracles (heuristic interactable prediction with beam-search fallback; fuses button/plat/door cascades as ripple edges)
• Grenade Vortex (GJ/GLJ): Grenade jumps and bounce-jumps integrated into Quantum Leaps with ammo-aware priors, purpose-tuned arcs, and phantom auto-tuning
• Mirage Minds (entropy-driven humanization with yaw/pitch jitter, glance-aways, and feint pauses)
• Teacher Mode debug impulses (102 show / 103 hide bot learning nodes, 104 dump spectral episodes)
• Speed Demon update (bunny hopping on straight runs and reflex projectile dodging)
• Predator Update: Strategic intelligence upgrade
  • Sound Navigation: Bots hear combat, item pickups, water splashes, and footsteps (with occlusion) and investigate
  • Curiosity: Bots shoot shootable buttons/doors/walls to discover secrets
• Smooth Steering: 3-frame angle averaging prevents jitter from pathfinder/whisker oscillation
• Sixth Sense: 360-degree awareness for items within 300 units (no facing check needed)
• High-Value Item Focus: Direct drive to powerups when close (<200 units) prevents strafing past
• Proactive Lookahead: 1-tick movement sim nudges bots away from dead-ends
• Mastermind Update: Proactive tactical combat
  • Pre-Fire: Bots shoot rockets at corners where you just disappeared
  • The Trap: Low-health bots stop running and ambush chasers with SSG/RL
  • Displacement: Bots knock enemies into lava/slime using splash damage
• Darwin Update: Adaptive self-learning through reinforcement
  • Natural Selection: Bots mark death locations as dangerous (avoided) and kill locations as glorious (attractive)
  • Weapon Specialization: Bots develop personal weapon preferences based on success/failure history
  • Stuck Learning: Navigation failures mark problematic nodes as difficult
  • Decay System: Danger fades fast (courage) while glory fades slow (nostalgia)
• Specter Gaze: Cinematic spectator camera system
  • Two-layer architecture: heavy math at Think rate, smooth interpolation every server frame (~72fps)
  • Drama-driven auto-switching (no random cuts — cameras follow the action)
  • Enhanced drama scoring: low-health duels and recent combat bonuses
  • Chase mode for first-person through bot's eyes
  • Toggle: impulse 105, cycle focus: impulse 106, chase: specter_chase 1
• Intelligence Update: Skill-differentiated humanization
  • Exponential aim jitter (skill 0 = ~30 deg, skill 5+ = perfect)
  • Reaction fire delay gates first shot after spotting enemy (skill 0 = 300ms, skill 4+ = instant)
  • Non-linear prediction depth (skill 0 = 3 steps, skill 5 = 6, skill 10 = 16)
  • Gradual retreat based on effective health (health + armor), not hard cutoff
  • Ambush uses effective health and skill-scaled patience (2-5s timeout)
  • Adaptive strafe timing (high skill holds good directions longer)
  • Multi-axis humanization: pitch jitter, glance-aways, variable hold-fire
  • Ammo-aware weapon switching (preemptive switch before running dry)
  • Combat bunny hopping (skilled bots hop while retreating mid-range)
  • Blind fire memory (stops pre-firing corners after 2 consecutive misses)
  • Item grab threat abort (abandons pickups when under fire at low health)
  • Vortex slime avoidance and skill-scaled rocket jump depth
• Navigation Humanization: Seven movement systems to eliminate robotic tells
  • Bunny hop rhythm variance (0.28-0.50s timing, 10-22 deg strafe, 30-50 accel)
  • Velocity momentum blending (lerp-based direction changes instead of instant snap)
  • S-curve turns (Hermite smoothstep ease-in-out on yaw clamping)
  • Graduated edge friction (two-tier braking: 64u=0.92, 32u=0.70)
  • Platform fidgeting (micro-drift and look-around while riding lifts)
  • Roaming speed variation (180-240 speed, corner slowdown, micro-pauses)
  • Swimming clumsiness (pitch wobble + sluggish velocity blend in water)
• Adaptive Tactics Update: Cross-system game sense with opponent modeling
  • Opponent profiling: 4-slot LRU tracker with EMA for aggression, weapon, and threat per enemy
  • Counter-weapon selection: reads enemy weapon to pick optimal counter (RL vs LG, LG vs RL, SNG vs SSG, SSG vs GL)
  • Continuous aggression score: 0.0-1.0 spectrum replaces binary fight-or-flee (HP, weapons, powerups, opponent profile, score pressure, multi-threat, match phase)
  • Multi-threat awareness: visible enemy counter during target scan; 2+ enemies feed aggression penalty (no hard combat suppression)
  • Match phase detection: SCRAMBLE/CONTROL/ENDGAME phases shift item priorities (weapons early, armor mid, powerups late)
  • Weapon sound inference: NOISE_WEAPON handler classifies heard weapons and adjusts investigation urgency by own health
  • Adaptive engagement distance: optimal range modulated by enemy weapon (+100u vs LG, -80u vs RL) and aggression score
• Optimization Pass: Structural performance improvements (~25-33% CPU reduction)
  • Missile linked list for dodge scans (replaces findradius over ~600 edicts)
  • Per-frame cvar caching, bot linked list in 6 hot-path functions
  • Arithmetic angle averaging, retreat/scavenge scan throttling
  • Enemy distance dedup, effective HP cache, edge friction early exit

• "Bot frags not shown" - MSG_UPDATEFRAGS sent correctly, likely fixed by 0-index fix
• "Extra SNG ammo" - bots use same ammo as players
• "Firing faster" - bots use identical attack timings
• "Respawn splash sound" - spawn uses correct teleport sounds
• "Floating after respawn" - spawn already uses MOVETYPE_STEP

• Score display at intermission - standard Quake behavior
• Scoreboard colors - engine limitation, modern ports handle correctly
• MultiSkin unreliable - code works, requires player.mdl with 16 skins

.\ci\build_mre.ps1

This compiles mre/, writes progs.dat into ci/mre/progs.dat, and deploys the same copy to launch/quake-spasm/mre/progs.dat so the runtime directory stays in-sync with the build. Manual build (fteqcc) + deploy:

cd c:\reaperai\mre
..\tools\fteqcc_win64\fteqcc64.exe -O3 -Tq1 -DQS_V6 progs.src
copy c:\reaperai\progs.dat c:\reaperai\launch\quake-spasm\mre\progs.dat /Y

Note: progs.dat currently exceeds the 32k global limit, so an enhanced QCVM/engine is required at runtime.

cd launch\quake-spasm
launch_reapbot_v2.bat 8 dm4

Manual command with logging:

quakespasm-sdl12.exe -game mre -condebug +developer 1 -listen 8 +maxplayers 8 +deathmatch 1 +map dm4

If QuakeSpasm reports progs.dat has wrong version number (7 should be 6), run via the bundled FTEQW engine instead:

cd c:\reaperai\launch\fteqw_win64
fteqw64.exe -basedir c:\reaperai\launch\quake-spasm -game mre -condebug +developer 1 -listen 8 +maxplayers 8 +deathmatch 1 +map dm4

# Full stability regression tests (~4 minutes)
.\ci\test_stability.bat

# Quick tests (~2 minutes)
.\ci\test_stability.bat --quick

Command to run stability regression tests: .\ci\test_stability.bat (add --quick for shorter runs).

Automated tests check for: edict overflow, runaway loops, route recursion, scoreboard overflow.

• ARCHITECTURE_CURRENT.md - Control flow and call graph
• REAPER_BOT_REBOOT_SPEC.md - Design and implementation spec (aligned)
• mre/README.md - Quick start and testing checklist
• mre/COMMUNITY_ISSUES.md - Issue tracker
• mre/CHANGELOG.md - Detailed changes
• mre/SOURCES.md - Research sources
• KNOWN_ISSUES.md - Current regressions (bots clipping, logging gaps) and reproduction notes

The reboot now includes predictive rollout systems for navigation and combat:

• Navigation Shadow Puppets: short-horizon beam search to avoid hazards and dead-ends.
• Combat Crucible: dual-rollout combat oracle that selects a first action using beam search.

Runtime cvars (see DEVELOPMENT.md for full details):

• shadow_debug (0/1) master debug
• shadow_nav_* and shadow_combat_* tuning knobs

• sv_mirage (0/1) master enable
• mirage_debug (0/1) developer logs
• Entropy-driven: mood drift, yaw jitter, pitch jitter, glance-aways, hold-fire feints
• No personas or heat maps — lightweight entropy model only

• Incremental mesh: flood-fill from spawns/items builds nodes and edges over time (no pre-bake).
• Phantom episodes validate edges; cursed nodes add cost; glory usage attracts paths.
• Mesh only takes over when a direct line to the goal is blocked; normal BotPath routing remains the baseline.
• Slime and lava both treated as hazards in navmesh edge validation.

• Detects large teleporter warps and fuses one-way "quantum" edges between entry/exit nodes.
• Chaining is permitted with loop limits; hazard exits are cursed, powerups boost usage.
• Tele edges are low-cost in Vortex A*.
• Lift nodes are sampled from func_plat/func_train with wait-cost biasing.

• When goals are blocked, Ripple probes nearby interactables (buttons/doors/plats).
• Heuristic trace + beam-search simulate shoot/touch/wait cascades and fuse one-way ripple edges.
• Vortex pathing can redirect to the interactable position before resuming the goal.
• Rocket jump beam depth scales by skill (6 steps at skill 0, 12 at skill 10).

• Extends Quantum Leaps with grenade jumps (GJ) and bounce-aware GLJ actions.
• Beam-search rollouts sample RJ/GJ/GLJ based on ammo, health, and purpose (lava/vault/high).
• Phantom episodes auto-tune GJ horiz/vert coeffs and GLJ launch angles.