Make Darktide bots as capable as VT2's modded bots (Grimalackt's Bot Improvements - Combat). Start with ability activation (already shipped), then add smart trigger heuristics, safety guards, and general behavior improvements.
- Tier 1 + Tier 2 ability activation for all 6 classes (whitelist removal + meta_data injection)
- Tier 3 item-based abilities (zealot relic, force field, drone — initial implementation)
- Runtime diagnostics (condition/enter/charge trace hooks, debug logging)
- Generic trigger:
enemies_in_proximity() > 0
- Sub-module refactor: thin
heuristics.luadispatcher + splitheuristics_context.lua/heuristics_*.lua, plusmeta_data.lua,item_fallback.lua,debug.lua(#25) - Per-career threat heuristics (#2): 18 functions (14 combat + 4 item) with per-ability activate/hold rules
- 142 unit tests via busted
- Tier 3 reliability fix (#3): all testable item abilities at 100% consume rate
- Structured JSONL event logging (#29) with
bb-log eventsanalysis subcommands - Item heuristics: per-ability rules for relic, force field, drone, stimm field
- Poxburster targeting (#34), ADS fix (#35), bot sprinting (#36)
- Charge/dash rescue aim (#10): rescue triggers + aim correction for zealot dash, ogryn charge, arbites charge
- 198 unit tests, M3 complete
- VFX/SFX bleed fix (#42), smart melee attack selection (#23), ranged fire fix (#31), warp venting (#30)
- Staff charged fire (#43, partial — p4 PASS in-game, p3 untested, p1/p2 need investigation)
- Grenade/blitz throw (#4): state machine for 19 templates, profile-driven. Standard/handleless/mine/knives/whistle/missile. Initial support only.
- Staff charged fire complete (#43): all 4 staves PASS — chain-only fire derivation fix for p1 Voidstrike + p2 Purgatus.
- Bot pinging (#16): bots ping elites/specials for human player.
- Distant special penalty (#19): melee target distance penalty for specials.
- Daemonhost avoidance (#17): suppress combat near dormant daemonhosts.
- 305 unit tests
- Batch 2 shipped: grenade heuristics + Psyker blitz follow-up (#4), hazard-aware abilities (#21), boss engagement refinement (#18), tiered log levels (#40), healing deferral implementation (#39), player-tag smart-target response (#48), and follow-up validation/logging cleanup.
- P0/P1 stabilization: animation crash guard (#50), ammo threshold (#51), melee horde bias (#52), Assail void throws (#61), grenade misaim (#62), unarmed defer, smart blitz targeting, ADS fire fix. Repo cleanup (scrapers migrated to hadrons-blessing). 461 tests.
- Default class-diverse bot profiles (#45): 4-class loadouts with per-slot settings, Tertium compat, cosmetic overrides.
- Full talent enrichment (#63): ~30 talents per class from hadrons-blessing builds, including abilities, keystones, and stat nodes. Bot-optimized build selection (Voice of Command veteran, Electro Shriek psyker, Gun-Lugger ogryn).
- Weapon blessings and perks (#63 phase 2): 2 T4 blessings + 1-2 T4 perks per weapon via synthetic
get_item_instanceoverrides. First mod to construct blessed weapons without player backend profiles. - Settings control surface (#6): category checkboxes, 4 behavior presets, feature gates, veteran semantic stance/shout gate.
- Heuristic dispatch refactor (#60), grenade fallback logging (#59), toggle safety audit (#57).
- Log throttle collision fix: 19 per-bot debug log keys were silently dropping multi-bot messages. Convention updated in AGENTS.md + logging.md.
- 518 unit tests.
- Combat-aware engagement leash (#47): coherency-anchored melee leash with already-engaged, under-attack, ranged-foray, and post-charge-grace overrides. 700+ override events validated.
- Non-veteran profile crash fix (#65):
set_profilehook blocks lossy network-sync overwrite on 1.11.x. - Mastiff-pounced priority (#55): score boost for pounced enemies + fix for zero-score gate when bot has no slot.
- Poxburster push (#54): bypass
_should_pushoutnumbered gate for poxbursters. - Rumbler VFX timing gap (#53): pre-call hook on loadout init.
- Healing deferral validated (#39): health station deferral confirmed in-game.
- Parser-level ability input validation:
_action_input_is_bot_queueablein shared_rules.lua fixes 1.11.x action input rejection for combat abilities. - Event log JSON fix:
_json_safe_numbersanitizes infinity/NaN before cjson.encode. - Debug logging: set_profile guard, record_charge, per-bot poxburster keys.
- 579 unit tests via busted.
Issues are tracked on GitHub.
Theme: bots perceive and react to combat situations better.
| # | Issue | Notes |
|---|---|---|
| 65 | P0: non-veteran profiles CTD on 1.11.0 | Done. set_profile hook blocks lossy network-sync overwrite. Validated on 1.11.3. |
| 54 | Push poxbursters | Done. Bypass _should_push outnumbered gate for poxburster breed + push logging. |
| 55 | Prioritize mastiff-pounced enemies | Done. Score boost for immobilized targets. Fixed score=0 gate bug (pounced enemies have no bot slot). Validated with 80+ pounce events. |
| 53 | Rumbler VFX timing gap | Done. Pre-call hook on loadout init restored (crash was from profiles, not VFX). |
| 47 | Combat-aware engagement leash | Done. Coherency-anchored leash: stickiness-limit extension, post-charge grace (validated), under-attack/ranged-foray overrides. 700+ override events in-game. |
| 39 | Healing deferral | Done. Validated: 80+ health station deferral events. Correct no-defer when human at full health. |
| — | Parser-level input validation | Done. _action_input_is_bot_queueable in shared_rules.lua: check parser sequence_configs before action handler validation. Fixes 1.11.x action input rejection. |
| — | Event log JSON fix | Done. _json_safe_number sanitizes math.huge/NaN before cjson.encode. |
Theme: fix regressions and behavior issues reported on Nexus after v0.9.0.
| # | Issue | Notes |
|---|---|---|
| 67 | P0: hook_require clobbering | Done. April 7 logs show the consolidated bt_bot_melee_action hook install and repeated melee choice ... lines, confirming the clobbering regression is fixed. |
| 68 | P1: Veteran class swap with other mods | Done. Yield guard now checks profile.character_id and profile.name (real backend characters always carry name; vanilla bot stubs use name_list_id instead), committed in 8520485. Validated from run console-2026-04-07-15.36.11: BetterBots preserved external profiles for bot slots 1–4 with real character_id values and logged allowed profile update (no _bb_resolved sentinel). Regression coverage in tests/bot_profiles_spec.lua. Closed. |
| 73 | P1: exception-safe shared state mutation | Done. Wrapped the temporary shared-state mutations in restore-on-error guards. Kept as defensive hardening; reopen only on actual repro. |
| 69 | P2: Mastiff-pinned target fixation | Done. Validated in run 0: friendly companion-pin penalties fired in both melee and ranged scoring. |
| 70 | P2: Arbites whistle ignores dog position | Done. Validated in run 0: invalid whistles were held and valid whistles still consumed charges. |
| 71 | P2: Ogryn grenade mid-horde | Done. April 7 event log shows grenade_ogryn_frag_block_melee_range below 4m and grenade_ogryn_frag_horde only above 4m. |
| 72 | P3: Configurable ammo policy | Done. April 7 logs exercised both the lowered ranged fire gate and ammo pickup decisions in runtime. |
Theme: bots coordinate with each other, protect interacting allies, and direct companions deliberately.
| # | Issue | Notes |
|---|---|---|
| 14 | Ability cooldown staggering | Done + validated in console-2026-04-11-17.41.41: textbook timeline at 17:50:09–17:50:18 — vet shout records aoe_shout at t0, two psyker shouts suppressed at Δ=2 s and Δ=4 s (fallback suppressed psyker_shout (team_cd:aoe_shout)), third psyker attempt at Δ=8 s fires through normally after the 6 s window expires. Two integration bugs found and fixed in d890695: variant talent name vs semantic_key in TeamCooldown.record(), and missing is_suppressed check in ability_queue.lua fallback path (the dominant activation path in solo play). Closed. |
| 37 | Objective-aware ability activation (P1) | Done + validated in console-2026-04-11-16.51: veteran_voc_protect_interactor queued 4× and ogryn_charge_block_protecting_interactor held 7× during the same run. Both the activate-defensively and suppress-aggressive sides confirmed. Closed. |
| 7 | Revive-with-ability (P1) | Done + validated in console-2026-04-11-16.51: 5 revive_candidate observed lines (psyker × 4, vet × 1; both knocked_down and netted interaction types) plus revive ability skipped (psyker_shout can_use_ability=false) proving the full hook → identity → ability check chain executes. Closed. |
| 49 | Arbites companion-command smart tag | Done + validated. 42+ companion-tagged events across 2 sessions plus bot 5 companion-tagged renegade_netgunner (reason: opportunity_target_enemy) in the latest run, with already_tagged dedup firing as expected. Closed. |
| 81 | Expand settings surface | Done + validated in console-2026-04-11-16.51: startup log line settings: preset=balanced, sprint_dist=12, chase_range=18, tag_bonus=3, horde_bias=4, smart_targeting=true, dh_avoidance=true resolves all 7 spec values. Always-emit-all-values mode landed in ed9f12b. Closed. |
| 83 | Settings UI: reorganize groups, factory functions, visual polish | Done. All 8 checklist items complete across 962a384, c7c9954, 5695e1f, ed9f12b. Group split, slot dropdown factory + deep-copy, Testing preset moved last, slider tooltip pattern, "Max" string fix, slot grouping with descriptions, gold/citrine group headers, event_log/perf_timing show_widgets gating. Plus golden mod_name with U+E048 mastery glyph. Closed. |
Theme: small, high-ROI polish. Trivial wins and one teammate-feel feature. v0.11.0 was released on 2026-04-15; the remaining live-validation issues are post-release follow-up, not release blockers.
| # | Issue | Notes |
|---|---|---|
| 32 | Mule item pickup | Closed 2026-04-15. mule_pickup.lua activates the dead vanilla book-carry path by mirroring inventory_slot_name -> slot_name and setting bots_mule_pickup on the side-mission book pickup templates. Tomes and grimoires each have their own BetterBots toggle (tome defaults on, grimoire defaults off); disabling a toggle also blocks pickup orders and prunes cached references for that type. Stale-unit cleanup runs unconditionally regardless of toggle state (fix on 2026-04-14 after review found the gate was silently skipping stale cleanup when grimoires were enabled). On 2026-04-15 BetterBots also added a post-vanilla mule-assignment override so nearby humans no longer suppress default book pickup. The same day, console-2026-04-15-18.01.55... added the authoritative runtime proof: mule pickup success: tome (bot=4) followed by stale-ref cleanup. Explicit tag -> pickup-order routing remains separate in #96. |
| 44 | Human-likeness tuning (Tier A) | Closed 2026-04-13. The old checkbox is replaced by split timing/leash profiles: human_likeness.lua now resolves off/fast/medium/slow/custom timing into BotSettings reaction times plus defensive/opportunistic jitter buckets, while engagement_leash.lua reads off/light/medium/strong/custom pressure-leash scaling instead of the old fixed half-leash model. console-2026-04-13-14.10.19 confirmed the runtime reaction-time patch and pressure leash scaling; jitter remains covered by tests but is not separately logged. Grenade/blitz timing, aim tuning, and dodge tuning remain out of scope. |
| 82 | Perf low-hanging fruit audit | Closed 2026-04-16. The low-risk audit is complete. Final branch cuts included same-frame Heuristics.resolve_decision(...) reuse between condition_patch.lua and ability_queue.lua, plus BT-side one-shot AbilityTemplates fetch + _MetaData.inject() so the BT path stops paying that metadata guard on every can_activate_ability evaluation. Closure is based on direct live evidence from console-2026-04-16-15.35.10...: resolve_decision cache hit ... lines prove the new reuse path is active in-game, and the same log contains a mission-end bb-perf:auto: 104.9 us/bot/frame total block. That validates the implementation, but not the original <80 us/bot/frame target; any deeper perf work moved to follow-up #99. |
| 87 | Sustained fire for flamers and held-fire weapons | Closed 2026-04-13. sustained_fire.lua now observes queued weapon_action inputs, arms per-unit sustained state for supported full-auto / stream paths, and injects required raw hold inputs through BotUnitInput._update_actions until the path goes stale or clears. Covered: flamer, Purgatus, recon lasguns, autoguns, autopistol, dual autopistols, bolter hipfire, heavy stubbers, and rippergun braced fire. console-2026-04-13-15.20.37... captured both the stream-route markers and the hold confirmation lines (holding sustained fire inputs) for Purgatus and flamer. Scope remains execution-only; ADS/brace choice still belongs to #41. |
| 89 | Bot grenade pickup heuristic | Closed 2026-04-15. ammo_policy.lua piggybacks the vanilla pickup_component.ammo_pickup slot for nearby small_grenade refills whenever every eligible human is above the configured grenade reserve threshold, while cooldown-only blitz users are excluded from grenade-pickup arbitration. console-2026-04-15-14.44.35... now contains the missing standalone world-pickup proof: grenade pickup permitted, grenade pickup bound into ammo slot, then grenade pickup success: small_grenade (bot=3, charges=0->3/3) and again ... (bot=3, charges=1->3/3). |
| 90 | Melee/ranged target type hysteresis | Closed 2026-04-16. target_type_hysteresis.lua now runs on the actual live path by post-processing BotPerceptionExtension._update_target_enemy, not by patching the dead bot_target_selection_template.lua table. BetterBots.lua owns the consolidated bot_perception_extension hook so hysteresis and poxburster post-processing do not clobber each other. Closure is based on console-2026-04-16-17.51.56...: 32 type flip ... lines, 46 type hold ... lines, 4 suppressed opposite-type switch ... lines, and the mission-end perf row target_type_hysteresis.post_process 143.000 ms total (11473 calls, 12.5 us/call). That resolves the earlier proof gap where the BT-side debounce fired but the math-layer hook never showed up in logs. |
| 91 | Bot weakspot aim MVP | Closed 2026-04-13. ranged_meta_data.lua injects vanilla-style aim_at_node = { 'j_head', 'j_spine' } for lasguns, autoguns, bolters, and stub revolvers when the template leaves attack_meta_data.aim_at_node unset. Existing aim_at_node values still win, and vanilla target_breed.override_bot_target_node behavior is unchanged. Live runtime confirmation now exists via `weakspot aim selected j_head |
| 93 | Grenade ballistic arc fix | Closed 2026-04-13. grenade_fallback.lua now resolves projectile locomotion data from the equipped grenade weapon template and mirrors vanilla Trajectory.angle_to_hit_moving_target(...) solving for supported gravity-affected manual-physics projectile families. console-2026-04-13-14.10.19 confirmed ballistic aim, queued wield, and charge consumption during live throws; closure used that runtime evidence plus manual gameplay confirmation that the remaining short-throw concern was gone. Covered: standard grenades, handleless grenades, Ogryn grenade throws, and zealot throwing knives. Flat fallback remains for broker_missile, psyker knives, whistle, smite, chain lightning, and mines. |
Theme: bots feel like teammates, not automatons. VT2 Bot Improvements parity. Mechanical polish + talent awareness + consumable-item coordination.
Release framing. v1.0.0 is scoped as the terminal release. Post-1.0 work may never ship — the milestone therefore includes everything feasible without architectural rewrites, not just the cheapest wins. Broad-scope follow-ups under individual issues are formally documented as scope-exit (see Post-1.0 section).
Execution order. Sprints are a logical dependency ordering, not calendar weeks. Foundation primitives (F1 talent context, F2 pocketable carry bootstrap) are load-bearing — F2 unlocks the carried-consumable MVP on its own.
| # | Item | Notes |
|---|---|---|
| F1 | Talent/buff context extension | Additive fields in build_context() — has_talent(), current_stacks(). Prerequisite for #38. Pure additive. |
| 13 | Navmesh charge/dash validation | Implemented 2026-04-18. Shared charge_nav_validation.lua validates the actual launch endpoint with NavQueries.ray_can_go(...) before committing ogryn_charge, adamant_charge, and zealot dash variants. Zealot dash resolves its targeted enemy from bot perception/smart-targeting, rescue charges validate the explicit ally aim point, and directional charges fall back to navigation_extension:destination() when no better endpoint exists. Wired into both BtBotActivateAbilityAction.enter and ability_queue.lua, with rescue aim applied only after validation passes so a blocked charge cannot leave BotUnitInput stuck on the ally position. Same-endpoint failures cache for 0.5s so GwNav queries do not repeat every tick, and the behavior is exposed behind enable_charge_nav_validation as a user kill switch. |
| 92 | Per-breed weakspot aim override | Hook BtBotShootAction._set_new_aim_target atop the #91 MVP allowlist in ranged_meta_data.lua:224, with live _aim_position re-evaluation for the two stateful cases. Implemented in code for Scab Mauler (j_spine), Bulwark exposed-head routing (j_head only when the shield is open or the bot is outside the 70° blocking cone, using the engine's flat block-angle math), and a provisional Crusher rear-arc proxy (j_head only when the bot is behind the target). The Crusher path is documented assumption, not rig-verified fact: the original back-of-head claim still lacks a confirmed node name in decompiled source. Independent of #41. |
| 86 | Tier 3 revive cover — timing investigation | Done 2026-04-17. Scope-exit at the enter hook confirmed by two independent investigations. PlayerCharacterStateInteracting.on_enter interrupts any in-flight ability/weapon action and force-wields slot_unarmed, killing Tier 3 sequences on revive entry. Viable-but-substantial architecture (approach-phase hook via on_refresh_destination + gated can_revive) documented on the issue and moved to Post-1.0. |
Not a Zealot-Martyrdom one-liner. BB ships three tuned builds in bot_profiles.lua; each needs heuristic touchpoints.
| Build | Talent marker | Touchpoint |
|---|---|---|
| Zealot Martyrdom | zealot_martyrdom |
Live healing suppression on stations/med-crates + zealot_invisibility low-HP panic disable; pocketable wound-cure path stays deferred with F2 |
| Psyker Warp Siphon / glass cannon | psyker_damage_based_on_warp_charge + psyker_warp_glass_cannon |
Raise peril vent threshold to preserve warp-charge-scaled damage |
| Psyker Venting Shriek cadence | psyker_shout_vent_warp_charge |
Shout as vent-trigger; cooldown shape differs from burst-damage shout |
| Veteran VoC + Focus Target | existing stance path | Verify tag ownership, then narrow ping override so Focus Target can still claim already-tagged priority targets |
Psyker Scrier's Gaze follow-up (#104) |
psyker_new_mark_passive, psyker_overcharge_weakspot_kill_bonuses, psyker_overcharge_reduced_warp_charge, psyker_overcharge_stance_infinite_casting |
Lower stance threat/density gates for aggressive mark/weakspot builds; widen upper peril ceilings for reduced-peril / Warp Unbound variants |
Ogryn Point-Blank Barrage follow-up (#104) |
ogryn_special_ammo_fire_shots, ogryn_special_ammo_armor_pen, ogryn_special_ammo_movement, ogryn_ranged_stance_toughness_regen |
Build-specific activation reasons for horde pressure, hard ranged targets, closer-range commitment, and low-toughness ranged sustain |
Further keystone/build extensions beyond the current shipped batch (Broker Chemical Dependency / Adrenaline Junkie, Ogryn Carapace Armor, wider Veteran specialization coverage, deeper Scrier vent logic) remain post-1.0.
| # | Issue | Notes |
|---|---|---|
| 38 | Talent-aware behavior | Code-complete 2026-04-18; first broader follow-up #104 landed 2026-04-19 and closed 2026-04-22 after live validation. The base MVP still covers Martyrdom live-healing suppression + low-HP Shroudfield panic disable, Psyker shout peril preservation keyed off warp-charge damage talents plus psyker_shout_vent_warp_charge, and Veteran Focus Target's one-shot tag reclaim for enemy_over_here_veteran. The first broader build-aware follow-up now also tunes Scrier's Gaze for aggressive mark/weakspot builds and reduced-peril / Warp Unbound variants, and tunes Point-Blank Barrage for Fire Shots, Armor Pen, movement, and toughness-regen branches. Detection still degrades cleanly when talents are absent (non-BB profiles). |
| # | Issue | Notes |
|---|---|---|
| 41 (narrow) | Weapon-family close-range classifier | Code-complete 2026-04-18; first explicit family-expansion follow-up #105 landed 2026-04-19 and closed 2026-04-22 after live validation. Narrow family policy landed in ranged_meta_data.lua, then wired into both target_type_hysteresis.lua and the vanilla BtBotShootAction._should_aim hook. Supported close-range families are now flamer, Purgatus (forcestaff_p2_m1), shotgun, heavy stubber, autopistol/dual autopistols, and rippergun. Under close pressure those families keep ranged target type instead of falling back to melee; ADS suppression applies to the true hipfire families (including autopistols), while Purgatus and ripperguns preserve their non-hipfire fire paths. Broad enemy-aware fire cadence and any autogun subgroup expansion stay post-1.0. |
| 33 (narrow) | Activate_special melee | Code-complete 2026-04-18; follow-ups landed 2026-04-19, 2026-04-20, and 2026-04-22; chain-family follow-up #103 closed 2026-04-22 after live validation. melee_attack_choice.lua now caches supported weapon-special metadata during BtBotMeleeAction.enter and prepends special_action before the chosen attack when the current family-specific rule says the target is worth it. The first post-1.0 pass added chain-family toggle_special weapons (chainaxe_, chainsword_, chainsword_2h_) for elite/captain/monster/boss or super-armor targets. The next pass split the old powered bucket into real families: 1H power swords now arm in live combat windows instead of only elite/special cases, Zealot 2H power swords resolve their toggle_special path correctly, force swords stay targeted, thunder hammers now widen to armored/heavy elites plus captain/monster/boss, and Ogryn latrine shovels (ogryn_club_p1_m2/m3) now fold for high-health or armored targets, forcing heavy follow-ups on super-armor and armored high-health targets. First-pass ranged specials also landed for supported shotgun special-shell loaders; mauls, other Ogryn melee specials, and bash/bayonet ranged specials remain post-1.0. |
Sprint 4 stopped being "invent a missing primitive" once the decompiled source was checked closely on 2026-04-18. Vanilla already has the carry side (pickup_component.mule_pickup, BotBehaviorExtension._refresh_destination, PocketableInteraction.stop); BetterBots needed to patch supported pocketables into that path, then layer conservative bot policy and use/deploy sequencing on top.
| # | Item | Notes |
|---|---|---|
| F2 | Pocketable carry bootstrap | Code-complete 2026-04-18; in-game validation pending. New module pocketable_pickup.lua patches supported pocketables (ammo_cache_pocketable, medical_crate_pocketable, the three combat stims, and syringe_corruption_pocketable) into the existing vanilla mule path, blocks proactive grabs while a human still has the matching slot open, and hooks update_dispatcher.lua for conservative wield/use/place follow-through. |
| 24 (a) | Medicae discipline | Code-complete 2026-04-18; in-game validation pending. healing_deferral.lua owns the human-first medicae policy, Martyrdom carve-out, and station/deployable deferral hooks. The original Sprint 4 MVP shipped conservative bot-side onset rules here; #97 later narrowed that to "any missing amount once humans are above reserve" for non-Martyrdom bots without reviving the broad wound-cure / ally-handoff scope. A distinct corruption-only onset and a station-charge-aware reserve model are still not shipped. |
| 24 (b) | Stim usage | Code-complete 2026-04-18; in-game validation pending. Supported combat stims self-use on high-threat combat entry after the carried stim is wielded into slot_pocketable_small; syringe_corruption_pocketable is also pickable now and self-uses conservatively when the bot itself needs healing/corruption relief while ally-give stays deferred. |
| 24 (c) | Pocketable wound-cure / ally distribution | Deferred post-1.0 on 2026-04-18. The carry primitive exists, but reliable give-to-ally / wound-cure behavior is not load-bearing for v1.0.0 and the vanilla handoff path is too narrow/flaky for an MVP claim. |
| 88 | Deployable crate carry + deploy | Code-complete 2026-04-18; in-game validation pending. Ammo + medical crates now ride the vanilla mule carry path and auto-deploy from slot_pocketable only when the bot is safe, at least two allies are in coherency, and the team actually needs the resource. |
| # | Issue | Notes |
|---|---|---|
| 56 | Communication wheel response | Code-complete 2026-04-18; in-game validation pending. Narrow Solo Play MVP in com_wheel_response.lua: hook Vo.on_demand_vo_event so vanilla + ForTheEmperor wheel events share one detection path, temporarily force aggressive after com_cheer, and treat com_need_ammo / com_need_health as short-lived human-priority overrides in ammo_policy.lua / healing_deferral.lua. Need Help convergence and location-directed movement stay out of scope because BetterBots still has no movement-order layer. |
| 96 | Smart-tag item interaction bridge | Code-complete 2026-04-18; in-game validation pending. smart_tag_orders.lua hooks SmartTagSystem.set_contextual_unit_tag for first-time item pings and SmartTagSystem.trigger_tag_interaction for already-tagged items, then routes explicit item-tag interactions for ammo, world grenade refills, books, and supported pocketables back into BotOrder.pickup(...), reusing existing MulePickup/PocketablePickup policy gates and choosing the nearest eligible bot. Health stations and location-style interactions stay out of scope for the MVP. |
| 97 | Non-book resource arbitration | Code-complete 2026-04-18; in-game validation pending. Ammo, ammo-driven grenade refills, and world grenade pickups already used the intended "human reserve first, then any missing amount" onset on the current branch; the remaining gap was medicae. healing_deferral.lua now promotes health-station demand for any damaged non-Martyrdom bot once humans are above reserve, instead of inheriting vanilla's heavy-damage threshold or BetterBots' old corruption/>80%/last-charge vetoes. Med-crates keep the same human-first gate and vanilla's existing any-healable-damage onset. |
| 101 | Weapon-slot wield timeout on BB-locked swap | Code-complete 2026-04-18; in-game validation pending. Grenade + item fallbacks now query shared BetterBots slot-lock state and fast-retry with reason = "slot_locked" instead of waiting out the full wield timeout when another BB sequence is holding a different slot. Fits coordination-polish neighborhood; not load-bearing. |
| # | Issue | Notes |
|---|---|---|
| 98 | Sparse metadata hardening | Code-complete 2026-04-18; in-game validation pending. Targeted boundary guards landed in melee_attack_choice.lua, melee_meta_data.lua, and ranged_meta_data.lua: truthy non-table attack_meta_data is now treated as invalid at the boundary instead of being indexed as if it were sane, while the metadata injectors replace malformed values only for the enabled lifetime of the feature and restore the original value on disable. Regression coverage now includes non-table top-level melee selection plus malformed existing attack_meta_data on both melee and ranged injectors. No blanket pcall, no broad refactor. |
| 102 | ActionInputParser zoom/unzoom noise | Closed 2026-04-18. ranged_meta_data.resolve_vanilla_fallback() no longer invents "zoom"/"zoom_shoot" on templates that do not actually expose aim inputs; it keeps explicit vanilla action start_inputs, derives brace/charge aim paths when they exist, and otherwise returns nil aim fields. weapon_action.lua now clears stale scratchpad zoom/unzoom fields when the current template has no aim chain and drops parser-proven unsupported queued zoom / unzoom inputs on the currently wielded template before they reach the engine parser. |
| 99 | Perf benchmark protocol | Closed 2026-04-18. BetterBots already had the reusable recorder; the missing piece was a credible gate. docs/dev/debugging.md now defines the v1.0.0 benchmark protocol: use mission-end bb-perf:auto: totals from three combat-heavy Solo Play runs on the same build, with only Performance timings enabled, and evaluate the median rather than a single cherry-picked dump. Current validated reference band is 104.9, 113.8, and 124.5 µs/bot/frame total, so the old <80 µs/bot/frame target is retired; v1.0.0 accepts median <= 125 with no single run > 140. Reopen only on user-visible perf complaints or future runs outside that band, then investigate ability_queue.decision, grenade_fallback, sprint.update_movement, and ammo_policy.update_ammo. |
| 85 | Combat ability identity refactor | Closed 2026-04-18. The shared combat_ability_identity.lua resolver already landed, duplicated Veteran-specific sniffers are gone from leaf modules, and regression guards now enforce both table drift and "no duplicate local resolver" invariants in tests/startup_regressions_spec.lua. Later live validation already exercised the important shared-template paths: #7 proved the revive hook → identity → ability check chain on real revive candidates, and #14 proved semantic cooldown routing for Veteran shout. No further v1.0.0 code work is justified here. |
| — | Full cold-boot soak | Both mod load orders, grep DMF warnings in raw console, Auric mission runs. |
| — | Nexus package + changelog + outreach | Release mechanics. |
Runway cuts are now mostly moot. The planned code/doc items for v1.0.0 are either landed or explicitly scope-exited; the remaining release work is validation and packaging, not feature triage.
Do not cut: Sprint 4's shipped MVP (F2 bootstrap + #24a/b + #88 carry/deploy), #33-narrow melee identity, #38 keystone layer, #13 + #92 mechanical polish. These are load-bearing for "v1.0.0 was worth shipping."
Theme: architectural upgrades and research-track items. If post-1.0 happens, each is a mini-project.
| # | Issue | Notes |
|---|---|---|
| 22 | Utility-based ability scoring | Replace boolean heuristics with spline-interpolated utility curves. Darktide has native utility.lua + bot_utility_considerations.lua — framework exists, needs wiring. Architectural upgrade. |
| 28 | Built-in bot profile management | Absorb Tertium4Or5 functionality. Profile selection + loadout preset support. Only pursue if upstream remains unpatched. |
| 80 | Grenade/blitz tactical evaluator | Shared grenade/blitz decision object, family-specific targeting/placement, Arbites dog vs Lone Wolf split, and execution-time revalidation tied to original tactical intent. Planning docs: docs/superpowers/specs/2026-04-08-grenade-blitz-tactical-evaluator-design.md, docs/superpowers/plans/2026-04-08-grenade-blitz-tactical-evaluator.md. References #49 and is intentionally narrower than #22. |
| 84 | User-authored bot profiles | Integration with hadrons-blessing for user-defined bot builds. Design-heavy, no concrete scope yet. |
| 86 | Tier 3 revive cover | Moved 2026-04-17 after Sprint 1 timing investigation confirmed BtBotInteractAction.enter is unusable (engine interrupt on interacting state entry). Viable architecture: approach-phase hook in revive_ability.on_refresh_destination + gate bt_bot_conditions.can_revive while item sequence mid-flight. Ships instant-variant shield + drone; relic stays excluded (5.6s unwield + slot lock). Full plan on the issue. |
Use this as the execution map, not as a replacement for issue acceptance criteria. ROI means player-visible impact plus risk reduction divided by implementation and validation cost. needs-testing on GitHub means code/support is present but live proof is missing.
| Rank | Issue | Class | ROI call / next action |
|---|---|---|---|
| R1 | #108 Human revive priority |
Needs testing | Code-complete 2026-04-28: revive_ability.lua now assigns the nearest live bot to a knocked-down human at the pre-BT destination-verification seam, forces revive_with_urgent_target, and refreshes movement so vanilla ally-aid pathing outranks urgent/priority enemy pathing. Needs cold-boot Solo Play validation with debug logs. |
| R2 | #17 Daemonhost avoidance |
Validate/fix | Rage-quit class failure. Add decisive first-action stage/aggro_state evidence, then either close or patch the remaining dormant-state classifier gap. |
| R3 | #106 Perf cap miss |
Implement | Every future feature inherits this cost. Start with grenade fallback idle cadence, then context augmentation caching if needed. |
| R4 | #100 Scenario validation harness |
Implement | High developer ROI. A narrow /bb_scenario MVP would turn daemonhost, revive, poxburster, grenade, and weakspot claims into repeatable checks. |
| R5 | #96 Smart-tag item bridge |
Validate | User-visible command path; current blocker is positive live evidence, not architecture. Confirm smart-tag pickup routed ... after the liveness fix. |
| R6 | #43 Voidblast p1 charged fire |
Validate/fix | Narrow Psyker build payoff. Close only after voidblast anchor locked plus voidblast charged fire override or real trigger_explosion evidence. |
| R7 | #92 Per-breed weakspot aim |
Validate/limit | Validate Mauler spine first. Keep Crusher/Bulwark expansion deferred until rig/node evidence is real. |
| R8 | #33 Remaining weapon specials |
Implement slices | Continue only high-value families: Ogryn specials, mauls, rippergun/bayonet/bash where they change survival or damage. Leave flashlight/parry theater alone until evidence says otherwise. |
| R9 | #41 Weapon/enemy-aware ranged behavior |
Implement slices | High potential but broad. Keep family-scoped after perf/harness; avoid a general fire-control rewrite. |
| R10 | #107 Barrel avoidance |
Validate first | Likely vanilla already handles triggered-barrel AoE threats. Instrument and patch only a proven gap. |
Lower ROI for now: #80 and #22 are architectural and should wait for #100; #86 is cool but less important than #108; broad #24, #28, and #84 carry support and UX burden; #8 is DLC-gated.
Broad-scope cuts (scope-exit, captured under parent issues):
- Broad
#24: complex healing-item ping negotiation beyond F2 primitive + medicae + basic distribution - Broad
#33: ranged weapon specials (bayonet, pistol-whip, racking slide) and richer per-family target policy beyond the current powered-vs-chain split - Broad
#41: full enemy-aware fire cadence, dynamic gestalt per target type, plus any autogun subgroup audit beyond the current family set - Broad
#92: replace the provisional Crusher rear-arc proxy with a rig-verified node or stronger live evidence - Keystone/build extensions beyond the current shipped batch: deeper Scrier's Gaze vent logic, Broker Chemical Dependency / Adrenaline Junkie, Ogryn Carapace Armor stack mgmt, broader Veteran specialization coverage, and any future Brain Burst proc-cooldown-aware follow-up beyond the shipped
psyker_smite_on_hitde-prioritization carve-out
| # | Issue | Blocker |
|---|---|---|
| 8 | Hive Scum ability support | DLC-blocked (Hive Scum / broker archetype not owned) |
| 17 | Daemonhost avoidance | Code + tests shipped v0.6.0, but still validation-gated. First real DH spawn on 2026-04-11 exposed a grenade/blitz-path gap; heuristic carve-out staged in 03ce4fd+ffe7c6b. A later 2026-04-15 log still showed bad pre-aggro behavior because BetterBots was only consulting aggro_state; the current branch now uses daemonhost stage when available and treats any non-aggroed stage as dormant across combat, ping, companion-tag, and player-tag-boost paths. Later 2026-04-23 and 2026-04-24 logs contain real daemonhost spawns, but no positive dormant-suppression evidence; the 2026-04-24 engagement is inconclusive until first-action stage/aggro_state is logged. |
- Don't break what works. Vanilla bot combat (melee, shoot, revive, rescue, follow) must remain functional. Every change is additive.
- Per-ability, not per-class. Trigger heuristics are per ability template, not per archetype. A Zealot with Dash needs different rules than a Zealot with Stealth.
- Precise triggers, eager usage. Bots should use abilities frequently but only when the trigger conditions genuinely apply. Community consensus (VT2 and Darktide) strongly favors maximizing ability uptime — conservation frustrates players more than occasional misfires. Heuristics should be confident, not rare.
- Observable. Debug logging traces every activation decision. If a bot does something wrong, the log should explain why.
Heuristics and feature ideas are sourced from:
- VT2 Bot Improvements - Combat (Grimalackt) — per-career threat thresholds, revive-with-ability, elite pinging, boss engagement, melee selection
- VT2 Bot Improvements - Impulse Control (Squatting-Bear) — ability suppression, anti-waste conditions
- VT2 decompiled source — 14-level BT, utility-based scoring, 90+ considerations, item management
- Darktide community (Fatshark forums, Steam, Reddit) — prioritized pain points
- Darktide decompiled source (v1.10.7) — untapped perception signals, blackboard data, cover system, formation logic
- Per-class tactics docs (
docs/classes/*-tactics.md) — community-sourced USE WHEN / DON'T USE / proposed bot rules
See docs/related-mods.md for detailed mod analysis and docs/classes/*-tactics.md for per-ability heuristics.
- M1 (v0.1.0): Tier 1 + Tier 2 abilities activate in solo play. Published on Nexus.
- M2 (v0.2.0–v0.3.0): Per-career threat heuristics + Tier 3 reliability + structured event logging.
- M3 (v0.4.0): Ability quality + bot fixes — suppression, charge rescue, Psyker overcharge, revive protection, poxburster targeting, ADS fix, bot sprinting.
- M4 (v0.5.0): Ability polish + weapon fixes — VFX/SFX bleed, melee/ranged meta_data, warp venting, staff charged fire (partial).
- M5 (v0.6.0): Scope expansion — grenade/blitz, staff charged fire complete, bot pinging, daemonhost avoidance, distant special penalty.
- M5-batch2 (v0.7.0): Grenade heuristics + Psyker blitz, ping anti-spam, hazard-aware abilities, boss engagement, healing deferral, player-tag response.
- v0.7.1: P0/P1 stabilization — animation crash guard, ammo threshold, melee horde bias, Assail void throws, grenade misaim. 461 tests.