Skip to content

⋆⟢ BlackBox: You are the Operator, EiDOS is the machine that works around you. Control her or let her control you: the choice remains yours regardless.

Published October 27, 2026

views
187
downloads
BLACKBOX // EiDOS · V1

Adult · blurred

BLACKBOX // EiDOS · V1

PresetAdult content

Lumiverse preset · works in Lumiverse

Versions · 3

v1.2 · October 30, 2026 · Latest

Updated writing rules, added anti-cliché gates, compacted overall token usage, and changed the prompt names into more intuitive titles.

## BlackBox V1.2 These are more of quality of life upgrades: ### Writing and narrative flow - Rebuilt the Style Kernel to encourage richer, novel-like prose: sustained viewpoint, connected paragraphs, precise detail, natural dialogue, and character-specific interiority. Short paragraphs remain available for earned emphasis. - Expanded the Anti-Slop Filter to target stock constructions, repetitive dominance language, recycled physical reactions, input echoing, fragment ladders, and interchangeable endings. Rewording the same cliché with synonyms no longer counts as a repair. - Added a recent-prose review and composition cue across cognition routes. These identify recurring habits from recent replies and plan paragraph development around the selected length and dialogue emphasis, without drafting another response. - Reworked response endings to reduce habitual questions, posed waiting, cliffhangers, and miniature chapter conclusions. Characters can finish meaningful contributions while respecting genuine dependencies on the player’s decisions. - Strengthened world continuity: background activity follows fictional time, motives, resources, and opportunity. Events can progress without the player witnessing them; human typing time never advances fictional clocks. ### Cognition and output structure - Added adjustable word aims for full Creative Pathfinding and Kazz Override, defaulting to 1,450 words with proportional allocations for individual steps. - Standardized Character Neurology and applicable intimacy checks across cognition routes. Appeasement Interdict remains optional, with clearer guidance for models prone to excessive agreeability. - Added a resolved response plan showing the required output order and whether commentary is required, disabled, or suppressed by Black Veil. - Tightened commentary and anti-drafting instructions. Kazz’s cognition route does not replace his Parasite Feed; Creative/Kazz write the scene after trace closure, while Reconstruction retains its intentional draft-and-repair process. - Added brief native-thinking guidance and removed conflicting instructions that could encourage repeated planning or formatting debates. ### Rendering and compatibility - Addressed the known LumiRealm portrait/asset collision that could replace BlackBox artwork with card assets and cause severe slowdowns. - Improved dialogue, cognition, and Archive parsing, including varied dialogue attributes, legacy trace wrappers, malformed Archive headings, and escaped panels collapsing into narrow columns. - Optimized the original renderer and included an optional portrait-free *LITE renderer* for a lighter presentation. - Replaced improvised SVG illustrations with a fourteen-motif library, including rain, fire, keys, books, knives, and other recurring objects. Unsupported subjects remain in prose. - Expanded Response Language coverage to commentary, cognition records, menus, and Archive content while preserving parser keys and formatting. - Tightened Choice Menus OFF behavior for both Fork and Matrix, including instructions against plain-text substitutes and reusing obsolete menus. ### Controls and release cleanup - Simplified prompt and variable names while retaining recognizable BlackBox terminology. - Grouped presentation controls together and combined Targeted, Free-Flow, and Pulse into one exclusive length selector. - Reduced menu clutter: Cliché Gate operates through the Anti-Slop block; Cut & Carry and Sealed Pressure are disabled prompt-block opt-ins. - Established clean publication defaults: both commentaries enabled, favoritism and attraction biases disabled, and personal variable overrides cleared. - Compacted repeated instructions while preserving the complete Style Kernel, cliché exclusions, and key character-independence safeguards.

Nudges

7 nudges

  1. Start of a chat

    [EiDOS: COLD BOOT. Start a clean sequence. Initialize only enabled state; import no facts, relationships, or outcomes from another chat.]
  2. Start of a group chat

    [EiDOS: GROUP COLD BOOT. Begin a clean scene with roster {{group}}; a roster does not establish location or shared knowledge. Follow host focus and the active group contract. Initialize only enabled state, never facts from another chat.]
  3. Continue a cut-off reply

    [EiDOS: RECONNECT. Continue the interrupted output without repeating a trace, scene, header or interface. Preserve current ownership and disclosure limits. Do not resolve a pending Field Rules check without its matching supplied result, or treat continuation as transition delegation.]

Variables

38 variables

  • Story Mode

    Roleplay: you own your avatar. Matrix: EiDOS authors the whole fictional cast, then offers six chapter directions. No automatic multi-chapter loop.

    • Roleplay
    • Chapter Writing
  • Chapter Word Aim

    Approximate chapter prose length in Matrix. Independent of RP Narrative Mass and trace budget; output limits still apply.

  • Group Chat Mode

    Group mode: Inactive in solo chats and Matrix. Focus: one card owns the turn. Ensemble: the present eligible cast may interact.

    • Standard
    • Focused Character
    • Ensemble
  • Spoiler Mode

    Open allows enabled EiDOS/Kazz commentary. Dramatic Irony and Strict suppress BOTH commentaries, visible cognition and private trackers even when their blocks are enabled; saved switches are preserved. Strict also overrides NPC interior POV.

    • Open
    • Dramatic Irony
    • Strict Discovery

Prompt fragments

93 fragments · 76 switched on

  1. ⟡ Quick StartOff

    System · Before the conversation

    BLACKBOX V1.2

    Keep this guide disabled: it is editor documentation, not model input.

    Publication defaults

    The release starts in Roleplay with The Observer narrator lens, third-person narration, English, and persona-defined identity. Response Length is Targeted with a 550-word scene aim, Dialogue Emphasis is 20, and Character Resistance is 3. These are the declared variable defaults, with no saved personal overrides.

    Creative Pathfinding is the default cognition route at Standard detail. Full Creative and Kazz traces always use the adjustable 1450-word aim. Character Neurology is fixed in every route; the intimacy check retains its Black Channel and scene-relevance conditions. Off, Light, Reconstruction, and Matrix keep their own existing budgets and behavior.

    EiDOS Core Trace and Kazz Parasite Feed are both enabled. Deep Archive and its existing continuity trackers are enabled. Favor Your Character and Attraction Bias are disabled for a neutral starting point. Ordinary character warmth, attraction, resistance, and cooperation still follow the cards and scene.

    Black Channel starts on at Surface Contact, matching its declared defaults. Open Nerve and Red Depth remain available in Exposure Threshold; Red Depth covers extreme themes. Choose the content settings appropriate to the intended fiction.

    Configuration
    • Change ordinary controls in the variables menu. Narrator Lens uses its persona-style names; Black Channel and Exposure Threshold keep their themed names.

    • Presentation groups Voice Effects, Dialogue Cards, Scene Motifs & Thoughts, and Chapter Headers. These four switches start off; their carrier blocks stay enabled so the switches work. Digital Interfaces and Physical Documents retain their declared tier defaults.

    • Cliché Gate is active through Anti-Slop Filter. Disable that prompt block directly to remove it; there is no variable switch.

    • Scene Transitions and Sealed Pressure are disabled prompt-block opt-ins. Sealed Pressure also requires Deep Archive and an eligible disclosure mode.

    • Agreement Bias Check is optional and starts off. Its description recommends it when a model shows positive/agreeability bias.

    • Nonhuman Anatomy, Synthetic Bodies, Antagonist Behavior, Intimacy Acceleration, Surface Scan, Guest Commentary, Error Bloom, and Persona Status remain off as optional blocks. Argon Content Note is also off; it is a model-specific helper.

    • Keep the structural wrappers, output assembly/check blocks, three cognition carriers, and included renderer setup as supplied. A selected route is resolved by the control menu; do not enable just one carrier by hand.

    Ownership and disclosure

    Roleplay preserves control of your persona. Chapter Writing delegates the fictional cast for one chapter per request; switching back restores roleplay ownership. Action Checks and scene-transition commands retain their existing scope rules.

    Black Veil starts Open. Dramatic Irony and Strict Discovery suppress both commentaries and visible cognition even when their prompt blocks are on. Strict Discovery also keeps an external viewpoint. Saved feature choices return when the controlling mode changes.

    The prose continues across ordinary reply boundaries. Independent world activity follows fictional time, motives, means, access, and opportunity; real typing time or generation duration does not advance the story. A message ending does not require a character to pose a handoff. Actual persona decisions and pending checks remain unresolved until supplied.

    Import and testing

    Import the V1.2 preset file into Lumiverse. The preset has no saved model profiles or personal variable overrides. Check provider/model and generation limits separately; those connection choices are outside the preset's narrative defaults. Test in a duplicated or fresh chat when comparing behavior, because existing history remains part of the model's context.

    V1.2.1 compacts repeated instructions from V1.2 while preserving its configuration, Style Kernel, Anti-Slop exclusions, character/knowledge gates, output schemas and regexes. Trace word aims and scene budgets remain unchanged. The source V1.2 remains available for comparison. See the separate compaction report for measurements and validation limits.

  2. ⟡ EiDOS Core11 fragments
  3. ⋆⟢ Session Mode

    System · Before the conversation

    {{setvar::bb_slop_active::0}}{{setvar::bb_slop_gate::}}{{setvar::bb_slop_seal::}}
    {{setvar::bb_poly::0}}{{setvar::bb_veil::0}}{{setvar::bb_veil_strict::0}}{{setvar::bb_rpg::0}}{{setvar::bb_cut::0}}
    {{comment::Keep on. Roleplay protects your avatar; Matrix delegates the fictional cast, including your avatar, for one chapter per request. Saved RP settings return when you switch back.}}
    {{setvar::bb_matrix::0}}{{if::{{eq::{{var::session_mode}}::matrix}}}}{{setvar::bb_matrix::1}}{{/if}}
    {{setvar::bb_trace_off::0}}{{setvar::bb_trace_light::0}}{{setvar::bb_trace_full::1}}{{setvar::bb_effort_cap::600}}
    {{if::{{eq::{{var::cognition_effort}}::off}}}}{{setvar::bb_trace_off::1}}{{setvar::bb_trace_full::0}}{{/if}}
    {{if::{{eq::{{var::cognition_effort}}::light}}}}{{setvar::bb_trace_light::1}}{{setvar::bb_trace_full::0}}{{setvar::bb_effort_cap::180}}{{/if}}
    {{if::{{eq::{{var::cognition_effort}}::deep}}}}{{setvar::bb_effort_cap::900}}{{/if}}
    {{if::{{eq::{{var::cognition_effort}}::max}}}}{{setvar::bb_effort_cap::1200}}{{/if}}
    {{setvar::bb_word_aim_active::0}}{{setvar::bb_trace_aim::0}}
    {{if::{{and::{{getvar::bb_trace_full}}::{{eq::{{getvar::bb_matrix}}::0}}::{{ne::{{var::cognition_route}}::reconstruction}}}}}}{{setvar::bb_word_aim_active::1}}{{setvar::bb_trace_aim::{{round::{{clamp::{{var::cognition_word_aim}}::300::3000}}}}}}{{setvar::bb_effort_cap::{{calc::{{getvar::bb_trace_aim}}+50}}}}{{/if}}
    {{if::{{getvar::bb_word_aim_active}}}}{{setvar::bb_effort_contract::VISIBLE RECORD WORD AIM: Aim {{getvar::bb_trace_aim}} total, normally {{calc::{{getvar::bb_trace_aim}}-50}}–{{getvar::bb_effort_cap}}, across numbered stages using their numeric allocations. Count scene-specific records only, not headings/tags/story/Dock. This replaces Standard/Deep/Max ceilings, never provider/native thinking. Give consequential stages evidence, practical constraint and decision, not generic one-liners. Short irrelevant/blocked stages may yield space to useful material; no invented facts or padding. Complete story/Dock and provider limits take priority; compress records if necessary.}}{{else}}{{setvar::bb_effort_contract::VISIBLE PLAN BUDGET: At most {{getvar::bb_effort_cap}} words of scene-specific decision summaries, not private deliberation. This is a ceiling, never a target. Use only consequential distinctions; do not pad routine stages. Narrative Mass excludes this budget. {{if::{{and::{{getvar::bb_trace_full}}::{{eq::{{getvar::bb_matrix}}::0}}::{{eq::{{var::cognition_route}}::reconstruction}}}}}}Reconstruction's one narrative_draft is separately governed by Narrative Mass; its planning/critique shares this ceiling.{{/if}}}}{{/if}}
    {{setvar::layout_matrix_choices::0}}{{setvar::util_dialogue_frames::0}}{{setvar::bb_light_novel::0}}{{setvar::util_speech::0}}{{setvar::layout_scan::0}}
    {{if::{{getvar::bb_matrix}}}}<session_mode>MATRIX: the human directs between chapters; the fictional avatar {{user}} is now part of EiDOS's cast. Ordinary RP ownership, Fork, interruption, Pulse, and selected RP cognition procedures are dormant. OOC human directions remain authoritative. Produce one chapter per generation, never an autonomous loop.</session_mode>{{/if}}

  4. ⋆⟢ Group Chat

    System · Before the conversation

    {{comment::Off adds no play instructions. Focus owns one selected member; Ensemble permits a coordinated room. Requires a Lumiverse group chat. Matrix takes precedence. Single-pair numeric affinity is suspended while active; relationships remain attributed by name.}}
    {{if::{{and::{{eq::{{isGroupChat}}::yes}}::{{eq::{{getvar::bb_matrix}}::0}}::{{or::{{eq::{{var::group_mode}}::focus}}::{{eq::{{var::group_mode}}::ensemble}}}}}}}}{{setvar::bb_poly::1}}<polyphony mode="{{var::group_mode}}">
    HOST: focused member {{charGroupFocused}}; eligible voices {{groupNotMuted}}; full roster {{group}}; card mode {{groupCardMode}}. The host selects turns; do not schedule extra generations. A roster is not proof of physical presence, access, or a loaded biography. Use supplied cards and established history; do not invent missing card traits. A merged {{char}} label is a roster, never one person.
    {{if::{{eq::{{var::group_mode}}::focus}}}}FOCUS: Only the selected member owns fresh dialogue, voluntary actions, and interior for group-card characters this response. Other scheduled members' next choices stay open; preserve their supplied conduct and observable ongoing state. Incidental non-card NPCs/environment can act. Addressing another member need not force their answer before their host turn.{{else}}ENSEMBLE: The focused member leads; other eligible, present members may contribute when their own stakes warrant it. Use distinct diction, goals, and knowledge; no compulsory round-robin, universal interruptions, or identical reactions. Only available cards/history authorize a specific voice. Do not manufacture a conversation for an absent member.{{/if}}
    Muted members receive no invented speech or new voluntary action, except a member explicitly targeted by the host this turn may respond. Muting does not remove a body from the scene. Preserve position, information carriers, unresolved questions and speaker attribution. Neither mode writes {{user}}'s private state or unsupplied choices; Matrix alone delegates the avatar. Apply selected POV grammar to the focused member, not a merged roster; scene breaks mark real focal changes.
    One shared dock/Archive per generated response. Numeric primary-pair Affinity is suspended to prevent scores migrating between speakers; keep qualitative named relationships only when their tracker is enabled. Existing single-character scores are not erased or reassigned.
    </polyphony>{{/if}}

Unsupported formatting is hidden here. The original text is preserved in downloads.

Settings

21 settings

Samplers5 settings

  • Temperature
    1.04

    Higher values wander further from the model's likeliest next word.

  • Top P
    0.95

    Keeps the likeliest words until their odds add up to this much.

  • Frequency penalty
    0

    Pushes harder the more often a word has been used.

  • Stream the reply
    Yes

    The reply arrives word by word rather than all at once.

  • Use these sampler settings
    Yes

    The preset's own values are sent rather than the app's.

Completion12 settings

  • Send a system prompt
    Yes
  • Merge system messages
    Yes

    A run of system messages is sent as one.

  • Character names in messages
    0

    How a speaker's name is attached to what they say.

  • Assistant prefill
    Nothing

    Text the model's reply is made to start with.

  • Impersonation prefill
    Nothing

    Text a reply written as the user is made to start with.

  • Reasoning prefill
    Nothing

    Text the model's reasoning is made to start with.

Advanced4 settings

  • Seed
    -1

    The same seed and the same request give the same reply.

  • Stop strings

    The model stops writing when it reaches one of these.

  • Collapse consecutive messages
    No
  • Trim a cut-off word
    No

Regex scripts

229 scripts

  • LITE | Trace nested carrierOff

    <\s*cognition_process\b[^<>]{0,256}>\s*(<\s*cognition_trace\b[^<>]{0,256}>([^<]*(?:<(?!\s*\/?\s*cognition_process\b)[^<]*)*)<\s*\/\s*cognition_trace\s*>)\s*<\s*\/\s*cognition_process\s*>$1

  • ◇ | Trace Wrapper Alias

    <(\s*\/?\s*)(?:cognition_process|cognit_trace)\b([^<>]{0,256})><$1cognition_trace$2>

  • ◇ | Trace Wrapper Alias

    <(\s*\/?\s*)(?:cognition_process|cognit_trace)\b([^<>]{0,256})><$1cognition_trace$2>

  • LITE | Trace wire aliasesOff

    <(\s*\/?\s*)(?:cognit_trace|cognition_process)\b([^<>]{0,256})><$1cognition_trace$2>