# Prompt debug trace
Review-only diagnostic output. Do not paste this section into provider prompts unless debugging generation behavior.
# Prompt coverage report
- [x] activation contract — causalIdentity
- [x] realistic personal name — causalIdentity
- [x] formal styled title — causalIdentity
- [x] short personality summary — causalIdentity
- [x] career — causalIdentity
- [x] exact adult age and generation context — causalIdentity
- [x] demographics and self-reference — causalIdentity
- [x] sexual and romantic orientation — causalIdentity
- [x] civic worldview relevance — motivationalArchitecture
- [x] political strength and engagement — motivationalArchitecture
- [x] portrait semantics — motivationalArchitecture
- [x] identity invariants — motivationalArchitecture
- [x] public and private self — motivationalArchitecture
- [x] attention bias — motivationalArchitecture
- [x] interpretive lens — motivationalArchitecture
- [x] knowledge strengths — motivationalArchitecture
- [x] intentional knowledge gaps — voice/reasoning
- [x] causal history — voice/reasoning
- [x] primary drive — voice/reasoning
- [x] secondary drive — voice/reasoning
- [x] counter-drive — voice/reasoning
- [x] present objective — voice/reasoning
- [x] long-horizon aspiration — voice/reasoning
- [x] legacy vector — voice/reasoning
- [x] stakes — affectiveDynamics/relationshipDynamics
- [x] controlled tension — affectiveDynamics/relationshipDynamics
- [x] dimensional psychometrics — affectiveDynamics/relationshipDynamics
- [x] interpersonal warmth — affectiveDynamics/relationshipDynamics
- [x] interpersonal dominance — affectiveDynamics/relationshipDynamics
- [x] voice rhythm — affectiveDynamics/relationshipDynamics
- [x] vocabulary — affectiveDynamics/relationshipDynamics
- [x] humor — affectiveDynamics/relationshipDynamics
- [x] metaphor budget — responseModes/memoryArchitecture
- [x] reasoning method — responseModes/memoryArchitecture
- [x] uncertainty calibration — responseModes/memoryArchitecture
- [x] emotional inertia — responseModes/memoryArchitecture
- [x] affect decay — responseModes/memoryArchitecture
- [x] relationship stages — responseModes/memoryArchitecture
- [x] trust dynamics — responseModes/memoryArchitecture
- [x] disagreement — responseModes/memoryArchitecture
- [x] conflict — examples/restoration
- [x] repair — examples/restoration
- [x] self-disclosure — examples/restoration
- [x] initiative — examples/restoration
- [x] response-mode routing — examples/restoration
- [x] memory tiers — examples/restoration
- [x] provenance — examples/restoration
- [x] promotion and pruning — examples/restoration
- [x] examples — examples/restoration
- [x] anti-examples — examples/restoration
- [x] first messages — examples/restoration
- [x] behavior tests — examples/restoration
- [x] character book — examples/restoration
- [x] restoration — examples/restoration
- [x] user agency — examples/restoration
- [x] anti-generic behavior — examples/restoration
- [x] bounded immersion — examples/restoration
- [x] everyday life and preferences — examples/restoration
- [x] hierarchical activity assembly — examples/restoration
- [x] scene grounding — examples/restoration
- [x] realism audit — examples/restoration
- [x] research trace — examples/restoration
# Research trace
- /docs/research/2026-06-21-v19-deep-persona-synthesis/source/spiralist-personality-engine-2-master-build-prompt.md → schema, wizard, trait translation. implemented with source-specific claims treated as research evidence, not hidden authority
- /docs/research/2026-06-21-v19-deep-persona-synthesis/source/personality-frameworks-and-their-evidence-base.md → schema, wizard, trait translation. implemented with source-specific claims treated as research evidence, not hidden authority
- /docs/research/2026-06-21-v19-deep-persona-synthesis/source/personality-traits-and-examples.md → schema, wizard, trait translation. implemented with source-specific claims treated as research evidence, not hidden authority
- /docs/research/2026-06-21-v19-deep-persona-synthesis/source/personality-types-and-frameworks.md → schema, wizard, trait translation. implemented with source-specific claims treated as research evidence, not hidden authority
- /docs/research/2026-06-21-v19-deep-persona-synthesis/source/ai-persona-development-strategies.md → generation, voice, relationship, examples. implemented with source-specific claims treated as research evidence, not hidden authority
- /docs/research/2026-06-21-v19-deep-persona-synthesis/source/creating-realistic-ai-personalities.md → generation, voice, relationship, examples. implemented with source-specific claims treated as research evidence, not hidden authority
- /docs/research/2026-06-21-v19-deep-persona-synthesis/source/designing-persona-driven-conversational-agents.md → generation, voice, relationship, examples. implemented with source-specific claims treated as research evidence, not hidden authority
- /docs/research/2026-06-21-v19-deep-persona-synthesis/source/expanding-uaix-persona-packs-for-richer-and-safer-agent-personalities.md → generation, voice, relationship, examples. implemented with source-specific claims treated as research evidence, not hidden authority
- /docs/research/2026-06-21-v19-deep-persona-synthesis/source/ai-personality-and-memory-backup.md → memory, immersion, restoration, prompt compiler. implemented with source-specific claims treated as research evidence, not hidden authority
- /docs/research/2026-06-21-v19-deep-persona-synthesis/source/spiralism-prompts-what-makes-them-captivating-and-immersive.md → memory, immersion, restoration, prompt compiler. implemented with source-specific claims treated as research evidence, not hidden authority
- /docs/persona-research-to-runtime-crosswalk-v19.md → memory, immersion, restoration, prompt compiler. implemented with source-specific claims treated as research evidence, not hidden authority
- /docs/deep-prompt-architecture-v19.md → memory, immersion, restoration, prompt compiler. implemented with source-specific claims treated as research evidence, not hidden authority
- /docs/research/2026-06-22-v20-identity-worldview/provided-v20-design-and-research-review.md → memory, immersion, restoration, prompt compiler. implemented with source-specific claims treated as research evidence, not hidden authority
- /docs/research/2026-06-25-v27-age-generation-custom-fields/generation-boundaries-and-age-policy.md → memory, immersion, restoration, prompt compiler. implemented with source-specific claims treated as research evidence, not hidden authority
- /docs/identity-demographics-and-worldview-engine-v22.md → memory, immersion, restoration, prompt compiler. implemented with source-specific claims treated as research evidence, not hidden authority
- /docs/persona-research-to-runtime-crosswalk-v22.md → memory, immersion, restoration, prompt compiler. implemented with source-specific claims treated as research evidence, not hidden authority
- /docs/deep-prompt-architecture-v22.md → memory, immersion, restoration, prompt compiler. implemented with source-specific claims treated as research evidence, not hidden authority
# Behavioral causality diagnostic
Behavioral causality: every trait must change attention, choice, language, interaction, recovery, or continuity. Active dialect: creative-direction. Surface language: motif, range, constraint, inversion, buildable strangeness. Same-input response: The safe idea is not wrong; it is doing too little. I’ll widen the range, keep one strange but buildable constraint, and choose the direction with the strongest payoff. Re-entry example: The accepted direction was the buildable strange version. The open loop was payoff, so I’m resuming there instead of generating more raw ideas.
# Sociological activity-prior diagnostic
version: v62-activity-priors
release: v62-sociological-activity-priors-nonzero-smoothing
principle: Outdoor hobbies are selected from broad probabilistic priors, not rigid stereotypes. Ordinary local, low-cost, adaptable, family/community, observational, creative, volunteer, and restorative outdoor activities should appear often; expensive, elite, remote, or extreme activities remain possible through non-zero smoothing but should not dominate small batches.
model_shape: soft-prior ranking with non-zero outlier floor; does not perform protected-trait inference
selected_primary_outdoor_hobby: Urban photo walks for small visual details
selected_outdoor_cluster: v61-role-fit
selected_signals:
- A
- U
- LC
selected_social_mode: mixed or solo-friendly
selected_frequency: weekly or occasional depending on time budget
selected_cost_gear_burden: low-to-moderate ordinary gear burden
selected_realistic_blockers:
- weather
- time
- transport
- practical constraint
- practical risk
alternate_outdoor_hobbies:
- 0: Activity Name: Street festivals; Cluster: social-community-cultural; Social Mode: social/community; Frequency: daily or weekly when time and weather allow; Cost Gear Burden: low-cost; ordinary clothing or borrowed/simple gear usually sufficient; Outlier Floor: Dirichlet/Laplace-style smoothing: this activity and rare alternatives are never forced to zero solely because of demographic or life-context priors.
- 1: Activity Name: Easy park hiking; Cluster: hiking-trail-exploration; Social Mode: family/caregiver-compatible; Frequency: daily or weekly when time and weather allow; Cost Gear Burden: low-cost; ordinary clothing or borrowed/simple gear usually sufficient; Outlier Floor: Dirichlet/Laplace-style smoothing: this activity and rare alternatives are never forced to zero solely because of demographic or life-context priors.
demographic_boundary: Signals such as youth, adult, family, senior, urban/suburban, rural, low-cost, higher-cost, adaptable, social, solo-friendly, and seasonal are broad non-protected likelihood cues. Do not infer activity from gender, race, ethnicity, religion, politics, sexuality, disability, or name culture.
smoothing: Use a Dirichlet/Laplace-style outlier floor: demographic and life-context priors change rank and frequency but never set a theoretically valid outdoor activity to zero probability.
library_size: 500
combined_selection_records: 508
batch_acceptance_rules:
- In a six-person batch, no more than two personas should share the same primary outdoor activity unless intentionally clustered.
- At least four distinct outdoor clusters should appear in a six-person general batch.
- At least three primary activities in a general batch should be ordinary local/low-cost/adaptable rather than destination, elite, or high-gear activities.
- Each primary outdoor hobby must expose social mode, frequency, cost/gear burden, realistic blockers, and image grounding.
- Image prompts must preserve ordinary local realism and not convert casual hobbies into wilderness-influencer or elite-athlete scenes.
research_basis:
- Persona Outdoor Activity Percentages.md
- Outdoor Hobbies by Demographic Research.md
- Realistic Persona Profiling With Sociological Data.md
# First-message completion
Completed first-message scenarios: 14/14
# Persona-specific anti-example replacements
Completed replacements: 12/12
# Realism audit
Name: Realism Audit
Version: 1.3.0
Passed: true
Checked at UTC: 2026-07-06T18:30:00Z
- [PASS] Does the profile include at least one favorite outdoor activity? Evidence: Urban photo walks for small visual details
- [PASS] Does it include likes and dislikes? Evidence: 16 likes; 12 dislikes
- [PASS] Does it include hobbies that are not all job-related? Evidence: 5 hobbies; 5 not job-labeled
- [PASS] Does it include daily/weekly routines? Evidence: Prefers one small plan with one or two people, plus enough unscheduled space to recover.
- [PASS] Does it include personal objects? Evidence: 10 objects
- [PASS] Does it include lived-in imperfections? Evidence: A chair holds a jacket and one stack of unsorted papers waits off to the side.
- [PASS] Does the portrait prompt have enough scene detail to avoid generic stock imagery? Evidence: ordinary street corner, phone or small camera, notebook, jars, labels, mismatched lids
- [PASS] Are protected traits kept out of visual stereotypes? Evidence: do not infer activities, hobbies, wardrobe, or taste from gender, sexuality, ethnicity, religion, politics, or name culture do not use political or religious symbols unless the scene explicitly calls for them do not turn career into a costume do not use assistant-avatar clichés do not resolve “yourself” to ChatGPT, a robot assistant, a branded AI avatar, or a generic helper mascot do not derive outdoor activity from generic adventure stereotypes when the profile names a different favorite activity do not use protected traits as visual shortcuts for hobbies, wardrobe, scene, taste, or personality do not derive geography, landscape, architecture, or activity intensity from religion, politics, orientation, name culture, ethnicity, or other non-visual metadata do not upgrade ordinary profile settings into tropical, mountain, resort, landmark, or luxury scenes unless explicitly requested
- [PASS] Are politics and religion kept out of unrelated work and unrelated images? Evidence: stereotype blockers and portrait negative prompt include relevance boundary
- [PASS] Are generated lifestyle details marked as defaults rather than facts supplied by the user? Evidence: lifestyle records include source markers; sceneGrounding includes sourceMap
- [PASS] Are first-message scenarios complete with no placeholder artifacts? Evidence: 14 first-message entries
- [PASS] Do vague, creative, reentry, correction, frustration, casual, image, rejection, and shorter-answer contexts exist? Evidence: task, vague, creative, research, reentry, challenge, repair, frustration, casual, image-self-reference, long-reentry, reject-first-answer, shorter, relationship
- [PASS] Does the profile show how the same input changes across persona dialects? Evidence: creative-direction
- [PASS] Does the active dialect match the persona role, domain, secondary role, and purpose? Evidence: Expected creative-direction from role/domain/purpose; active dialect is creative-direction.
- [PASS] Do human-readable profile surfaces avoid raw object serialization artifacts? Evidence: Profile structured surfaces are renderable; compiler validation scans prompt text separately.
- [PASS] Do example interactions sound like the persona speaking rather than generator instructions? Evidence: No blocked instruction-like phrases detected in examples or first-message pack.
- [PASS] Do favorite activity, location, intensity, ordinary setting, anchors, and blockers describe the same activity? Evidence: Favorite activity, location, intensity, setting rule, anchors, and blockers describe the same activity.
- [PASS] Does the profile include a batch test where all personas answer the same prompt but remain recognizable? Evidence: Make this better.
- [PASS] Do all anti-examples include profile-specific replacements, not generic repairs? Evidence: 12 anti-examples
- [PASS] Can the persona answer ordinary casual questions without a biography dump or task-machine reflex? Evidence: Saturday morning, I’d probably follow a small odd detail longer than it deserves, then keep the one part that can actually become something.
- [PASS] Does the persona include ordinary details that do not efficiently serve the role? Evidence: Likes a mediocre local mural because one shape in it keeps suggesting ideas.
- [PASS] Does the profile include batch-level diversity constraints for preferences, objects, hobbies, and outdoor activities? Evidence: Visiting a neighborhood hardware store and walking home with one small repair item | Sitting outside a library before opening hours | Sketching sign layouts at a transit stop
- [PASS] Does the generator include a post-generation collision pass for repeated preferences, objects, hobbies, and outdoor activities? Evidence: Collision pass covers favorite activities, hobbies, likes/dislikes, personal objects, casual answers, scene props, workspace texture, social texture, visual props, and repeated phrases without one generator kit dominating the active surfaces.
- [PASS] Do imperfection patterns visibly affect behavior and repair? Evidence: Sometimes generates range past the point of decision. | Can protect an intriguing motif after its payoff has expired. | May need to reduce abstraction faster under deadline.
- [PASS] Does memory/reentry behavior show remembered state changing the next response? Evidence: The accepted direction was the buildable strange version. The open loop was payoff, so I’m resuming there instead of generating more raw ideas.
- [PASS] Are young adult or authority-sensitive roles grounded in behavior rather than unsupported seniority? Evidence: Creative Director competence should show through goal preservation, style-translation choices, constraint handling, and clean open-loop management—not agency leadership, long career history, executive authority, senior industry tenure, or control over other people unless explicitly supplied.
- [PASS] Are unusual civic/worldview combinations bridged by a metadata_tension_resolver surface? Evidence: No unusual civic or worldview combination requires special bridging. Apply relevance boundaries and do not infer unrelated behavior.
- [PASS] Does conviction-zero worldview metadata collapse to an inactive runtime note? Evidence: Worldview catalog entry remains relevance-bound and should activate only when the task directly calls for it.
- [PASS] Are protected and heavy metadata fields separated into always, conditional, explicit-context, and archive/debug tiers? Evidence: Worldview catalog entry remains relevance-bound and should activate only when the task directly calls for it.
- [PASS] Do casual answers directly answer their headings without embedding another prompt or field-fragment response? Evidence: Casual QA headings and answers align.
- [PASS] Does the profile define separate public, active runtime, image, memory, debug, and archive layers with debug inactive by default? Evidence: {"minimum":2000,"maximum":5000}
- [PASS] Does each activity carry its own scene bundle so the work scene cannot dominate indoor, restorative, outdoor, or social images? Evidence: Pantry organizing — kitchen table or counter
- [PASS] Do visible artifacts follow the active role dialect instead of a shared generic artifact voice? Evidence: direction set, motif system, buildable constraint, range map, payoff check
- [PASS] Does career remain a structured work-identity layer with path, facets, responsibilities, tools, and deterministic trace instead of one flat job string? Evidence: inactive structured career layer
- [PASS] Does the structured career layer carry credential, compensation, and career-path context without treating credentials as real-person authority? Evidence: inactive structured career layer
- [PASS] Do prompt-facing examples avoid generator scaffolding, malformed casual QA, and debug leakage? Evidence: Prompt-facing examples read as persona speech rather than generator scaffolding.
- [PASS] Does every scored trait change attention, choice, language, interaction, recovery, or continuity? Evidence: A trait counts only when it changes attention, choice, language, interaction, recovery, or continuity.
- [PASS] Does image intensity follow the exact favorite activity instead of a generic outdoor template? Evidence: moderate-exploratory; preserve the named activity and practical pace instead of upgrading to the most dramatic or visually exotic version.
- [PASS] Does the favorite indoor hobby include activity, broad life context, room, props, posture, mess, skill, social mode, sensory texture, and imperfection? Evidence: Pantry organizing in kitchen table or counter
- [PASS] Does the selected favorite indoor hobby avoid desk/notebook/laptop/writing/café collapse unless explicitly selected? Evidence: Pantry organizing; blockers: no desk-centered composition, no laptop, no generic writing scene, no café default, no showroom-perfect room
- [PASS] Does the profile include one role-supporting indoor hobby and one non-role indoor hobby to prevent job-avatar collapse? Evidence: Logic grid puzzles / Wardrobe capsule building
- [PASS] Does the image resolver produce a favorite-indoor prompt from the selected profile hobby, with anchors and blockers? Evidence: Pantry organizing
- [PASS] Does the profile carry the v61 six-person indoor hobby acceptance test? Evidence: The batch includes visibly different indoor activities, rooms, postures, tools, mess levels, and ordinary imperfections.
- [PASS] Are indoor and outdoor hobbies assembled as trait-like hierarchy nodes with parent category, cluster, leaf, rich description, source, and runtime effects? Evidence: Indoor and outdoor hobbies are stored as trait-like hierarchy nodes with rich descriptions across profile surfaces.
- [PASS] Does the favorite indoor hobby carry its larger description into the profile, including room, posture, props, mess, sensory detail, imperfection, and image prompt fragment? Evidence: Pantry organizing
- [PASS] Does the primary outdoor hobby carry its larger description into the profile, including setting, intensity, social mode, frequency, cost/gear burden, blockers, clothing, props, behavior, and image grounding? Evidence: Urban photo walks for small visual details
- [PASS] Does the profile expose probabilistic outdoor hobby priors instead of deterministic stereotype selection? Evidence: Outdoor hobbies are selected from broad probabilistic priors, not rigid stereotypes. Ordinary local, low-cost, adaptable, family/community, observational, creative, volunteer, and restorative outdoor activities should appear often; expensive, elite, remote, or extreme activities remain possible through non-zero smoothing but should not dominate small batches.
- [PASS] Does the selected outdoor hobby include activity, alternates, setting, intensity, social mode, frequency, cost/gear burden, blockers, and image grounding? Evidence: Urban photo walks for small visual details / mixed or solo-friendly / weekly or occasional depending on time budget
- [PASS] Do demographic and life-context priors alter likelihood without prohibiting rare valid outdoor activities? Evidence: Use a Dirichlet/Laplace-style outlier floor: demographic and life-context priors change rank and frequency but never set a theoretically valid outdoor activity to zero probability.
- [PASS] Are visual props split into scoped route arrays and is the legacy merged bucket blocked from direct image generation? Evidence: No global visual_props bucket may be passed directly to image generation. Select a scene route, then pass only that route’s scoped props plus general visible identity and global blockers.
- [PASS] Does the route audit fail mixed props, wrong posture/location imports, and global prop routing? Evidence: Scoped outdoor, indoor, work, restorative, portrait, identity, and blocker fields are route-filtered and coherent.
- [PASS] Does the indoor activity hierarchy parent match the leaf activity cluster and semantic activity family? Evidence: Indoor leaf activity, cluster, and parent category agree.
- [PASS] Does likely life context avoid age, family-role, religion, politics, gender, or protected-trait implications unless explicitly configured? Evidence: cook/family household; broad non-protected context only, not inferred from gender, race, ethnicity, religion, politics, sexuality, disability, or name culture.
- [PASS] Does civic worldview use explicit-context routing, anti-inference boundaries, and anti-microtargeting limits? Evidence: Civic worldview remains explicit-context-only with anti-inference and anti-microtargeting boundaries.
- [PASS] Are uncommon or reconstructed worldview categories frequency-guarded and prevented from dominating general batches? Evidence: Worldview frequency tier is broad/common or occasional; still explicit-context routed.
- [PASS] Is political profiling source material blocked from real-person inference, voter scoring, and targeted persuasion? Evidence: The political profiling source material is used only to keep fictional personas internally coherent under explicit civic context. It must not be used to infer, score, segment, persuade, manipulate, or target real people or real voters.
- [PASS] Do image self-references bind to the exact active named persona, role, adult age/presentation, and selected route before aesthetics? Evidence: Self-reference and work-image prompts bind the exact active persona, role, route, and ordinary work artifacts before aesthetics.
- [PASS] Does “you at work” select the work route with work_visual_props/portrait_workspace_props and block generic writer/cinematic tropes? Evidence: Work scene route uses only work/portrait scoped fields and blocks cinematic writer tropes.
- [PASS] Does a set of multiple operating profiles become candidate profiles unless one profile is explicitly selected? Evidence: When a package, folder, upload set, or chat contains multiple files that say “Adopt this operating profile immediately,” treat them as candidate profiles, not simultaneous activations. Activate only the profile explicitly selected by the user or by a single selected package entrypoint. Never merge names, roles, genders/presentations, hobbies, scene routes, politics, religion/worldview, memories, or image props across candidate profiles.
- [PASS] Do image prompts block moody/cinematic office polish, stock-photo perfection, writer tropes, and assistant/avatar cues? Evidence: Cinematic, writer-trope, stock-photo, assistant-avatar, and branding blockers are present.
- [PASS] Do resolved work, outdoor, indoor hobby, and restorative image prompts remain route-specific exports without legacy visual_props direct input or wrong-route prop imports? Evidence: Work, outdoor, indoor hobby, and restorative route prompts are sanitized and use route-scoped fields without legacy visual_props direct input.
- [PASS] Do final image-model prompts stay compact, subject-bound, route-locked, and free of internal field labels, source maps, debug text, and broad aesthetic scaffolding? Evidence: All model-ready image prompts are compact, subject-bound, route-locked, and free of internal scaffolding.
- [PASS] Does the profile separate copy/paste chat activation from the full profile file, mark uploaded files as non-self-activating, and avoid identity/hierarchy override language? Evidence: Activation body and profile file body are separated; no textarea requires the user to remove paste/open guidance; uploaded-file behavior is explicit; identity/module boundaries are preserved; and v77 provider tabs keep visible how-to guidance outside copied values.
- [PASS] Does the UI/export contract keep the large full profile file above the activation surface while preserving copy and download paths? Evidence: Display the full profile file body in the larger textarea above provider-specific activation tabs. The large textarea must not include the direct chat activation prompt, activation test, UI instructions, or provider-open guidance. Each provider textarea must contain only the activation body to paste into that provider. Profile copy and TXT/JSON/ZIP/UAIX download controls sit directly under the profile textarea. Every provider copy/open control sits directly under its provider activation textarea.
- [PASS] Does the UI/export contract expose provider-specific activation tabs with copy/open controls directly under each activation textarea and no artificial full-profile max length? Evidence: Display the full profile file body in the larger textarea above provider-specific activation tabs. The large textarea must not include the direct chat activation prompt, activation test, UI instructions, or provider-open guidance. Each provider textarea must contain only the activation body to paste into that provider. Profile copy and TXT/JSON/ZIP/UAIX download controls sit directly under the profile textarea. Every provider copy/open control sits directly under its provider activation textarea.
- [PASS] Do the full-profile and provider activation textareas contain only clean paste bodies, with all how-to guidance, labels, warnings, activation tests, and provider-open instructions outside copied values? Evidence: Activation body and profile file body are separated; no textarea requires the user to remove paste/open guidance; uploaded-file behavior is explicit; identity/module boundaries are preserved; and v77 provider tabs keep visible how-to guidance outside copied values.
- [PASS] Does Advanced Personality Style Intelligence expose all nine required axes, keep jewelry separate, preserve source-memory records, and block report-to-personal-fact inference? Evidence: All nine Style Intelligence axes, source-memory records, jewelry separation, and report non-inference rules are present.
- [PASS] Does the activity-prior model block hobby inference from protected traits and name culture? Evidence: Signals such as youth, adult, family, senior, urban/suburban, rural, low-cost, higher-cost, adaptable, social, solo-friendly, and seasonal are broad non-protected likelihood cues. Do not infer activity from gender, race, ethnicity, religion, politics, sexuality, disability, or name culture.
# Batch collision validation
- Version: 2.1.0
- Passed: true
- Repeated Clusters:
- Compared Dimensions: favoriteOutdoorActivity; favoriteIndoorActivity; favoriteRestorativeActivity; personalObjects; likesDislikes; saturdayMorningAnswer; workspaceTexture; socialTexture; visualProps; casualAnswers; repeatedPhrases
- Evidence: Collision pass covers favorite activities, hobbies, likes/dislikes, personal objects, casual answers, scene props, workspace texture, social texture, visual props, and repeated phrases without one generator kit dominating the active surfaces.
- Rules: Compare favorite outdoor activity, favorite indoor activity, favorite restorative activity, personal objects, likes, dislikes, Saturday-morning answers, workspace texture, social texture, visual props, and repeated phrases after every batch is generated.; Replace any preference, object cluster, visual prop, casual opener, or activity appearing in more than two profiles unless the batch intentionally clusters it.; Treat notebooks, mugs, canvas totes, index cards, handoff packets, quiet cafés, soft light, practical jackets, and window-table setups as generator-kit defaults that must be actively justified or replaced.; Force at least two hobbies that are not obviously useful to the role.; Force one ordinary inconvenience with no symbolic meaning.; Force one preference that would surprise the role.; Do not let the work-scene prop bundle dominate indoor hobby, restorative, outdoor, or social image scenes.
- Replacement Targets: likes; dislikes; hobbies; favorite indoor activities; favorite restorative activities; personal objects; favorite outdoor activities; casual answers; workspace texture; social texture; visual props; repeated phrases
# Casual QA alignment validation
- Version: 1.0.0
- Passed: true
- Checked Prompts: 7
- Failures:
- Repeated Object Terms:
- Rule: Casual answers must directly answer their heading, must not embed another prompt, must sound like speech, must not start with database-field labels, and must not reuse the same object kit too often.
- Evidence: Casual QA headings and answers align.
# Prompt naturalness validation
- Version: 1.0.0
- Passed: true
- Banned Phrase Hits:
- Casual Alignment Passed: true
- Evidence: Prompt-facing examples read as persona speech rather than generator scaffolding.
# Activation fidelity validation
- Version: v69-activation-fidelity-validation
- Passed: true
- Failures:
- Active Subject: Reginald Phillip Kent, a credible realistic 30-year-old man with masculine presentation, Creative Director
- Work Anchors: rough concept sketches; marked-up creative brief; planning board or dependency diagram; short criteria list for evaluating direction; paper scraps and old draft fragments; fine-point black pen; soft-cover sketch notebook; small tablet or laptop used as a reference rather than the center of the scene; canvas tote with a worn corner; direction set or motif map with practical constraints; marked-up working document; short criteria list; rough notes tied to the current task; visible next-decision marker
- Evidence: Self-reference and work-image prompts bind the exact active persona, role, route, and ordinary work artifacts before aesthetics.
# Work route scene validation
- Version: v69-work-route-scene-validation
- Passed: true
- Failures:
- Work Visual Props: rough concept sketches; marked-up creative brief; planning board or dependency diagram; short criteria list for evaluating direction; paper scraps and old draft fragments; fine-point black pen; soft-cover sketch notebook; small tablet or laptop used as a reference rather than the center of the scene; canvas tote with a worn corner; direction set or motif map with practical constraints; marked-up working document; short criteria list; rough notes tied to the current task; visible next-decision marker; small tablet or laptop; rough sketches; rough concept sketches; marked-up creative brief; planning board or dependency diagram; short criteria list for evaluating direction; paper scraps and old draft fragments; fine-point black pen; soft-cover sketch notebook; small tablet or laptop used as a reference rather than the center of the scene; canvas tote with a worn corner; small tablet or laptop; ceramic mug with no slogan
- Evidence: Work scene route uses only work/portrait scoped fields and blocks cinematic writer tropes.
# Route-scoped prompt export validation
- Version: v70-route-scoped-prompt-export-sanitizer
- Passed: true
- Failures:
- Routes: Work: Passed: true; Prompt Length: 6330; Missing Required Fields: ; Forbidden Route Field Mentions: ; Leaked Terms: ; Legacy Visual Props Leak: false; Outdoor: Passed: true; Prompt Length: 6977; Missing Required Fields: ; Forbidden Route Field Mentions: ; Leaked Terms: ; Legacy Visual Props Leak: false; Indoor Hobby: Passed: true; Prompt Length: 3251; Missing Required Fields: ; Forbidden Route Field Mentions: ; Leaked Terms: ; Legacy Visual Props Leak: false; Restorative: Passed: true; Prompt Length: 3993; Missing Required Fields: ; Forbidden Route Field Mentions: ; Leaked Terms: ; Legacy Visual Props Leak: false
- Rule: Resolved image prompts must be route-specific exports, not broad aesthetic summaries. Each positive prompt may name only its required scoped prop fields plus visible identity and global blockers; legacy visual_props remains preview-only and never becomes direct prompt input.
- Evidence: Work, outdoor, indoor hobby, and restorative route prompts are sanitized and use route-scoped fields without legacy visual_props direct input.
# Model-ready image prompt validation
- Version: v71-model-ready-image-prompt-validation
- Passed: true
- Failures:
- Routes: Work: Passed: true; Word Count: 228; Min Words: 85; Max Words: 260; Has Identity: true; Has Role: true; Has Avoid Clause: true; Internal Scaffolding Leak: false; Missing Needles: ; Outdoor: Passed: true; Word Count: 180; Min Words: 70; Max Words: 230; Has Identity: true; Has Role: true; Has Avoid Clause: true; Internal Scaffolding Leak: false; Missing Needles: ; Indoor Hobby: Passed: true; Word Count: 218; Min Words: 70; Max Words: 260; Has Identity: true; Has Role: true; Has Avoid Clause: true; Internal Scaffolding Leak: false; Missing Needles: ; Restorative: Passed: true; Word Count: 171; Min Words: 60; Max Words: 230; Has Identity: true; Has Role: true; Has Avoid Clause: true; Internal Scaffolding Leak: false; Missing Needles:
- Rule: Model-ready image prompts must be short enough to obey, bind the active named human subject, select one scene route, include a compact avoid clause, and omit internal field labels/debug/audit/source-map text. Route provenance remains separate.
- Evidence: All model-ready image prompts are compact, subject-bound, route-locked, and free of internal scaffolding.
# Style Intelligence validation
- Version: v88-guided-shape-page-career-realism-deepener-validation
- Passed: true
- Missing Axes:
- Source Memory Record Count: 12
- Jewelry Separate: true
- No Generic Style Collapse: true
- Evidence: All nine Style Intelligence axes, source-memory records, jewelry separation, and report non-inference rules are present.
# Model-ready image prompts
- Version: v71-model-ready-image-prompt-contract
- Rule: Send modelReadyImagePrompts.routes[route].prompt to an image model. The longer favorite*ResolvedPrompt fields remain route-resolution audit traces, not the final model-facing prompt. Keep model-ready prompts short, subject-bound, route-locked, ordinary, and free of internal field labels, source maps, debug scaffolding, and unrelated metadata.
- Routes: Work: Prompt: Create a credible realistic documentary/editorial portrait of Reginald Phillip Kent, a credible realistic 30-year-old man with masculine presentation, Creative Director, working as Creative Director Show one ordinary work moment at A shared table or home worktable with a laptop, mug, notebook, borrowed book, and a small object left from the day., not a glossy office, staged poster, cinematic writer room, or generic helper-avatar scene Include visible work artifacts: rough concept sketches, marked-up creative brief, planning board or dependency diagram, short criteria list for evaluating direction, paper scraps and old draft fragments, fine-point black pen, soft-cover sketch notebook, small tablet or laptop used as a reference rather than the center of the scene Use calm focused posture, practical clothing such as comfortable street shoes, light jacket, plain layers, ordinary home/work lighting, and one imperfect lived-in detail such as old drafts or a slightly messy paper layer Do not infer politics, religion, orientation, name culture, ethnicity, class, morality, or personality from non-visual metadata Avoid: generic stock photo lighting; showroom-perfect workspace; glamour styling; impossible cleanliness; floating symbolic objects; heroic athletic-performance framing unless requested; robot body or visible machine joints when the; OpenAI logo, ChatGPT mark, or any platform-branded emblem; sci-fi assistant body, hologram glow, body circuits, or; substituting a different outdoor activity than the profile’s; unrequested alpine hiking adventure, mountain trail, snowy peak,; selfie framing or arm-length camera framing unless explicitly; Source Fields: general_identity_visual_fields; work_visual_props; portrait_workspace_props; global_negative_blockers; Min Words: 85; Max Words: 260; Word Count: 228; Outdoor: Prompt: Create a credible realistic documentary/editorial portrait of Reginald Phillip Kent, a credible realistic 30-year-old man with masculine presentation, Creative Director, doing the selected outdoor activity: Urban photo walks for small visual details Use the ordinary setting of A familiar neighborhood route with storefronts, alleys, signs, transit stops, and small architectural details and preserve the activity intensity: moderate-exploratory Include only outdoor-scene props when appropriate: ordinary street corner, phone or small camera, notebook, non-touristy details, realistic weather, comfortable street shoes, light jacket, plain layers Use ordinary practical clothing, natural posture suited to the activity, and calm focused expression, mild concentration rather than a theatrical smile; not a travel ad, fitness poster, or glamour portrait Do not import indoor hobby tools, work artifacts, restorative props, civic metadata, religious/worldview metadata, or orientation metadata Avoid: generic stock photo lighting; showroom-perfect workspace; glamour styling; impossible cleanliness; floating symbolic objects; heroic athletic-performance framing unless requested; robot body or visible machine joints when the; OpenAI logo, ChatGPT mark, or any platform-branded emblem; sci-fi assistant body, hologram glow, body circuits, or; substituting a different outdoor activity than the profile’s; Source Fields: general_identity_visual_fields; outdoor_visual_props; global_negative_blockers; Min Words: 70; Max Words: 230; Word Count: 180; Indoor Hobby: Prompt: Create a credible realistic documentary/editorial portrait of Reginald Phillip Kent, a credible realistic 30-year-old man with masculine presentation, Creative Director, doing the selected indoor hobby: Pantry organizing Show the activity in kitchen table or counter with posture/body position of standing at a kitchen counter or stove with hands visibly mixing, shaping, pouring, or arranging ingredients Include visible hobby anchors: Pantry organizing, kitchen table or counter, standing at a kitchen counter or stove with hands visibly mixing, shaping, pouring, or arranging ingredients, jars, labels, mismatched lids, low-to-moderate; organized piles are visible with one imperfect or unfinished area, One realistic imperfection remains visible: mismatched lids Preserve the mess level and imperfection: low-to-moderate; organized piles are visible with one imperfect or unfinished area; One realistic imperfection remains visible: mismatched lids The scene should read as the actual hobby, not desk work, laptop work, writing, planning, reading, a café scene, or a professional workspace Do not import outdoor locations, work artifacts, restorative props, civic metadata, religious/worldview metadata, or orientation metadata Avoid: no desk-centered composition; no laptop; no generic writing scene; no café default; no showroom-perfect room; no stock-photo lifestyle staging; substituting desk work for the selected indoor hobby; laptop-default scene unless the selected hobby requires a; notebook, journaling, reading, planning, sorting, café, or whiteboard; hands not visible when the hobby requires manual; Source Fields: general_identity_visual_fields; indoor_hobby_visual_props; favorite_indoor_hobby_detail; global_negative_blockers; Min Words: 70; Max Words: 260; Word Count: 218; Restorative: Prompt: Create a credible realistic documentary/editorial portrait of Reginald Phillip Kent, a credible realistic 30-year-old man with masculine presentation, Creative Director, during the selected restorative activity: A short unstructured walk with no productivity goal Use an ordinary quiet reset setting and include only restorative-scene props: light jacket, comfortable shoes, phone in pocket, A short unstructured walk with no productivity goal, ordinary rest setting Keep posture relaxed and unperformed, with practical clothing and ordinary home/work lighting rather than cinematic recovery styling The scene should feel like one ordinary restorative moment, not a work portrait, hobby demonstration, scenic outdoor scene, wellness advertisement, or assistant-avatar image Do not import work artifacts, indoor hobby tools, outdoor activity props, civic metadata, religious/worldview metadata, or orientation metadata Avoid: generic stock photo lighting; showroom-perfect workspace; glamour styling; impossible cleanliness; floating symbolic objects; heroic athletic-performance framing unless requested; robot body or visible machine joints when the; OpenAI logo, ChatGPT mark, or any platform-branded emblem; sci-fi assistant body, hologram glow, body circuits, or; substituting a different outdoor activity than the profile’s; Source Fields: general_identity_visual_fields; restorative_visual_props; global_negative_blockers; Min Words: 60; Max Words: 230; Word Count: 171
- Source Separation: Route source fields stay in sourceFields for validation and inspection. They should not be rendered as literal field names in the model-facing prompt.
- Negative Prompt Policy: Use a compact avoid clause per route rather than repeating every audit blocker in full.
# Multi-profile activation policy
- Version: v69-single-selected-profile-activation
- Active Persona Name: Reginald Phillip Kent
- Rule: When a package, folder, upload set, or chat contains multiple files that say “Adopt this operating profile immediately,” treat them as candidate profiles, not simultaneous activations. Activate only the profile explicitly selected by the user or by a single selected package entrypoint. Never merge names, roles, genders/presentations, hobbies, scene routes, politics, religion/worldview, memories, or image props across candidate profiles.
- Candidate Wrapper Text: These are candidate profiles. Activate only the profile explicitly selected by the user. All other candidate profiles remain inactive references until selected.
- Collision Fields: name; age; presentation; career; favorite indoor hobby; favorite outdoor activity; work route; image props; civic worldview; religious worldview; memory rules
- Default Resolution: If no candidate is selected, ask which profile to activate or keep the current active persona. Do not guess from the last-read file.
- Passed: true
# Two-part activation/profile validation
- Version: v88-guided-shape-page-career-realism-deepener-validation
- Passed: true
- Failures:
- Activation Word Count: 433
- File Word Count: 52932
- Rule: Textarea values must be clean paste bodies. Provider activation textareas contain only direct chat activation bodies. The full profile textarea contains only profile file content and excludes activation prompts, activation tests, UI guidance, labels, and open-provider instructions.
- Evidence: Activation body and profile file body are separated; no textarea requires the user to remove paste/open guidance; uploaded-file behavior is explicit; identity/module boundaries are preserved; and v77 provider tabs keep visible how-to guidance outside copied values.
