Home / Product / Atlas and Studio

Affirmology Studio Spec v1

Updated Jun 15, 2026 · Affirmology_Studio_Spec_v1.md

Affirmology Studio Spec v1

The internal tool where Sol and Jeff (later Colin and an inner circle) create, test, version, rate, and curate audios across many categories. Lives at studio.affirmology.ai, password protected. Goal: empower endless testing, keep what lands, drop what does not, push the good ones to the inner circle, and learn from feedback over time.

Who and where

• Users now: Sol and Jeff, each their own login. Later: Colin and a growing inner circle.
• URL: studio.affirmology.ai, gated. It works the same whether driven from the web page or from Jeff's Claude desktop (same backend), so Jeff can run projects from chat too.

Everything is chart-driven (core principle, non-negotiable)

Every audio the Studio produces is generated FROM a specific person's unique chart. Always. There is no chartless audio. The Studio is a set of plug-and-play STRUCTURES (reusable templates) that get customized to anyone's chart. We are building scalable systems.

• A render = a PERSON (their birth data and chart) + a STRUCTURE (a category plus a high-level concept direction). The chart supplies the content; the structure decides the shape: which systems to lead with, the focus, the tone, the length.
• "Concept" is HIGH LEVEL: the design of the audio TYPE, not a finished script. Example: "a before-bed audio that leads with the Moon and the Gene Keys shadow-to-gift movement, calming, about 8 minutes." The system applies that structure to the chosen person's chart.
• Whose chart: pick a saved person (Sol, Jeff) or enter birth data for anyone. Sol testing defaults to her own chart. This is exactly how a structure becomes reusable for the whole inner circle.
• Output: the grounded script, the ElevenLabs voice (Charlotte) now, and the report with the cosmic blueprint per section. Music bed comes later.

Categories (tabs), drawn from the existing docs

Grouped. Sol and Jeff can open any tab, start a project, and also create new categories and new project types.

• Foundational: the natal Sacred Audio (the Charlotte walk).
• Gene Keys journeys: full Activation Sequence; Venus Sequence only; Pearl/Prosperity (abundance) sequence only; single-sphere deep dive.
• Astrology, subconscious-focused: your chart read powerfully; current transits and the moment you are in; Solar/Lunar Return; Astrocartography; Vocational; Health and body.
• Relationships: Synastry; Business partnership; Family Sun-sign overlay; Parent-child / family-of-origin; Soul-family / joint awakening.
• Human Design: type, strategy, and authority deconditioning; single-center deep dive.
• Numerology: Life Path and Personal Year chapter read.
• Subconscious and somatic: before-bed subconscious reprogramming; EFT tapping (the kind Sol built with her mom and sister); crisis and grief companion; confidence and abundance; founder and brand confidence (the Sol-and-Jeff alignment suite).
• Integrative: "all systems agree on X" multi-tradition read.
• Conversational Q&A (future).
• Custom: a blank free-concept project, and "create a new category."

The flow for any project

1. Pick whose chart to use (Sol, Jeff, or enter a new person's birth data). Pick a category. Enter the high-level concept direction for this audio type (the structure: which systems lead, focus, tone, length). This is design, not a finished script.
2. Reflect-back step: the system echoes its understanding and offers 2 to 3 choices or questions (for example: lead with astrology or gene keys, tone, length, which sequence). She confirms or adjusts.
3. Generate: script in house style, then Charlotte voice mp3.
4. Result screen: listen, read the full script, see the report (in chart mode, the blueprint per section), download the mp3 and the report.
5. Feedback: a rating plus notes. Actions: Keep, Drop, save as a new Version, or Push to inner circle.

Cost discipline: script-first, voice on demand

The expensive step is the ElevenLabs voice (about 50 cents per 6-minute audio, more for long journeys). Scripts and build reports are pennies. So the Studio's primary loop is SCRIPT-FIRST. "Generate script" produces the chart-grounded script plus the team build report only, no voice, using a strong model (Sonnet or Opus) for the script and a cheaper model (Haiku) only for the back-office passes (build report, attribution map). We iterate, rate, and A/B at the script level for almost nothing. "Render audio" (ElevenLabs Turbo, half credits) is a SEPARATE, explicit, on-demand action used only on versions we choose to keep. ElevenLabs Pro (~200 audios/mo) suits the testing phase; voice long journeys deliberately.

The backend "build report" (team-facing, NOT client-facing)

Each render produces a report aimed at the person crafting the audios, not the end listener. Beyond mapping each script section to the chart element behind it, it must surface the higher-level thinking: what structure was chosen and why, which chart elements were used and why, which were deliberately NOT used and why, the tone and length decisions, and suggestions ("this could also be useful," alternative directions worth trying). This is what lets the team improve the audios fast and build a suite they trust enough to share with demo people and investors. Client-facing reports come later.

Feedback, A/B, and multi-person testing

• Typed feedback gets a number attached (the comfort vote 1 to 5), so qualitative notes become trackable signal.
• This is iterative: many versions before one is "push it." Support A/B, take a render and spin a variant that goes a deliberately different direction, and compare.
• Multiple people can experience and rate the same audio or version within a category (a render accumulates feedback from several testers, not just one).
• Sol and Jeff can create new categories and new structures at any time.
• Gene Keys structures to have ready: full sequence, Venus Sequence, the first four (Activation), and the deeper one.
• Leaderboard: a view scoring every audio and version by how much attention it has had (renders, versions, number of testers, total feedback) and by comfort scores, so we can see what is featured and polished versus untouched, and where to put effort next.
• Every build report ends with the techniques used (from Affirmology_TechniquesLibrary_v1.md), so testing and feedback reveal which neuroscience and meditation tools actually land for listeners. Proof in the pudding.

Tracking and library

• Every render is saved with its inputs, the recipe/version used, the script, the audio, and its feedback and status (kept, dropped, pushed).
• Versions are tracked per project so we can compare and improve.
• Storage is readable by both Jeff's and Sol's Claude: a Google Sheet (one row per render: category, concept, version, rating, status, links) plus the files. "Track everything."

Quality and accuracy

• Log every render and flag chart-accuracy problems (the Manifesting Generator bug is the model case).
• Build a known-chart test harness: gather public charts with known, published Human Design and astrology data, run our engine against them, and score the matches, so we catch and fix engine errors continuously.

Corpus (how the content is grounded)

• Today: script generation is grounded in our corpus (real A/B-tier interpretation snippets for the chart's elements) and the LLM fills gaps from its own knowledge. There is no live web search in generation.
• Next level (needs more API credit): deepen the corpus, especially transits and moon omens, and keep adding new, better data within reach. Current size about 12,905 records across 6 traditions.

Techniques and methods layer (connect the whole trove)

The Studio engine must not generate in isolation. It draws on two knowledge layers: 1. The chart layer (already wired): the 6 chart traditions in the corpus, grounding content in the person's unique chart. 2. A NEW techniques and methods layer: the neuroscience and meditation-technique knowledge we have already researched, structured into the corpus as its own "techniques" tradition so the engine can pull the right induction and method per the audio's structure. Sources we already have to ingest: Affirmology_Neuroscience_AudioEfficacy_Research_v1.md, Affirmology_PredictiveBrain_IdentityChange_Research_v1.md, Affirmology_DepthPsychology_JungianIntegration_Research_v1.md, Affirmology_TeachingAudio_TransitAware_Research_v1.md, Affirmology_GroupAudio_Research_v1.md, Affirmology_CrisisJourney_Research_v1.md, Affirmology_PlantMedicineJourney_Research_v1.md, Affirmology_AudioMastering_Production_Research_v1.md, plus the agent design in Affirmology_SubconsciousOS_AgenticArchitecture_v1.md and Affirmology_BackendArchitecture_v1.md. Jeff will also upload his own heart-coherence technique to add (and to voice in Charlotte). Building and populating this techniques store is its own task and needs API credit, like the rest of the corpus work. It runs from the laptop over Tailscale now; the Mac mini takes it over in a few days.

The generation step the Studio uses: chart layer + techniques layer + the structure directive (the concept) feed script_generator.generate_script(chart, ..., concept=...), which now accepts a structure directive, plus a team-facing build report.

Audio

• v1: voice only (ElevenLabs Charlotte).
• Later: one licensed long ambient bed (about 45 minutes, for long Gene Keys journeys) from a commercial library; the pipeline auto-trims and fades it to each audio's exact length.

Build stages

• v1 (now): chart-driven Studio. The gated page, a person/chart selector, category tabs, the high-level concept and reflect-back flow, a script grounded in that person's chart, Charlotte voice render, listen plus script plus report, the comfort vote and feedback, and the library. Two logins.
• v2: chart mode plus the full report with blueprint-per-section; push-to-a-real-person by birth data; Google Sheet tracking; Sol's shared Drive folder and morning sync.
• v3: Colin and inner-circle access; the known-chart accuracy test harness; the corpus deepening pass; the licensed audio bed.

Open decisions (defaults proposed)

1. Gating: Cloudflare Access (Zero Trust) with email logins for Sol and Jeff (recommended, scales to Colin), versus a simple shared password. Default: Cloudflare Access.
2. Tracking store: Google Sheet plus files (recommended), versus local files only. Default: Sheet plus files.
3. v1 scope: chart-driven. A person selector (Sol, Jeff, or new birth data) plus a structure (category + high-level concept direction), producing a script grounded in that person's chart, voiced by Charlotte, with the report. Voice only, no music yet. Default: yes.