Home / Ops / More in this area

Affirmology Audio Engine: Layered Understanding Architecture (v1)

Updated Jul 04, 2026 · Affirmology_AudioEngine_LayeredArchitecture_v1.md

Affirmology Audio Engine: Layered Understanding Architecture (v1)

Written 2026-07-04 with Jeff. How to use the council efficiently: do the expensive understanding ONCE, store it, grow it, and shape many cheap audios from it. No em dashes.

The core principle

The council's product is NOT an audio. It is a durable, structured UNDERSTANDING of the person. Audios are cheap shapings of that understanding. We never re-derive what we already learned; each new pull only does the necessary new work.

Expensive once. Cheap forever after.

The layers (each feeds the next)

• Layer 0 - Raw Blueprint. The chart across all systems (astrology, HD, Gene Keys, numerology, Vedic), transits, progressions. Deterministic, computed by the engine, near free. We already store this in the Profile.
• Layer 1 - The Understanding. The council's INTERPRETED findings, computed once (at the depth the person's engagement justifies). Not raw chart data: the synthesized insight. Structured and reusable:
• the salient findings ranked (salience scoring: Loud, Activated, Juicy, Wanted)
• the cross-system convergences ("four systems all say this")
• the aha moves identified per the registry (contradiction, reframe, wound, gift, timing, family, body)
• the wound, the gift, the gap, the North Node direction
• the story spine (the hero's tension and becoming) This is where the council spends money. It happens ONCE per person and updates only when new information arrives.
• Layer 2 - Context. Two sources: EXPLICIT (the user states it: "confidence," "my new business," "wealth") and ASSUMED (what the chart is shouting about, the intake-mismatch move). Context selects which Layer-1 findings to lead with and which placements to weight (a wealth reading weights 2nd house, Lakshmi/Vedic wealth signals, the money themes; a confidence audio weights something else). Current progression/transits ride here too (the in-season layer).
• Layer 3 - Learned. What we gather over time: probing questions, chatbot conversation, voice-note feedback, what landed and what did not. This grows Layer 1 (append, never redo) and tunes future audios. This is the "context is the missing key" fix: a human reader goes deeper by asking "what part of that lands," so we ask.
• Layer 4 - Presentation. Given Layers 1-3, choose the audio TYPE, the storytelling techniques, the transformational techniques, and the person's JOURNEY/MAP across several audios. This is structure selection + craft standard (the Hero's Story spec).

How audio types share the work

A wealth reading, a confidence audio, a Mirror, a bedtime track, an Origin Soul Song all draw from the SAME Layer 1 understanding. They differ only in Layer 2 (which facets, which weighting) and Layer 4 (type, craft, journey position). So the council's deep work is amortized across every audio a person ever gets. Categories are not separate builds; they are different SELECTIONS and PRESENTATIONS of one understanding.

The cost model (answers "chatbot cheap or expensive?")

• Layer 0 (chart math): near free.
• Layer 1 (the council understanding): the expensive event. Happens ONCE per person (or on a real update), not per audio.
• Chatbot Q&A in the app: CHEAP. It reads the stored understanding (retrieval plus a light model) and answers. It does NOT re-run the council. Cheap all day.
• Triggering a custom audio: the paid event, and even then it is a cheap SHAPING of Layer 1 (one or two calls), not a fresh council run, plus the voice render. So yes: chatbot talk is cheap; expense concentrates at custom-audio generation, and we keep even that cheap by shaping from the stored understanding.

The two investment tiers (Jeff's Layer 1 / Layer 2 plan)

• Tier 1 (light): the Origin Soul Song and its recognition. A shallow, cheap understanding: enough salient identity to make the song land. Given to everyone. Low spend.
• Tier 2 (deep): the full council Understanding. Built when a person goes deeper (beta, engaged, converts, or is a high-value connector). This is the expensive compute-once pass that then powers many cheap audios. Justified because they are invested. Jeff will justify a couple of extra gifts for the initial Origin-takers, because they are the highest-likelihood beta testers, connectors, and mission carriers, so investing in knowing them is worth it.

The probing-questions layer (build this)

A layer that asks purposeful questions (in the app chatbot, intake, or voice notes) whose purpose is to STORE deeper information and fill gaps in the understanding. Cheap to run (a light model asks and files answers into Layer 3), high payoff (every audio after is more customized). This is likely part of the missing key from the reading postmortem.

Does the current Profile serve this? (Jeff's question)

Partly. Today the Profile holds Layer 0 (chart data) plus ONE narrative deep read. That is a good foundation but not yet the reusable Understanding. To serve many audio types efficiently it needs to evolve: 1. Store Layer 1 as STRUCTURE, not just prose. Salient findings ranked, convergences, aha moves, wound/gift/gap, story spine, as fields the shaper can select from. Right now the "deep read" is one blob; we want queryable insight. 2. Add a Context store (Layer 2) and a Learned store (Layer 3) that grow over time, keyed to the person. 3. Version and append. New learnings update the understanding without recomputing it. The chart capture is already solid. The tweak is turning the single deep read into a structured, growing Understanding object.

Non-council levers (cost)

• Voice: Fish keeps readings cheap; ElevenLabs is the song-voice cost. Pursue the ElevenLabs free year, that is the real voice-cost lever.
• Model tiering: the council uses the strong model; the chatbot, the shaper, and the question-layer can use a cheaper model.

Open decisions for Jeff

1. Build the structured Layer 1 Understanding object now, or after the reading-craft rebuild lands?
2. What triggers Tier 2 (the deep understanding): a signup action, a second-audio request, a subscription, or manual for the first cohort?
3. Which few extra gifts for the Origin cohort, and at what depth?

Status 2026-07-04: DESIGN / discussion. Companion to Affirmology_HybridRenderArchitecture_v1.md (where the compute runs) and Affirmology_ReadingCraft_HeroStory_Synthesis_v1.md (how a reading sounds).