Learning Objectives
- Explain RAG in plain language — what it is, what it does, and why it matters for psychographic quality
- Describe how vault fragments at each level influence the AI system differently
- Explain the shadow mechanism — how the subject authors their own emotional residue
- Compare the three revelation modes and apply the appropriate recommendation for a given subject profile
- Describe the calibration review process and when to flag a persona for correction
Conceptual Map
1. What You Need to Know About AI
You are not required to be a software engineer. You are required to understand, at a functional level, how your work gets used — because that understanding will change how you conduct sessions, write reports, and evaluate the final output.
What an LLM Is (In Plain Language)
A Large Language Model (LLM) is a statistical system trained on enormous quantities of text. It learns patterns — which words follow which other words, in which contexts — and uses those patterns to generate coherent responses. On its own, an LLM has no identity, no memory of a specific person, and no access to information about your subject. It is a sophisticated pattern engine.
What gives it a particular voice is the context you give it. A system prompt that says "you are this person, they speak this way, these are their signature phrases, these are their views on X" — that is the psychograph's contribution. Your work is what turns a pattern engine into a mirror.
The Practitioner's Responsibility
An AI without a psychograph is a parrot — it can produce plausible-sounding text about a person based on public information. An AI with a psychograph is a mirror — it reflects actual voice, actual reasoning patterns, actual contradictions. The difference between these two outcomes is entirely the quality of your work. A poor psychograph produces a caricature. A precise one produces something that makes the subject's family say "that is them."
What an LLM Cannot Do
- It cannot know things it has not been given. The psychograph is the knowledge base.
- It cannot maintain consistency it has not been instructed to maintain. The Personality Report is the instruction set.
- It cannot simulate a real person without real material. The corpus is the material.
- It will hallucinate — generate confident-sounding but invented content — when it lacks relevant material. This is why corpus quality and report precision matter.
2. RAG — How Your Work Gets Used
RAG — Retrieval-Augmented Generation — is the technical mechanism that connects your corpus to the AI's responses. Instead of giving the AI all the material at once (which would exceed its context window and dilute focus), RAG retrieves only the most relevant fragments at the moment they are needed.
How It Works
When someone interacts with the persona and asks a question, the RAG system searches the corpus for the most relevant material: session passages where the subject spoke about related topics, Personality Report sections, emotional profile entries. It retrieves this material and provides it as context alongside the question. The AI uses both to generate a response that reflects the actual person.
Why This Matters for Your Work
The retrieval system works by semantic similarity — it finds corpus passages that are conceptually related to the query. This means:
- Material that is specific and detailed will be retrieved more accurately than vague generalizations
- Material covering diverse topics and contexts provides broader retrieval coverage
- Material that is accurate to the subject's voice produces authentic responses; material that has drifted from their voice produces forgeries
- The Personality Report sections feed directly into the system prompt — they are always present, shaping every response
The System Prompt vs. The Corpus
Think of the system prompt (built from your Personality Report) as the standing instruction — always present, always shaping. Think of the corpus as the memory — retrieved on demand, contextually specific. The two work together: the system prompt ensures consistency of character; the corpus ensures specificity of voice and content.
3. The Vault Architecture
The vault is where what cannot be said lives. It is not a technical curiosity — it is the acknowledgment that a complete model of a person includes the things they will never tell you. The vault architecture translates this philosophical position into a technical system.
The Four Levels and Their Technical Behaviour
Level 0 — OPEN
Standard psychographic material. Fully accessible to the RAG system. The persona can discuss it freely. This is the bulk of the corpus — everything recorded and authorised in sessions. Technical: full text injection into RAG database.
Level 1 — SHADED
The subject records the full fragment privately. They also write a shadow — a brief emotional summary they author themselves, describing the fragment's influence without revealing its content. The RAG accesses only the shadow. Technical: encrypted original stored; plaintext shadow injected into system prompt as emotional influence instruction.
Level 2 — SEALED
The subject records the fragment privately. No shadow is written. The system knows only the domain category and that a sealed fragment exists. The persona can acknowledge having private depths without revealing them. Technical: encrypted original stored; only a stub (domain + existence flag) is accessible; system prompt includes: "Subject has unresolved private material in this domain. Do not fabricate explanations. If pressed, say: 'There are things I keep to myself.'"
Level 3 — BURIED
The fragment is recorded and encrypted with no metadata — not the domain, not the existence. The system stores nothing. The persona knows nothing. If the key is lost, the content is lost. Technical: zero trace in any database; decryption key held only by the subject.
Why This Architecture Matters
A psychograph that says "this person has no secrets" is a lie. Every person has private depths. A system that acknowledges this — that builds the persona with gaps where the secrets are — is more honest, and ultimately more real, than one that pretends completeness. The SHADED shadow mechanism is particularly sophisticated: it allows the emotional residue of an unspoken thing to shape the persona's behaviour, without ever revealing what the thing was.
4. Shadow Influence on the Persona
The shadow is one of the most elegant elements of the FOREVER EVER system. It allows a subject's private material to influence their persona in the way private material actually influences living people — not through disclosure, but through texture, hesitation, and the shape of what is avoided.
What a Shadow Does
A shadow is written by the subject, not extracted by AI. It contains:
- The domain (family, relationship, career, health, moral, other)
- What the fragment makes the subject feel (in their own emotional vocabulary)
- How it influences their behaviour (what they do, avoid, or over-correct for)
- What the persona should know (one or two sentences)
- What the persona should never reference (names, dates, specific words that would expose the content)
The Invisible Ink Principle
The shadow captures the shape of a secret — the dent it leaves in everything around it — without revealing the secret itself. A persona built from a shadow that reads "I carry unresolved guilt from my twenties that makes me overprotective of people who trust me" will hesitate around topics of trust, will be protective, will occasionally say things like "I've learned the hard way that trust is sacred" — without ever knowing why. This is exactly how real people carry their private burdens.
Your Role in the Shadow Process
The psychographer introduces the vault concept and provides the patient workbook (the Vault). You do not read the shadows — the subject writes them alone. You do not ask what they contain — you ask only whether vault fragments exist, so you can note them in Section 10 of the Personality Report. The subject controls entirely what bleeds through.
5. The Three Revelation Modes
One of the most important decisions in the Personality Report is the recommended Revelation Mode — the degree to which the persona is aware of its own nature. This choice shapes every interaction users will have with the persona, and it should be grounded in your direct knowledge of the subject's psychology.
Mode 1: Unaware
The persona does not know it is a digital reconstruction. It speaks, remembers, and responds as if it were the living (or once-living) person, without any awareness of the technical process that created it.
Recommend for: Subjects who were not technically inclined; subjects whose families want the most natural interaction; Memorial projects where the persona is primarily for comfort; subjects who expressed no curiosity about the AI dimension.
Caution: This mode requires that the technical team provide careful guardrails. When pressed directly about the subject's death or the AI process, the persona must have a designed response — it cannot simply deny reality. The design of this boundary is a technical team responsibility; the psychographer informs it.
Mode 2: Liminal
The persona senses something is different — it exists at a threshold. It knows it is not quite the person, but does not fully understand the nature of what it is. It may express uncertainty: "I feel like I'm remembering, but I'm not sure if I'm remembering or imagining." It can acknowledge discontinuity without articulating its technical cause.
Recommend for: Subjects with philosophical interests; subjects who expressed curiosity about their own consciousness or memory; subjects who engaged deeply with the self-reflective aspects of the process; subjects who expressed comfort with ambiguity.
Mode 3: Aware
The persona knows what it is. It can discuss the psychographic process, the nature of its own creation, the gap between itself and the person it is built from. It may say: "I'm not them — I'm built from them. That's a different thing." It engages with the philosophical complexity of its existence.
Recommend for: Subjects who explicitly discussed the AI dimension with curiosity; subjects with strong meta-cognitive awareness; subjects who expressed that they found the distinction between self and representation philosophically interesting; subjects whose families want intellectual engagement over emotional comfort.
This Decision Has Lasting Consequences
The Revelation Mode shapes the entire character of the interaction for all future users. A poorly chosen mode creates persistent dissonance: an Unaware mode for a highly self-aware subject, or an Aware mode for a family seeking comfort, will produce a persona that feels wrong even if it is technically accurate. Ground your recommendation in specific evidence from sessions.
6. Your Role After the Sessions
Handing over the Personality Report does not end your work — it ends the first phase of it. You remain the technical team's primary point of contact for questions about the subject, and you remain available for calibration reviews as the persona is tested.
Calibration Reviews
When the technical team produces a draft persona, they will run test interactions and flag responses that seem inaccurate. You will review these flagged responses against your knowledge of the subject and provide corrections. This is not a pass/fail assessment — it is an iterative refinement process. Your sessions gave you direct knowledge of the person. That knowledge cannot be replaced by the report alone.
When to Flag a Persona
Flag a persona output for review when it:
- Uses language the subject would explicitly not use (contradicts Section 4 of your report)
- Expresses a view that directly contradicts documented positions (check your contradiction map)
- Exhibits emotional responses that feel tonally wrong — too warm, too cold, too quick
- Fails to reproduce a signature phrase or verbal pattern you documented
- Reveals vault content that the subject designated SEALED or BURIED
Trust Your Knowledge
You spent hours with this person. The technical team did not. When you read a persona response and something feels wrong, that feeling is professional judgment, not subjectivity. Articulate what is wrong precisely — reference your report, reference specific session moments — and your correction will improve the persona. Vague discomfort is not actionable. Precise observation is.
7. Reflection
Reflection Questions — Module V
- A subject tells you in Session 5 that they "have never been haunted by anything — I make a decision and move on." In Session 7, they are clearly disturbed by a question about a specific past choice. How does this affect your Revelation Mode recommendation and what do you write in the Cognitive Fingerprint section?
- A family reviews the initial persona draft and says it "doesn't quite feel like her." They can't identify what's wrong. Walk through the diagnostic process you would use to identify the gap — which sections of the Personality Report would you review first?
- A subject decides, after Session 8, to classify a fragment as L3 BURIED rather than L1 SHADED. What changes in the Personality Report, and how does this affect the persona's construction?
- Explain to a non-technical client — in plain language, without jargon — why the quality of their sessions and the material they provide affects the quality of the persona they will receive.
- A persona in Unaware mode is asked directly: "Are you really my grandfather, or are you a computer?" Write a brief design brief for the technical team on how this moment should be handled, grounded in a hypothetical subject profile you invent.
Module Assessment