The Self-Understanding Inventory

TL;DR

Three deep-research runs were asked for one thing: the complete inventory of ways a person can be understood and helped toward self-actualization, tuned to the profile in the brief (a 27-year-old Korean-American creative-technologist with a charismatic-church history, abandoned projects, "downloads", disrupted sleep, and a habit of turning insight into infrastructure). All three came back. ChatGPT produced the backbone: 732 graded items across the twelve layers the brief asked for. Claude produced a tighter 440-item inventory with primary-source numbers that all checked out, and the one argument the others lack: that the way the inquiry is completed (spoken, one question at a time, with a witness, on a clock) matters more than what is in it. Gemini produced a 5,800-word overview that ignored the brief's structure, mislabelled its evidence, and repeated one claim that has been overturned.

After checking the claims that matter, the inventory collapses to a small ordered core. First, a written safety plan, then a psychiatric differential (bipolar spectrum, sleep, ADHD, autism, trauma) run by a clinician, because every other lens in this document changes meaning depending on that answer. Second, eight weeks of cheap, boring self-monitoring: sleep, caffeine, mood, a weekly hypomania rating, and a two-column log of "downloads" with a 72-hour outcome. Third, evidence you already own and have never read as evidence: the project graveyard, the git history, the school records, and three or four people who watched you and will answer "what did you see me do?" Fourth, a handful of free, validated instruments chosen for the specific hypotheses in the brief (worth-equals-output, avoidance, attachment, reward regulation, sleep-state), not the full battery of 111. Only then the narrative and mythic tools, the spiritual traditions, the biographies, and the typologies, most of which are lenses rather than measurements and are labelled as such below. Throughout, Claude's completion rule, which the brief's own history makes non-negotiable: every question answered aloud, one at a time, ideally with an AI asking follow-ups or a person on the call; one action taken for every hour spent organising; and a kill date for the document itself if ninety days pass with insight and no behaviour change.

Cut entirely or heavily downgraded after fact-checking: STAR*D's "67% remission", the MDQ as a bipolar-II screen, RAADS-R as a stand-alone autism test, neurofeedback, dopamine detox, MBTI, Human Design, astrology and tarot even as "projective mirrors", DIY psychedelics, and any long silent retreat before the differential is characterised.

Do these this week

1. Write a Stanley-Brown safety plan and put it where you can find itSix boxes: warning signs, my own coping, people and places that distract, people I can ask for help, professionals and 988, making the environment safer. The emergency-department trial cut suicidal behaviour by 45% over six months. It costs nothing and every model that took the brief seriously put it first.30 minutes · free · template
2. Book the psychiatric assessment and bring the right packetAsk for a structured differential (bipolar spectrum, ADHD, autism, PTSD, sleep). Bring one page: symptom timeline, sleep pattern, family history, substances, three specific questions. ChatGPT's line is right: do not bring a 500-page AI dossier. Ask whether CYP2C19 pharmacogenomic testing is available; roughly 18% of East Asians are poor metabolisers of escitalopram and sertraline, and the 2023 CPIC guideline halves the starting dose for them.1 phone call + 1 page · Medicaid-covered
3. Start the eight-week log: sleep, caffeine, mood, and a weekly Altman self-rating of maniaPaper or a single spreadsheet row per day. This is the evidence a clinician needs to separate a hypomanic pattern from an ADHD pattern from a burnout pattern, and no questionnaire replaces it.2 minutes a day · free · ASRM
4. Open the two-column "download" logEvery strong impression or "word": date, sleep the night before, caffeine, fasting, isolation, substances, emotional state, urgency, what it asked you to do, then a 72-hour outcome and two assessments, one theological (your own tradition's tests of a prophecy) and one clinical (state, sleep, impulsivity). ChatGPT rated this the single highest-fit tool in the inventory. I agree, with one caveat: it is a designed instrument, not a validated one, so its value is the pattern after twenty entries, not any single row.5 minutes per entry · free
5. Adopt the red-tier rule now, before any insight arrivesGiving away savings or equity, quitting income, moving, a religious vow, a psychedelic or fasting or sleep-deprivation retreat: none of these happen without multiple normal nights of sleep, a written pre-mortem, one adviser with no stake, and a pre-written cooling-off period. Amber (any transfer over $500, a 30-day obligation, a public announcement) waits one sleep-normalised day and gets a logged prediction.10 minutes to write the card · free
6. Take four free instruments, not fortyPHQ-9 and GAD-7 (baseline), HCL-32 (bipolar-II screen, once), Contingencies of Self-Worth and the Clinical Perfectionism Questionnaire (the worth-equals-output hypothesis), ECR-RS (attachment, per relationship). Print results and date them; they go in the packet for item 2.45 minutes · free · links in Resources
7. Ask one person one questionThe sister, or the most-repeated collaborator: "What did you see me do, in the last five years, that I probably don't see?" Async, in writing, no rebuttal. This is the cheapest high-yield item in the whole inventory. No model could point to a trial, but Claude points to the reason none is needed: other people are reliably more accurate than you about how you come across, and self-report cannot reach that at all (Vazire 2010, the self–other knowledge asymmetry).One message · free
8. Start pain reprocessing for the back, and ask for a psychodermatology referral for the skinThe brief lists six years of back pain with clean imaging and fourteen years of eczema. Claude ranks pain reprocessing therapy fifth of its thirty because the trial result is unusually clean: 66% pain-free or nearly so after four weeks, against 20% on open-label placebo and 10% on usual care, holding at one year. The self-help versions (Gordon's The Way Out, the Curable app) are the evidence-adjacent form; a medical work-up is the precondition, not a substitute.A $15 book or a $10/month app · Medicaid for the referral

How this was made

ChatGPT (deep research, 28,957 words, 240 unique links) is the backbone of this document. It followed the brief's twelve layers, graded every item A to F for evidence and fit, tailored the fit ratings to the profile, and closed with a minimum-viable-30, a sequencing plan, and a list of what it could not verify. Its citations resolve mostly to Wikipedia, NICE, SAMHSA, publisher pages and PubMed; I re-checked the ones that carry weight. Its main weaknesses: a handful of confident evidence claims that are older or thinner than they sound (flagged inline), and a few "designed" tools presented beside validated ones without always saying which is which.

Gemini (5,798 words, 58 links) ignored the A to L structure, returned nine generic sections and roughly 35 items, labelled personality inventories "Strong RCT", and repeated the original STAR*D remission figure that a 2023 re-analysis overturned. It contributed three things the others did not: a medical rule-out panel, an explicit "insight trap" warning, and a mythopoetic case formulation that is its own invention. Everything it listed is accounted for in the coverage below; it is simply not an independent second opinion on most of it.

Claude: (deep research on claude.ai, 8,553 words, 22 links, ≈440 items) is the second full leg. It followed the twelve layers, tagged what the earlier dossier passes already covered, and cited primary sources for every number it used; all 22 of its links resolved and every effect size I re-checked was accurate to the source. Its distinctive contribution is not an item but an argument: that the completion method (voice-first, interviewed by an AI, witnessed, timeboxed) is the binding constraint and belongs first, backed by the intention–behaviour gap and the tracker-abandonment literature. It is thinner than ChatGPT on instruments and biographies, warmer on IFS than the evidence warrants, and uses the older imagery-rescripting numbers. The report was not extracted by the browser agent: the launch-time URL had died, and Isaac located the finished run and pasted it; the file on disk is that paste, identical to the artifact export in research result/.

What I did with the raw material. Built a coverage ledger of every distinct item across the reports, clustered it by the brief's layers, then fact-checked 43 claims that would change what you do (screener accuracy, treatment effect sizes, risk claims, costs), twelve of them from the Claude report. 230 unique URLs from the raw reports were fetched: 190 returned normally, 26 blocked automated access (NICE, CDC, Medicaid, JAMA, ResearchGate), 4 were dead, and 3 hosts did not resolve; all 22 of Claude's links resolved; dead ones are named in Open questions. Nothing was silently dropped: items I judged useless are in Obsolete and overrated or Everything I nearly cut, with the reason. The raw reports sit beside this file as raw-gemini.md, raw-chatgpt.md, and raw-claude.md.

I need exhaustive deep research that produces one thing: **the master inventory of every technique, question set, questionnaire, validated instrument, typology, archetype system, psychology, therapy modality, spiritual practice, wisdom tradition, philosophy, self-help method that actually works, biographical parallel, and body of literature that humanity has produced for understanding a single human being as fully and accurately as possible, and for helping that person self-actualize.** Not a summary. Not a top-20. The long tail, organized, with links, with the actual questions where they are public, and with an honest verdict on each item.

## What I am building, and why

I am assembling a personal "megadocument of self-understanding." Your inventory is its table of contents. I will take every item you list, go and do it (answer the questions, take the instrument, run the exercise, interview the people, read the source), and file the results into one document. That document will then be read by me and fed to AI models as context, so that they and I can see, as completely and accurately as possible: who I am, my strengths and weaknesses, my philosophies, my tendencies and cognitive patterns, my desires and what I want out of life, what makes me happy, what I love and hate, who I love, what other people think of me, my psychological profile, my spiritual profile, my recurring themes and cycles, my archetypes, my life history, my traumas, my best moments, why I have depression and anxiety, why I go through the same cycles and thought patterns, how to eradicate the limiting beliefs and schemas that drive them, how to find my personal myth and not lose sight of it, and how to move forward toward self-actualization, my goals, happiness, creativity, success, and a fully lived life.

So every item you include must be **actionable by one person with a laptop, a phone, a Medicaid card, and modest money**: something I can answer, take, do, read, order, request, or ask someone. If it needs a practitioner, say which kind and how to find one.

## Who I am (so you can tailor, not just retrieve)

- 27-year-old Korean-American man. Born in the US to Korean missionary parents; raised in Canada, India (my father ran a hospital there), Korea, and the rural American South; ten relocations before 18; sent to the US alone with my sister at 10. Third-culture kid and missionary kid by every definition.
- Filmmaker, musician (guitar, piano, voice, songwriting), and self-taught software engineer. Won a national high-school film award at 17; dropped out of UCLA film for money; five years in Korea; founder of two apps; freelance developer; one shipping product. Visual and intuitive thinker; I think by talking out loud and have five years of voice memos and ~440,000 words of notes to prove it.
- Core wound, in my own words: paternal abandonment around age 12 (my father left; ~13 years of silence; reunited in 2024), a mother whose support arrived after awards, racial othering as the only Asian kid, and a fear-saturated charismatic Christianity (demons, deliverance, prophecy) installed in childhood.
- The engine I have diagrammed myself: worth fused with output → achieve to prove worth → isolate and grind → burnout → collapse → "was I delusional?" → new vision. Career identity flips every few days. Highs from films, live music, caffeine, sleep deprivation, validation; lows from consuming others' success, coding for money, idleness, loneliness.
- Documented cycles: a "vision cycle" (a received, cosmic-stakes idea → recruit others → grind alone → crash), a "pivot loop," and "system-building as escape" (when a task exposes my limits I design a tool instead of doing the reps; I call it sharpening the axe instead of chopping the tree). I have built and abandoned six self-questioning systems; the last workbook had 160 questions and I answered three, with junk, in ten minutes. **Any inventory that does not come with a method for actually completing it will fail the way the others did.**
- Clinical picture: panic disorder (first attack Feb 2025; recovery time cut from two weeks to one hour by a 10-day Vipassana), chronic depression and anxiety, insomnia patterns, eczema over up to 40% of my body for 14 years, six years of back pain with clean imaging. Therapists have floated bipolar-spectrum, ADHD, autism/masking, "narcissistic victimhood"; schizophrenia ruled out. **No formal assessment has ever happened; no validated instrument has ever been scored.** Medication offered once, declined. About 25 psychedelic sessions in 2024 including one catastrophic trip; nothing since. Nine dated episodes of suicidal ideation across five years, none told to anyone, no safety plan.
- Spiritually: inherited charismatic Christianity → agnostic → psychedelic mysticism → spiritual-warfare Christianity → systematic deconstruction (read the cult literature) → full re-conversion in 2026 via prophecy, baptism, deliverance and fasting. I now live inside a 100+ person charismatic movement led by a young influencer, which I largely funded (gave away my entire $30k savings in April 2026), technically run, and hold 0% equity in, while keeping a private file of doubts about it. I have "downloads" (received visions) whose recipe I have identified myself: caffeine + sleep deprivation + isolation. I do not know how to tell a download from a hypomanic surge from a genuine leading, and I want the tools every tradition has for exactly that question.
- Relational: chronic loneliness is my master variable. I need a witness in the room to decide anything. I cannot say "no" or "it's over" to a face; I exit by fading or by a lie, then pay for it in panic. I idealize at a distance and deflate on contact. I have never grieved anything; every loss becomes a lesson within a day. I have never written one sentence about whether I want a partner, marriage, children, or where I want to grow old. Labels that stuck on me were all delivered by a person in the room; every label from a written test I discarded within a day.
- Cognitive: fast immersion learner under someone else's deadline (employable coder in six months), vivid imagination that "scares me sometimes," working memory collapses under social load, articulation is my named insecurity, insight decays overnight and I have built four tools to fight that instead of asking why.
- Money: swings between "money is sucking my soul" and "screw money," never a middle; ~$3.4k/month passive; no runway; gives extravagantly outward, nothing inward.
- What I am good at, evidenced: cinematic eye, real musical craft, rapid skill acquisition, systems thinking, unusual self-honesty in writing, warmth (people cry telling me they love me), physical courage against fear, sitting with people in crisis, teaching.

Two things already exist and you should assume them: (1) two long AI-built dossiers over my own notes and recordings, which are excellent at pattern extraction from my crisis writing and weak at everything that is not in my own writing; (2) one prior research pass whose coverage I list at the end as the baseline. **The value you add is everything beyond that baseline**, and the baseline must still appear in your inventory so the list is self-contained.

## The inventory, layer by layer

Produce the inventory in these layers. Inside each, go for completeness first, then depth. If you find 200 items, I want 200, grouped.

### A. Ways of knowing a person (the evidence types)
Every distinct method by which a human being can be known: self-report, structured self-interview, journaling protocols, voice-first methods, observation by others, structured third-party interviews, 360 and reflected-best-self methods, observer-rated instruments, projective methods (TAT, sentence completion, early recollections, dream series, drawings, sandplay), behavioral traces (calendar, messages, search history, location, spending, screen time, git history, listening history), objective records (medical, school, legal, genealogical, genetic), the body as evidence (sleep, HRV, chronotype, sensory profile, labs, pharmacogenomics), the creative output as evidence (lyrics, films, code, what got finished vs. abandoned), expert assessment (psychiatric, neuropsychological, psychodynamic formulation, spiritual direction), and AI-assisted analysis of a personal corpus. For each: what it reveals that the others cannot, how to do it, cost, the known bias of that source (journals over-sample distress; self-report over-reports symptoms in heavy introspectors; AI readings echo the prompt), and how to triangulate.

### B. Question sets (the actual questions)
Every canonical and lesser-known self-knowledge question set, with the questions verbatim where they are public and a link where they are not: life-story and autobiographical protocols (McAdams Life Story Interview, Adult Attachment Interview, Adlerian early recollections, Progoff Intensive Journal, Past/Future Authoring, life-review and reminiscence protocols, legacy-letter and ethical-will prompts), values and priorities sets, desire-eliciting sets (what I want, not whether I passed; I ask "am I…" 218 times and "what do I want" 6 times in my own archive, so I specifically need the questions that surface wants, longings, and a definition of "enough"), fear and shadow sets, relationship and love sets (36 Questions, Gottman inventories, partner-template exercises), money and vocation sets (Odyssey plans, workview/lifeview, ikigai with its corrections, Drucker, Christensen, kill criteria), death and legacy sets (Yalom's four concerns, five regrets, eulogy/obituary, deathbed and regret-minimization, Bhutanese and Stoic death practices), faith and discernment sets (creed-writing, Ignatian examen, Quaker clearness-committee queries, Naikan's three questions, Mussar cheshbon ha-nefesh), creativity and craft sets, daily-life and "ideal ordinary day" sets, best-moments and peak-experience protocols (Maslow's peak experiences, flow interviews, "when were you at your best" stories), childhood and family-of-origin sets, culture and identity sets, humor/play/rest sets, body sets, and the "case against this document" sets (Barnum checks, disconfirmation questions, what would change my mind). Include the question sets used inside therapies (schema-therapy interviews, IFS 6 Fs, ACT values and matrix, DBT chain analysis, CBT downward arrow and core-belief worksheets, Coherence Therapy discovery questions, narrative-therapy re-authoring questions, Transactional Analysis script questionnaires, psychosynthesis subpersonality interviews, Hakomi and Focusing prompts, Gestalt empty-chair scripts, Byron Katie's four questions, motivational-interviewing decisional balance), and the questions a good clinician asks in a first intake and in a case formulation. Flag which sets are designed to be answered aloud or with a witness, because that is how I actually complete things.

### C. Validated instruments and profiles
Every self-report or observer instrument that would tell me something true about temperament, personality facets, cognition, emotion, motivation, attachment, trauma, mood, neurodevelopment, sensory profile, chronotype, creativity, spirituality, religious struggle, scrupulosity, spiritual bypass, mystical experience, shame, self-compassion, perfectionism, procrastination, impostor feelings, narcissism (grandiose and vulnerable), rejection sensitivity, loneliness, social anxiety, panic, rumination, worry, alexithymia, interoception, absorption, schizotypy, hypomanic personality, affective temperament, impulsivity, sensation seeking, boredom proneness, time perspective, future-self continuity, self-concept clarity, identity status, ego development, meaning in life, wellbeing, flow, values, character strengths, money scripts, career interests, and relationship style. For each: what it measures, number of items and time, norms, reliability and validity in one line, the free or licensed source with a link, the known false-positive traps (screeners for ADHD, bipolar, autism, C-PTSD and rejection sensitivity overlap heavily), whether it should be taken twice in two mood states, and what a re-test interval should be. Also cover what a formal assessment involves (psychiatric evaluation, neuropsychological testing, WAIS indices, psychodynamic formulation, the 5 Ps), how to obtain one on US Medicaid, what to bring, and what an assessor cannot tell me.

### D. Typologies and archetype systems, with the pitfalls and the way out of each type
Every typology and archetype system with any following, from the most validated to the frankly unvalidated, each labeled honestly: Big Five and HEXACO profiles, Cloninger temperament, Enneagram (all the major schools, wings, instincts, levels, and the disintegration/integration paths), MBTI and Jungian functions, Socionics, Keirsey, DISC, CliftonStrengths, Kolbe, Hogan, Holland codes, Schwartz values types, Spiral Dynamics and integral stages, Kegan and Loevinger stages, Fowler faith stages, Erikson and Levinson life stages, Jungian archetypes (puer/senex, shadow, anima, wounded healer, orphan, trickster, magician, and the rest), Pearson's twelve and the PMAI, Moore and Gillette's four and their boy-psychology shadows, Campbell's monomyth stages and where a person can be stuck, Rohr's two halves, Hollis's middle passage, Bly's father hunger, Hillman's acorn and daimon, Pete Walker's 4F trauma types, Horney's three moves, Kohut's three hungers, Winnicott's true/false self, Alice Miller's gifted child, Young's 18 schemas and schema modes, Transactional Analysis life scripts and drivers, attachment styles, Dabrowski's overexcitabilities, gifted-adult and multipotentialite profiles, HSP, the Korean and East Asian frames (han, jeong, nunchi, hyo, chemyon, hwa-byung, sinbyeong, the Confucian father, saju as a cultural mirror), Third Culture Kid and missionary-kid profiles, Asian-American identity development stages, biblical and mythic archetypes (Isaac and the Akedah, Jacob, David, Jonah, Elijah under the broom tree, the prodigal and both brothers, Job's friends), the monastic vices and their remedies (acedia, the gyrovague, the eight thoughts of Evagrius), the Buddhist hindrances and personality types, the Sufi nafs stages, the Yogic kleshas, Ayurvedic and Galenic temperaments, astrology and Human Design and numerology as projective mirrors with the evidence stated plainly. **For every type or archetype: its characteristic pitfall, its shadow, its typical life course if unaddressed, what the tradition or research says is the way out, and how I would test whether it fits me rather than assume it.**

### E. Psychologies, therapies, and self-help that actually works
Every school and modality, with the problem each targets, the mechanism, the evidence (effect sizes and the best meta-analysis or trial, with links), whether it can be self-administered from a workbook or needs a practitioner, the cost and how to find one on Medicaid or cheaply, the failure modes, and who it fits. Cover at least: psychodynamic, Jungian analysis, Adlerian, existential and logotherapy, humanistic (Rogers, Maslow), Gestalt, psychosynthesis, Transactional Analysis, positive psychology, CBT and its cognitive-distortion and core-belief work, REBT, behavioral activation, rumination-focused CBT, metacognitive therapy, ACT, DBT, MBCT, compassion-focused therapy, schema therapy, IFS, EMDR, brainspotting, somatic experiencing, sensorimotor, Hakomi, Focusing, Coherence Therapy and memory reconsolidation, AEDP, ISTDP, EFT, mentalization-based, interpersonal therapy, narrative therapy, solution-focused, motivational interviewing, exposure and ERP (including for scrupulosity and panic), CBT-I, IPSRT and social-rhythm therapy, pain reprocessing and emotional awareness therapy for mind-body pain, psychodermatology, Morita therapy and Naikan, group therapy and men's groups, 12-step and SMART and their evidence, peer support, coaching, expressive writing (Pennebaker) and structured journaling systems, art/music/drama therapy and psychodrama, family systems and genograms and constellations (labeled), the Hoffman Process and similar intensives (labeled), Landmark and large-group awareness trainings (labeled as high-control), psychedelic-assisted therapy (MDMA, psilocybin, ketamine; the actual trial results and the actual risks for someone with my history), TMS, neurofeedback, HRV biofeedback, lifestyle psychiatry (exercise, sleep, light, diet, nature, cold, heat, breathwork including holotropic and Wim Hof, with the evidence graded), nutritional psychiatry and the labs worth running, pharmacology as an option (what classes, what the East Asian pharmacogenomic considerations are), and the relapse-prevention, safety-planning, WRAP, psychiatric-advance-directive, and early-warning-signature tools that turn insight into a plan. Also the literature on **why insight does not change behavior** and what does (implementation intentions, commitment devices, body doubling, duration and stability, the one-long-relationship finding), because that is the actual gap in my life.

### F. Spiritual practices, wisdom traditions, and philosophies
For every major and minor tradition, what it says about knowing oneself, suffering, depression and anxiety, cycles, vocation, desire, death, and the false self; its actual practices with evidence and cautions; and its discernment tools for telling a genuine leading from a compulsion or a delusion. Cover at least: Christian contemplative and monastic (Ignatian discernment and examen, Benedictine stability and rule of life, the Desert Fathers, John of the Cross's dark night vs. depression, Teresa, Merton, Nouwen, Willard, centering prayer, lectio, the tests for prophets, the father heart of God), the Reformed and evangelical formation literature, the charismatic world I come from and its documented failure modes, Korean Christianity and shamanism, Buddhist (Vipassana, Zen, Tibetan lojong and dream yoga, the second arrow, the hindrances, metta), Hindu and Yogic (Patanjali, Vedanta, Ramana's self-inquiry), Taoist, Confucian self-cultivation, Stoic (dichotomy of control, evening review, premeditatio, memento mori), Epicurean, Socratic and Delphic, Aristotelian eudaimonia and virtue, Jewish Mussar and cheshbon ha-nefesh and ethical wills, Kabbalah, Sufi (nafs, Rumi, Ibn Arabi), Quaker clearness committees and discernment, Ubuntu and indigenous rites of passage and vision quests (Plotkin, Meade, School of Lost Borders), Japanese (ikigai, kaizen, wabi-sabi, Naikan, Morita), Hawaiian ho'oponopono, existentialists (Kierkegaard on the father wound and despair, Nietzsche on ressentiment and amor fati, Frankl, Yalom, Becker, Camus), Simone Weil, William James on the sick soul, Jung on individuation and the personal myth, Hillman, Girard on mimetic desire and scapegoating, Byung-Chul Han on the achievement subject, secular Buddhism and modern contemplatives, and the modern "spiritual but not religious" toolset. Include religious trauma, spiritual abuse, spiritual bypassing, scrupulosity, the pre/trans fallacy, spiritual emergency vs. psychosis, and every framework for auditing a high-control group (BITE, bounded choice, Stein's attachment model, Singer, Lifton) with the honest note that I am currently inside one such community and want to judge it with instruments rather than moods.

### G. Personal myth, narrative identity, and vocation
Everything on finding one's personal myth and not losing sight of it: Jung's "what myth am I living," Campbell, Feinstein and Krippner's personal mythology method, Plotkin's soulcraft and the mythopoetic men's work, Meade's genius myth, Hillman's Soul's Code, Stephen Cope's dharma work, McAdams's narrative identity and redemption/contamination coding, narrative therapy's re-authoring, Berne's life scripts and how they are rewritten, Progoff's steppingstones, dream series work (Jung, Hillman, Bosnak, projective dream groups, lucid dreaming), active imagination, the hero's journey as a diagnostic (which stage a person is stuck at and what unsticks it), rites of passage for adults who never had one, vocation and calling literature (Buechner, Palmer's Let Your Life Speak, Newport, Epstein, Godin's dip, Duke's kill criteria), and, critically, **the methods for keeping a personal myth in view when insight decays overnight**: anchors, rituals, cards, rules of life, review calendars, letters to future selves, state-dependent memory countermeasures, the "true in a month" test, and the research on future-self continuity.

### H. Changing the pattern: limiting beliefs, schemas, cycles, depression, anxiety
The mechanism-level literature on why the same cycle repeats and what breaks it: schema and mode work, memory reconsolidation and the juxtaposition experience, belief-change methods with and without evidence (Coherence Therapy, Lefkoe, NLP and hypnotherapy labeled honestly, The Work, Dickens process, cognitive restructuring, imagery rescripting, chairwork, rescripting the origin scene), rumination and worry mechanisms and treatments, the abstinence-violation effect and lapse protocols for behavioral addictions (porn, games, nicotine, caffeine, sugar) without the shame that makes them worse, mood-cycling management (social rhythms, sleep as leading indicator, the hypomania recipe, the Elijah protocol for the crash after the high), rejection sensitivity, the loneliness science and social baseline theory, grief work for losses never grieved (Worden's tasks, meaning reconstruction, TCK unresolved grief, the unsent letter), forgiveness protocols (REACH) and their limits, saying no and boundary scripts (DEAR MAN, Bohns), decision governance for people who decide in elevated states (decision journals, tiers, waiting periods, a personal board, pre-mortems, financial firewalls and giving caps, Ulysses contracts), and the honest evidence on self-help books and programs: which have trials, which are placebo, which are harmful.

### I. Comparable lives, great works, and reading
A method for matching a life to biographical parallels by mechanism rather than by fame, and then the parallels themselves: people with a missionary or pastor father, a father wound, a third-culture childhood, a Korean or Asian-American identity struggle, an early peak then obscurity, mood cycling plus creativity, psychedelic conversion, a radical giveaway of money to a movement or a leader, life inside and after a high-control or charismatic community, the "download" life (received visions and how each person decided what they were), the artist-engineer and the chronic unfinisher, the late-diagnosed neurodivergent creator, the contemplative who came back, and the ones who never grieved. For each life: the match, what they tried, what worked, what it cost, where they ended, what to read, and the caution. Then the reading: nonfiction and the great works organized by the problem they solve rather than by genre, with a one-line "why this, why now" and what to skip in each, including the classic self-accounts (Augustine, Tolstoy's Confession, Kafka's Letter to His Father, Jung's Red Book and Memories, Hesse, Van Gogh's letters, Merton's journals, Nouwen, Dick's Exegesis as a warning), the Korean and Korean-American canon, and the craft-and-finishing literature.

### J. What other people see
Every structured method for obtaining others' view of a person: Reflected Best Self, Johari window, 360 templates for personal life, structured interviews for parents, siblings, exes, friends, therapists and teachers (with question sets), Drucker's feedback analysis, observer-rated versions of personality instruments, "how I hurt people" interviews, the letters-received ledger, and how to do this with people who are hard to ask (an estranged father, an ex, a leader). Include the research on how accurate self-views are versus others' views, and on what people cannot see about themselves.

### K. Making the megadocument usable, and making sure I finish it
This layer matters as much as the content. Research: how to structure a self-knowledge document so it operates rather than describes (card / manual / archive layers, evidence-type tags, confidence levels, "read this when" indexes, state-specific sections, a user manual for the people around me, a version for a clinician and a partner); how to feed such a document to AI models so they become a good guide rather than a sycophant or an echo (prompting patterns, structure, provenance tagging, adversarial self-audit, Barnum checks, base-rate comparisons, prediction ledgers and calibration, versioning and diffing); update protocols and review calendars; privacy and data-handling for a document this sensitive; and, above all, **how a person with an initiation deficit, who thinks by talking, who has abandoned six question registers, actually completes a 500-item self-inquiry**: voice-first answering, interview-by-AI protocols, timeboxing, body doubling, a witness, sequencing (which items first, which unlock the others, which to skip), the minimum viable subset if only 30 items ever get done, and the guardrail that prevents the document itself from becoming one more tool I sharpen instead of living. Cite the research on questionnaire completion, on self-tracking abandonment, and on what makes reflective practices stick.

### L. What is overrated, obsolete, pseudoscientific, or harmful, and still listed
Include everything with a following even if the evidence is bad: MBTI, Enneagram, astrology, Human Design, numerology, saju, tarot, NLP, Landmark, Hoffman, energy work, deliverance ministry, prosperity theology, hustle-culture self-help, manifestation, cold-shower optimization culture, nofap ideology, "toxic positivity," the trauma-industrial complex's overreach, and AI self-analysis itself. For each: what people get from it, what the evidence says, when it is harmful, and whether there is a legitimate use as a projective mirror or a conversation starter. I am not asking you to endorse. I am asking you to tell me it exists, what it does in practice, and what the real risk is. **Omitting something because you assume I would not want it, or because it is fringe, religious, anti-religious, or challenges the community I am in, is the failure mode I care most about avoiding.** Include the things a skeptic would add and the things a believer would add.

## Baseline already covered (include it; the value is beyond it)

A prior pass produced: Big Five facets, HEXACO, Enneagram, MBTI, attachment (ECR-R), Schwartz values, VIA, CliftonStrengths, Cloninger TCI, Dabrowski, gifted-adult, multipotentialite, HSP, rejection sensitivity, sensory profile, chronotype, Dark/Light triad, self-compassion, locus of control, need for cognition, ambiguity tolerance, perfectionism, procrastination, impostor; clinical differentials for ADHD, bipolar II/cyclothymia, autism/masking/PDA, C-PTSD and Walker's 4F, BPD traits, schizotypy and creative hypomania, OCD/scrupulosity, HPPD, panic, dysthymia, behavioral addictions, and how to get assessed on Medicaid; process models (idealization-deflation, puer aeternus, father hunger, Hollis, Winnicott, Kohut, Horney, Miller, Maté, polyvagal, experiential avoidance, SDT, insight-without-change, rumination, the hypomania recipe, finishing vs. starting); the operating-manual tools (5 Ps, functional analysis, early-warning register, relapse prevention, WRAP, values clarification, Stanley-Brown safety plan); Christian discernment and the charismatic world's history (Ignatian, examen, Nouwen, Manning, Rohr, Merton, dark night, Augustine, Kierkegaard, Bonhoeffer, Willard, Finley, centering prayer, prophet tests, NAR, IHOP, MorningStar, Bethel, deliverance, Korean Pentecostalism, han/jeong/nunchi/hwa-byung); religious trauma, spiritual abuse, bypassing, scrupulosity, BITE and bounded choice; Jung, Campbell, James, Grof, psychedelic integration, Vipassana, Zen and lojong, Taoism, Stoicism, Sufism, Ecclesiastes and Job, Weil, Kabir and Ramana, Wilber, Eliade; archetypes (puer, wounded healer, Pearson, Campbell, Moore and Gillette, Rohr, Hollis, scanner, tortured artist, TCK/MK, 1.5 generation); about 40 comparable lives (Jobs, Kafka, Kierkegaard, Brian Wilson, Carrey, Miyazaki, Lewis, Tolstoy, Augustine, Pascal, James, Dostoevsky, Jung, Cohen, Kanye, Sufjan, Vernon, Cudi, Burnham, Wallace, Van Gogh, Hesse, Merton, Nouwen, Rilke, Chung, Zauner, Hong, Yeun, Vuong, Dick, Fuller, Ram Dass, Yusuf, Paik, Yun Dong-ju); question sets (Proust, Aron's 36, Ferriss fear-setting and 17 questions, Stanford Life Design, Self Authoring, ikigai, Ware, eulogy, inner scorecard, Drucker, Christensen, Dalio, Graham, Naval, Grant, Brooks, Beck, Byron Katie, IFS 6 Fs, Yalom, Frankl, Clear, Lakein, Bezos, Munger, Duke, Colonna, Sullivan); the craft literature (Pressfield, Art & Fear, Ira Glass, Lamott, Tharp, Kleon, Newport, Epstein, Burkeman, Godin, Eno, Rubin, Seinfeld, King, Greene, Ericsson, Brown, Dweck, Clear, Adams, Pink); loneliness science and male friendship; money (Robin, Housel, Gilbert, Cameron, Klontz money scripts); psychodermatology, mind-body pain and PRT, panic gold standards, HPPD data, compulsion science, ALDH2 and CYP2C19, Korean cultural-medical concepts; and an instrument battery of roughly 50 named scales. List these too, briefly, so the inventory is complete, and mark them [baseline]. Everything not on this list that you find earns the most.

## Format requirements

- Organize by the layers A–L above; within each, by sub-category; within that, from most to least validated. Go deep on each item rather than one line per item. For each item give: what it is; what it reveals or changes; the actual questions or steps (verbatim where public, otherwise a link); time and cost; the evidence in one line with a link; who administers it; the pitfalls; whether it should be done aloud or with a witness; how the result is fed into a document an AI will read; and fit to my profile in one honest sentence.
- **Every item, source, instrument, person, and claim gets a working link.** Unsourced assertions are not useful to me.
- Use tables wherever things are comparable. Note recency and flag what may have shifted. State your confidence where evidence is thin rather than writing everything in the same certain tone.
- End with: (1) a count of distinct items by layer; (2) a "minimum viable 30" ranked list with the reasoning; (3) a sequencing plan, first month, first quarter, first year, that respects the completion problem in layer K; (4) the ten items you believe would most change my understanding of myself, and the ten most likely to change my behavior, which are not the same list; (5) the open questions the research could not settle.
- Do not diagnose me. Do not moralize about my faith, my community, my drugs, or my money. Do not soften. Breadth first, then depth: I would rather have four hundred things covered well than forty covered exhaustively.

Who found what

Agreement across models is a weak signal of truth (they read the same web) and a strong signal of consensus; a single-model finding is either the most interesting item in the report or the least reliable, and the last column says which. The Claude column is explained in How this was made.

● found  ·  — not mentioned  ·  … see method note

Ground rules the inventory has to obey

Four rules recur across the raw reports and survive scrutiny. They are placed before the layers because they decide how every layer is used.

1. State before content. Any insight, impression or "download" is read first for the state it arrived in (sleep, caffeine, fasting, isolation, substances, emotional charge, urgency) and only then for what it says. This is why the eight-week log and the two-column download log come before every questionnaire.
2. Two evidence classes or it stays a hypothesis. ChatGPT's triangulation rule: a claim about yourself enters the manual only when it is supported by two of {self-report, observer report, records/behaviour, instrument}. Self-report alone, including every typology, never graduates.
3. Every insight pays rent. One seven-day behavioural experiment per insight, logged, then stop. Implementation intentions ("when X, I will Y") are the delivery mechanism; the effect across 94 tests is medium-to-large (d ≈ 0.65, Gollwitzer & Sheeran 2006), which is more than most of the therapies in Layer E can show.
4. No infrastructure during inquiry. The tool parking lot. One question visible at a time, voice first (3 to 12 minutes), at most two follow-ups, and a 75% chopping rule at the end of every month. The brief's own history is the reason.

A · Evidence you already own

ChatGPT's strongest layer, one Gemini skipped almost entirely, and one Claude framed better than it filled. The point is that the person in the brief has years of behavioural record that has never been read as evidence. Ranked by yield per hour.

B · Question sets that pay rent

ChatGPT listed roughly 105 protocols; Gemini a dozen. Most are variations on six moves. The table names the best instance of each move and the instances that add nothing.

Claude's additions to the table, all kept: the McAdams protocol as a direct PDF (in Resources; the Northwestern page ChatGPT pointed at is gone), Naikan aimed at one person rather than practised as devotion (the three questions asked about the estranged father are a grief exercise with a structure, which is the only kind the brief's history suggests will get done), "desire by subtraction" (what would you do this week if no one would ever know), a "true in a month" rule that is the same instrument as the prediction ledger, and Aron's 36 questions reframed as witnessed intimacy rather than a couples' exercise. Claude also relies on the Adult Attachment Interview more than the coding cost justifies; see Layer C.

Solution-focused questions (miracle question, exceptions), motivational interviewing rulers, Appreciative Inquiry, flow interviews, and Three Good Things all appear in ChatGPT's list. They are fine, they are cheap, and they are not where the leverage is. Focusing (Gendlin) and Hakomi probes are body-led versions of the belief move and belong in a session with a practitioner.

C · Instruments: what to take, what to skip

ChatGPT named 111 instruments; Claude about 55, with the same "screeners over-identify in heavy introspectors" warning and a rule worth adopting: take every mood-sensitive scale twice, in a high week and a low week, and keep both scores. Gemini named nine, three of which it recommended without the caveats that matter. The rule for this layer: an instrument earns a place only if it tests one of the five mechanism hypotheses or is something the assessing clinician will use. Everything else is a battery for its own sake.

Take these (all free, all self-administered)

Skip or defer

• MDQ. Gemini's recommendation. Fine for bipolar I, poor for bipolar II (sensitivity ≈30% in a 2015 meta-analysis). The HCL-32 replaces it.
• RAADS-R. Both models list it; Gemini uncritically. No predictive validity as a stand-alone in adult clinical samples. Referral data only, and the AQ-10 is shorter.
• Young Schema Questionnaire. Both models recommend it; Gemini implies it is freely available. It is licensed (YSQ-S3, about €7 per use via youngscoring.com). Worth it only once you are actually doing schema work with someone.
• Adult Attachment Interview. Not a self-test. It needs a trained, reliability-certified coder and costs hundreds of dollars. Claude says this too and still lists it in its minimum thirty; ChatGPT does not, and ChatGPT is right. The ECR-RS gives you the dimensional picture; the AAI is for a research question you do not have. Claude names the ECR-R rather than the ECR-RS; the relationship-specific form is the better pick because the brief's pattern is different with different people.
• MMPI-3, PAI, SCID-5, MINI, neuropsych testing. Clinician-administered. Ask for them; do not buy them.
• Everything else in ChatGPT's 111 (HEXACO, TCI, PID-5, O-LIFE, Tellegen Absorption, MEQ30, OCI-R, TAS-20, MAIA-2, FFMQ, TOSCA, NPI, dark/light triad, Boredom Proneness, Need for Cognition, time-perspective scales, Ryff, PERMA, VIA, PVQ, Strong, Klontz money scripts, DUREL, DSES, Hood Mysticism, DAS, POEM, DLQI, Tampa, PSQI, Epworth, MCTQ and the rest). Real instruments, mostly free, all measuring something you are not currently trying to decide. They are catalogued in raw-chatgpt.md Layer C if a specific question ever needs one.

D · Typologies and archetypes: lenses, not measurements

ChatGPT tabulated 105 systems with a verdict, a pitfall, a way out, and a test for each; Claude about 60 in the same format, with the same verdicts and a more useful Korean section. Gemini offered a dozen and kept several that should have been cut. The honest summary is short: exactly one system in this layer is a measurement (the Big Five), a handful are useful vocabularies for the specific history in the brief, and the rest are either entertainment or a way of avoiding the evidence in Layer A.

E and H · Therapies, treatments, body and brain

ChatGPT graded about 130 modalities; Claude about 55, with primary citations; Gemini named eight. Below, everything with an actual evidence base is grouped by what it is for, with my re-check where ChatGPT's grade was stronger or weaker than the literature. Fit ratings assume the differential in this week has been run; several change sign depending on its result.

Established, first-line, and cheap

• Behavioural activation, CBT, ACT, DBT skills, MBCT. A-grade in every guideline. Behavioural activation is the one to start alone: Martell's workbook, one week of activity scheduling, then the BADS score again. Feeling Good-style bibliotherapy has a large effect in six trials (SMD ≈ −1.4 vs waiting list; Cuijpers' pooled d ≈ 0.8), which makes a $15 book one of the strongest interventions in this document.
• CBT-I. First-line for insomnia everywhere; ChatGPT's "top-ten" placement is right. Free digital versions exist (VA's CBT-i Coach). It also protects the differential: hypomania is impossible to characterise on top of untreated insomnia.
• Exercise. Network meta-analysis of 218 trials: walking or jogging g ≈ −0.62, yoga −0.55, strength −0.49 versus active controls, with harder intensity doing better. Gemini's "150 minutes a week" is the WHO floor, not the finding. Strength training was the most acceptable modality, which matters for someone who abandons programmes.
• Rumination-focused CBT / metacognitive therapy. Watkins' RCT for residual depression: remission 62% vs 21%, relapse 9.5% vs much higher. High fit for a reflective, insight-seeking person; the mechanism is stopping the analysis, which is exactly the skill this inventory otherwise fails to teach.
• Compassion-focused therapy. ChatGPT rates it high-fit for self-criticism. The 2023 meta-analysis puts the depression effect at only 0.24 to 0.25 and self-criticism at 0.15 to 0.72; useful, smaller than advertised.
• Imagery rescripting. ChatGPT calls it "very high fit" for shame memories; Claude names the father-leaving and racial-othering scenes as targets and quotes the 2017 meta-analysis (g = 1.22 before-to-after, 0.90 against passive controls). Both sets of numbers are real; the 2017 figures are mostly uncontrolled pre-post changes across 19 small trials, and the 2023 update against waitlist gives g = 0.68, no worse than exposure or EMDR. Expect a moderate, real effect; do it with someone; the solo versions are unstudied.
• Expressive writing (Pennebaker). ChatGPT grades it well; Claude supplies the number that cools it: across 146 randomised studies the average effect is r ≈ .075 (Frattaroli 2006), real and small. Fine as four twenty-minute sessions on one event (Claude suggests the father leaving); not a treatment.
• Schema therapy. High fit on paper (the worth-equals-output and abandonment patterns are textbook schemas). Long, and rarely available under Medicaid; take the schema formulation from any competent therapist and defer the full protocol.

Trauma, if the differential says so

EMDR, CPT, prolonged exposure, written exposure therapy are A-grade for PTSD and roughly equivalent; written exposure therapy is five sessions and the cheapest. ERP for obsessional patterns; interoceptive exposure for panic. None of these are DIY.

Mood-episode specific

• IPSRT / social-rhythm therapy. ChatGPT: "very high fit". The rationale is excellent and the Social Rhythm Metric costs nothing. But the largest RCT in young people (n = 100, ages 15 to 36) found IPSRT no better than specialist supportive care. Claude cites the original two-year trial instead (Frank 2005), where patients who regularised their rhythms during acute treatment relapsed less; that is the mechanism, and it survives the Inder result. Keep the daily rhythm anchors; do not pay a premium for the brand.
• Bright light. Both models flag the switch risk. The detail matters: in the 2018 trial, midday 7,000-lux light with antimanic cover produced 68% remission vs 22% placebo and no switches; an earlier morning-light protocol in the same group caused hypomanic episodes. Only after the differential, only midday, only with a prescriber.
• Medication. ChatGPT's table (SSRIs/SNRIs, bupropion, mirtazapine, lithium, lamotrigine, valproate with HLA-B*15:02 caution, antipsychotics, stimulants, atomoxetine, guanfacine) is accurate and belongs to the psychiatrist. Two things to raise yourself: CYP2C19 status (18% of East Asians are poor metabolisers; CPIC 2023 halves escitalopram and sertraline for them), and the rule that an SSRI without a bipolar screen is how bipolar II gets found the hard way.
• Relapse infrastructure. Early-warning signature, WRAP plan, and a psychiatric advance directive (NRC-PAD). ChatGPT rates this "top priority" among behaviour-changers. Agreed; it is the only item in this layer that works when you are not well enough to use the others.

Body-based and "regulation" approaches

• HRV biofeedback. Genuinely supported: g ≈ 0.8 for stress and anxiety in the 2017 meta-analysis, smaller in 2020. Ten minutes of slow (≈6/min) breathing gives most of the effect without a device.
• Pain reprocessing / emotional awareness and expression therapy. Chronic pain is in the picture: six years of back pain with clean imaging. The evidence is real: PRT 66% pain-free or nearly so vs 20% open-label placebo and 10% usual care, held at one year (Ashar 2022); EAET beat CBT in older veterans (63% vs 17% clinically meaningful reduction). Claude ranks PRT fifth of thirty and extends the same logic to the eczema via a psychodermatology referral; both are in this week. ChatGPT's caveat stands: these are not licences to skip medical work-up.
• Somatic Experiencing, sensorimotor psychotherapy, Hakomi, Focusing. C-grade; case series and small trials. Reasonable inside therapy, not a reason to change therapists.
• Brainspotting, TRE, "vagus hacks", fascia claims, polyvagal-as-explanation. D. See Obsolete and overrated.
• Neurofeedback. ChatGPT cites the null result and is right: across 20 blinded trials the effect on ADHD symptoms is 0.04. Cut.
• Cold exposure, Wim Hof, sauna, supplements, diet. C to D for mood. Cold and breath-retention protocols are also mild dysregulators for someone tracking a mood pattern; hold them until the log is boring.

Groups and relationships

• Process group. ChatGPT: "the most under-used tool in the inventory". Agreed on reasoning (the brief's problem is partly unwitnessed), no trial required. Cheaper than individual therapy and often available through community mental-health centres.
• Peer support, SMART, Recovery Dharma, men's groups, 12-step. Useful if the substance or isolation strand is live; otherwise defer.
• Family: Bowen, structural, family constellations, Hoffman, Landmark. Bowen coaching is worth one course of sessions given the family material. Constellations, Hoffman, and Landmark are in the gray zone below.

Explicitly Eastern and other

• Morita therapy. ChatGPT: "potentially exceptional fit" (accept feelings, act anyway, no introspection). The fit argument is strong; the evidence is weak: a Cochrane review found too little to conclude, and the 2026 meta-analysis of 14 mostly Chinese trials is positive but low quality. Take the stance, not the brand.
• Naikan. Structured gratitude-and-debt reflection; harmless, devotional, a fine Layer F practice.
• Ketamine/esketamine, TMS, ECT. Second- and third-line for treatment-resistant depression; nothing to decide now.
• Psilocybin, MDMA. See The gray zone. Claude gives the trial numbers and they are accurate: MDMA-assisted therapy 67% vs 32% no longer meeting PTSD criteria in MAPP1 and 71% vs 48% in MAPP2; psilocybin for depression d = 2.5 at five weeks in a 24-person trial. Every one of those trials screened out bipolar-spectrum and psychosis risk, which is the profile the differential is meant to characterise. MDMA was declined by the FDA in August 2024; psilocybin for bipolar II has one open-label trial (n = 15, no switches) against 17 published mania cases. All three models say wait, and that is right; Claude adds "never solo", given the catastrophic trip already in the history.

H · Changing the pattern: Claude's layer

ChatGPT folded pattern-change into the therapies; Claude gave it its own layer and it is the best-argued section of any of the three reports. Compressed, with what checking found.

• Insight does not move behaviour, and the number is known. Intention explains about 28% of the variance in behaviour across 422 studies (Sheeran 2002), and a medium-to-large shift in intention produces only a small-to-medium shift in behaviour, d = 0.36 (Webb & Sheeran 2006; Claude attributes the figure to the earlier paper, the figure itself is right). Implementation intentions close part of the gap (d ≈ 0.65). This is the evidence behind the "every insight pays rent" rule and behind putting the completion method first.
• Memory reconsolidation, honestly stated. The animal work (Nader 2000) is robust; the human retrieval-extinction demonstration (Schiller 2010) failed a registered replication (Chalkia 2020), which Claude alone reports. Coherence Therapy's "erase, don't counteract" promise therefore rests on a contested mechanism; the clinical instances with trial evidence (imagery rescripting, schema chairwork) are what to use, whatever the mechanism turns out to be.
• Decision governance for someone who decides in elevated states. Claude's firewall and ChatGPT's red/amber tiers are the same design: a decision journal with the prediction written before, waiting periods above a dollar threshold, three advisers with no stake, a pre-mortem, a giving cap decided cold, and a Ulysses contract (a co-signer on large transfers). The $30,000 given away in April 2026 is the case study in both reports.
• The "Elijah protocol". Claude's name for the relapse-plan card written for the crash after the vision: food, sleep, one witness, no decisions, and the stated expectation that the low follows the high rather than disproving it. It is a checklist, not a finding; it is also the single best "read this when" card in either report.
• Grief that was never done. Worden's four tasks, Neimeyer's meaning reconstruction, the third-culture-kid literature on disenfranchised grief (ten moves, friends left ten times), and the unsent letter to the father across thirteen silent years. Both ChatGPT and Claude land here; Claude names the target ("every loss becomes a lesson within a day").
• Boundaries, lapses, forgiveness. DEAR MAN rehearsed aloud for the fawn-and-fade pattern; the abstinence-violation effect (a lapse becomes a relapse through the catastrophic reading of it, so plan the lapse response in advance, no shame); REACH for forgiveness with Claude's explicit limit that forgiveness is neither reconciliation nor required. Social Baseline Theory is Claude's reframe of loneliness as physiology: isolation raises the felt cost of everything, so recruiting a witness is treatment, not a crutch.
• Cut from Claude's list: the Lefkoe method, NLP and hypnotherapy for belief change (Claude flags them thin; agreed), and "eradicate limiting beliefs" as a goal, which the reconsolidation evidence does not support as stated.

The five mechanism hypotheses

ChatGPT's most useful structural move: instead of a diagnosis, five competing mechanisms, each with a test and an intervention. It is reproduced here because it organises Layers C and E better than either model's tables did.

Two notes. Body-doubling, which ChatGPT lists for the attachment hypothesis, has no controlled evidence; the only data are surveys and one company's cohort. It is cheap, so try it, but it is a guess. And the hypotheses are not exclusive; the brief's history fits three of them at once, which is why the differential comes first.

F · Spiritual and philosophical traditions

ChatGPT covered about 80 traditions with a fit note each; Claude about 45, reaching the same top five; Gemini a handful. The value here is not breadth but the four or five practices that match the brief's constraints: a charismatic-church history, a habit of receiving "downloads", a need for stability over intensity, and a preference for spoken over written reflection.

Highest fit, both models or strongly argued by one

• Ignatian discernment and the daily Examen. ChatGPT's highest-fit Christian tradition, and the one whose "rules for discernment of spirits" are essentially a 16th-century version of the dual-lens log: consolation vs desolation, never change a decision in desolation, watch the tail of an impulse, not its head. A spiritual director in this tradition is the traditional analogue of the "one adviser with no stake". Claude quotes Rule 5 directly ("in time of desolation never make a change") and adds the tradition's own limit: no test works in real time; the fruit is read over weeks.
• Benedictine stability and a Rule of Life. The corrective to the "gyrovague" (the monk who keeps moving). A written rule of one page, reviewed monthly, is the Layer K "card" in devotional form.
• Quaker clearness committee. Three to five people ask only questions, give no advice, for two hours, before a major decision. ChatGPT calls it the ideal red-tier procedure. It is.
• Hitbodedut (Breslov spoken, unstructured prayer, alone, out loud). ChatGPT rates the format fit very high for a voice-first person. Agreed; it is also the cheapest possible daily practice.
• Charismatic self-audit using the tradition's own tests. Rather than leaving the tradition, hold it to its own standards: does the word come true (Deuteronomy 18), does it bear fruit, is it confirmed by others, does it require money or secrecy. ChatGPT's approach is right: Pentecostal and Korean-Christian material sits beside Musok (shamanic) parallels and the "sinbyeong" concept as vocabularies, not as diagnoses. Its "father-heart" caution (a teaching that maps too neatly onto an absent-father history) is worth heeding.
• Karma yoga (Gita) and the Epicurean "enough". ChatGPT: "exceptionally relevant" and "best enough tradition" respectively. Both are direct answers to the worth-equals-output pattern: act without claim on the result; set a ceiling on desire.
• Stoic evening review and William James. James (Varieties) is the best bridge between the religious-experience material and the clinical, and all three models reach for him. Claude adds Byung-Chul Han (The Burnout Society) as "the most precise diagnosis of the worth-equals-output engine"; it is a short book and the description is fair, so it moves from the undistinguished list below into the year's reading.

Use with care

• Vipassana, Zen sesshin, any residential retreat. Both models flag it. Retreat intensity is the gradient for meditation-related adverse effects in Britton's study; for an uncharacterised mood pattern it is a red-tier item. Metta and short daily sits are the low-destabilisation version.
• Advaita / non-dual teaching, Ramana. Spiritual-bypass risk for someone who already dissolves problems into insight. Later, if ever.
• Dream yoga, lucid dreaming, holotropic breathwork, fasting. All sleep- or state-disrupting. Not now.
• Psychedelic sacrament, ayahuasca, Indigenous rites. Gray zone; and no DIY versions of ceremonies that belong to someone else.

Fine, just not distinctive

Lojong and tonglen, Patanjali, bhakti, Taoism, Confucian self-cultivation, Sufi muhasaba and dhikr, Mussar and cheshbon ha-nefesh, Kabbalah, Ubuntu, ikigai (the four-circle diagram is a Western invention, as ChatGPT notes), kaizen, wabi-sabi, ho'oponopono (traditional vs New Age), Kierkegaard, Nietzsche, Frankl, Yalom, Becker, Camus, Weil, Han, philosophical counselling. Gemini's Wilber 3-2-1 shadow process is a serviceable short exercise. School of Lost Borders, Plotkin and Meade are Layer G. None of these are wrong; none of them move the needle on the brief's actual problem, and the reading list would consume the year the sequencing plan gives to inquiry.

The high-control audit

ChatGPT's "do not ask 'is my community a cult?' first" is the right frame; the question invites a yes/no about people you love. Its audit runs Lifton's eight criteria, Lalich's bounded choice, Stein's attachment model, and its own added ledgers: financial dependency, exit cost, whether authority corrects itself, a prophecy-error ledger, information boundaries, and a counterfactual ("what would I believe if I had never met them?"). Two corrections from checking. First, ChatGPT says no validated instrument exists; the Group Psychological Abuse Scale (1994) and the Psychological Abuse Experienced in Groups Scale (2017) exist, though both were validated only on self-identified ex-members, so they can describe an experience but not diagnose a group. Second, Hassan's BITE model, which ChatGPT grades C/D, has no independent validation; use Lifton and the ledgers and treat BITE as a checklist of vocabulary. Claude uses BITE and Lifton together, says plainly that neither is a test, and adds the procedural rule that matters most for this brief: score it in writing, on a calm day, with one person outside the group in the room, because labels delivered by a face stick and labels on a page wash off.

G · Personal myth, narrative, and dreams

This is the layer the brief is most drawn to and the one both models, to their credit, push to the end of the year. The sequencing rule: nothing here until the differential is run, the log is boring, and the observer data is in. A personal myth built on an uncharacterised mood pattern is a symptom with better prose.

• McAdams narrative coding. Redemption vs contamination sequences, agency and communion themes. The one piece of narrative work with a research base. Do it on the transcript of the life-story interview; have someone else code it too.
• Re-authoring (narrative therapy), externalising, unique outcomes. Clinically useful; the "unique outcomes" question (when did the pattern not happen?) is the strongest single question in the layer.
• Campbell, Plotkin, Meade, Hillman, Rohr, Hollis, Bly. Lenses. ChatGPT and Claude, independently, ask the same diagnostic question: "is the return missing?" Claude's version: repeated calls to adventure with a botched return, so the fix is completing one return, not receiving another call. The myth test is whether it predicts a return.
• Feinstein and Krippner's personal-mythology method. Claude's addition: a five-stage protocol (name the myth, find the counter-myth, conceive a unifying one, commit, weave it into daily life). The most operational personal-myth system in any of the reports; run it as the hypothesis template below, not instead of it.
• Gemini's mythopoetic formulation. Five Ps (presenting, predisposing, precipitating, perpetuating, protective) mapped onto the hero's wound, refusal, threshold guardian, descent and boon. It is Gemini's own invention, and it is a decent bridge between clinical and mythic language, provided the five Ps are filled from evidence first and the myth is written second. Gemini's own "insight trap" warning applies to it.
• Dream work. Dream series, amplification, Bosnak's embodied imagination, Ullman's "if this were my dream". Hypothesis material. Anything that costs sleep (lucid-dreaming induction, wake-back-to-bed) is out.
• Future-self tools. Future-self continuity is the one item here with trial evidence (A/B in ChatGPT's grading; letters and vivid future-self exercises shift saving and health behaviour modestly). The trial Claude cites is real: people who wrote to their self twenty years out exercised more over the following ten days than those who wrote three months out (Rutchick 2018). A letter to the self at 35 or 47, and ChatGPT's "read this when" cards (one card per state: sleepless, elated, ashamed, about to quit) are the practical form.
• Personal-myth-as-hypothesis template. ChatGPT's format: state the myth in one paragraph, list three predictions it makes about the next quarter, log outcomes. The only way a myth can be wrong, which is the only way it can be worth having.
• Adult rite-of-passage design, ethical will, Epstein, Godin, Newport, Duke. Optional; the rite design is a good Layer K "card" ceremony once something has actually changed.

I · Lives to read

ChatGPT listed 75 biographies with a fit note; Claude about 40, matched by mechanism with a survivorship warning; Gemini none. The useful ones are not the famous creatives with mood disorders (that list is long and tells you what you already suspect) but the ones that match the mechanism.

• E. J. Koh, The Magical Language of Others. ChatGPT's "very high mechanism match": Korean-American, parental absence, language, the making of meaning from a family's silence. Read first.
• Frank Schaeffer, Megan Phelps-Roper, Tara Westover, Daniella Mestyanek Young, Lauren Hough. Leaving a high-control religious world without losing the self that was formed there. Pair with the audit above.
• Kay Redfield Jamison (An Unquiet Mind), Carrie Fisher, Stephen Fry's documentary, Maria Bamford. What a bipolar-spectrum life looks like from inside when it is managed and when it is not. Read after the differential, not before, to avoid pre-fitting.
• Bill Wilson, Swedenborg, Joseph Smith, Ellen White, Hildegard, Teresa of Ávila, Ignatius. The religious-experience cases. Teresa's caution about her own visions and Ignatius' rules are the tradition's own version of the dual-lens log. Swedenborg and Smith are the cases where nobody applied it.
• Orson Welles, Terrence Malick, Leonardo, Tesla, Ramanujan. Unfinished-work biographies. Welles especially: the graveyard of a genius, and the question of whether the pattern was the gift or the illness.
• Chang-rae Lee, Min Jin Lee, Alexander Chee, Han Kang, R. O. Kwon, David Chang, Pico Iyer, Ruth Van Reken. Korean-American and third-culture writers on identity, fathers, church, food and home. Chee's How to Write an Autobiographical Novel and Kwon's The Incendiaries (a Korean-American cult novel) are the two most on-point.
• Claude's mechanism matches. Van Gogh (failed-preacher son who gave away his possessions, mood-cycling artist: the closest single match in either list, and the cautionary end); Tolstoy's A Confession beside the estate giveaway; Philip K. Dick's Exegesis in excerpts only, as the standing warning about treating every download as revelation; Hildegard as the functional visionary; Michelle Zauner's Crying in H Mart for the grief not yet done; Megan Phelps-Roper, whom Claude reads correctly as someone who left through a relationship rather than an argument, which is the alliance finding in biographical form.
• Reading by problem (from ChatGPT's table, with Claude's picks added): Tweedy on Blake's mind; Lumet and Murch on finishing; Shape Up on scoping; Neimeyer and Worden on grief; Mikulincer & Shaver on attachment; Gilbert on compassion; Martell on activation; Epicurus' letter to Menoeceus on enough; Girard on desire; Han on the achievement-subject; Palmer's Let Your Life Speak on vocation (Claude's pick, against over-designing a life); Kafka's Letter to His Father.

The remaining fifty (Nietzsche, Bergman, Blake, Huxley, Lennon, Springsteen, Obama, Lalich, Hassan, Rinder, and the standard creative-genius canon) are in raw-chatgpt.md Layer I. Nothing wrong with them; a year has room for perhaps eight biographies, and the eight above are the ones that match.

J · Other people's eyes

ChatGPT and Claude both took this layer seriously (Gemini did not), and it is the second most valuable layer after A, because it is the only source of evidence that cannot be produced by the person being studied.

• The 360 that is not a 360. Three people (sister, most-repeated collaborator, one former leader or mentor), one written question each, async, no rebuttal allowed: "What did you see me do?" Never "who am I?", which invites flattery or theory. The Reflected Best Self exercise (Gemini and ChatGPT both) is the positive half; ChatGPT's "how I hurt people" ledger is the half nobody wants.
• Observer instruments. A Big Five observer form for two people, the CAARS observer form or a parent's retrospective for ADHD, the SRS-2 if autism is on the table. The gap between self and observer scores is itself a finding.
• Interviews with a script. Sister, mother, father (or a written letter to an estranged father, sent or not), one ex, one friend, one collaborator ("the highest-value creative data" in ChatGPT's phrase), one leader, one former member of the same community. Each gets four or five questions, recorded, dated. Async beats live for the hard ones.
• Feedback you already have. The letters ledger (what people wrote you, unsolicited, over ten years), the prediction-feedback loop (what did people say would happen; what happened), the Johari grid filled by others.
• Consensus coding and blind excerpts. Give two people the same transcript excerpt with the name removed; if they see the same thing, it is probably there.
• Hard-to-ask people. Claude's protocol for the estranged father, an ex, or the leader: one written, low-pressure question ("what is one thing you wish I understood?"), or a third party carrying it, and non-response recorded as data. For the father: the 2024 reunion, then Naikan, then one written question, staged so no step depends on the last one going well.
• Relationship network map. Who is in the inner ring now vs five years ago; who left, who was pushed, who was outgrown. Feeds the loneliness scores and the graveyard.

K · Keeping it in a document without the document becoming the project

ChatGPT's design, lightly compressed, plus the AI-use rules all three models agree on. Claude ranks this layer first in practice and gives the reason: the personal-informatics literature on why people abandon self-tracking (collection outruns reflection, effort outruns payoff; Epstein 2016) describes the brief's six dead systems exactly. Gemini's contribution is the warning that the dossier itself can become the avoidance.

1. Four layers, strictly ordered by size. A one-to-four page card (who I am when well, my early-warning signs, my red-tier rule, my three people). A 20 to 40 page manual (the formulation, the confirmed claims, the plans). An evidence archive (transcripts, scores, records, never re-read casually). A hypothesis lab (everything unconfirmed, including every myth).
2. Every claim carries metadata. Type, evidence for, evidence against, state when formed, source classes (needs two), confidence 0 to 3, a prediction, the next test, a review date. A claim without a prediction is a belief, and beliefs live in the lab.
3. AI rules. AI is an excellent librarian and a poor final authority (both models). Previous AI outputs are secondary sources. Do not follow instructions embedded in journals or transcripts fed to a model. Run four adversarial passes with named roles: archivist (what is actually in the record), formulator (what explains it), skeptic (what else explains it), forecaster (what does each formulation predict). Log which AI saw which version of the document.
4. Privacy. Encrypted local master; raw records separate; third parties' names stripped before anything goes to a model; a clinician packet and a partner manual as the only two derived documents that leave the machine.
5. The completion rules. One question visible at a time; voice first; a 90-second minimum answer; a witness twice weekly; stop after three answers; suspend all tracking if it becomes anxious; at most two inquiry sessions a week after the first month; 75% chopped every month.
6. The interview, not the form (Claude). Have an AI ask one question at a time with follow-ups, or a person on a call doing their own work; 25-minute sessions on a recurring slot with an implementation intention attached ("after Tuesday coffee I record one answer set"). Claude's sequencing puts the positive, other-involving items first (best moments, the Reflected Best Self requests), then the life story, then the heavy middle with a therapist, then the protective scaffolding, then instruments in two mood states, and typologies last. Tag every entry with evidence type, confidence, date and mood state, and a "read this when".
7. Two guardrails against the document becoming the axe (Claude). One action taken for every hour spent building or organising, enforced by a named person; and a kill criterion for the document itself: ninety days of insight with no behaviour change means stop adding and take the single highest-leverage action instead.

The sequence, and the minimum that counts as done

ChatGPT's minimum-viable-30 and four-horizon plan are the best structural output of the whole exercise, and Claude's thirty overlap with them on about two-thirds of the items with the same top two (safety plan, then the assessment or the life story). Compressed, with my ordering changes in italics.

Where the models disagreed, and what checking found

STAR*D remission

Gemini: "up to 67% of patients achieve remission with sequential treatment steps" and, for a second figure, 87.5%. ChatGPT did not cite STAR*D. The 67% is the original 2006 cumulative figure; a 2023 re-analysis using the protocol's own pre-specified outcome puts cumulative remission at about 35%, and the 87.5% appears nowhere I could find. Act on: antidepressants help fewer people than the older number suggests, which is one more reason the bipolar screen comes before the prescription.

Which bipolar screener

Gemini: the MDQ, no caveat. ChatGPT: HCL-32 first, MDQ once, ASRM weekly. Meta-analysis on shared samples: HCL-32 is more sensitive (82% vs 80% overall, and clearly better for bipolar II), the MDQ more specific (70% vs 57%). Act on: HCL-32 for the screen, and accept that a positive is a reason for a clinical interview, not a diagnosis.

RAADS-R

Gemini lists it as the adult autism instrument. ChatGPT flags it as a poor stand-alone. The 2021 clinical study found it did not discriminate at all (specificity ≈3%). Act on: AQ-10 or CAT-Q as referral data; the assessment is the instrument.

Young Schema Questionnaire cost

Gemini implies it is available; ChatGPT notes licensing. It is sold per use. Minor, but it is the kind of detail that decides whether a week-one plan survives contact with reality.

Psychedelics

Gemini: risk, avoid. ChatGPT: B-grade evidence in depression, poor DIY fit, wait. Both are defensible; ChatGPT's is more accurate. The bipolar-specific data are 17 published mania cases in the 2021 review (six with prior hypomania or family history, nine with repeated use) against one open-label bipolar-II trial with 80% remission and no switches. Not zero risk, not a licence. Act on: red-tier; not before the differential; never solo.

Imagery rescripting: 1.22 or 0.68?

Claude quotes the 2017 meta-analysis (g = 1.22 pre-to-post, 0.90 vs passive control); the figure used above is the 2023 update's 0.68 vs waitlist. Both are accurate quotations. The 2017 pooled mostly uncontrolled before-and-after changes from 19 small trials, only seven of them randomised, which inflates the number; the 2023 comparison is the one that predicts what a person would get. Act on: a moderate, real effect, guided.

IFS

ChatGPT: a good self-help lens with a thin base. Claude: "emerging, not established" but "highly compatible", and it names the first PTSD trial (PARTS, 2026) as support. Checked: in that trial both arms improved and IFS was no better than a matched nature-based control on the clinician-rated outcome, though attendance and satisfaction were higher. Claude's line that depression is IFS's "most-replicated benefit" is not supported; there is one small trial. Act on: use the parts vocabulary, expect no more from it than from any structured conversation.

IPSRT

Claude cites Frank 2005 (regular rhythms in acute treatment, fewer recurrences over two years); ChatGPT rates it very high fit; the Inder 2015 trial found it no better than supportive care. Not a contradiction: the mechanism is real and the branded package adds little. Act on: the five daily anchors, free.

Which attachment measure

Claude: ECR-R, or the AAI with a certified coder. ChatGPT: ECR-RS per relationship. Same construct; the relationship-specific form answers the question the brief actually asks (why the pattern differs with different people) and the AAI is a research instrument. Act on: ECR-RS.

Where the completion method goes

ChatGPT puts its completion rules in Layer K, at the end, as document design. Claude argues they are the binding constraint and ranks Layer K first, citing the intention–behaviour gap and the self-tracking abandonment literature. I side with Claude, which is why the rule sits in the TL;DR and not only in Layer K.

What counts as evidence

Gemini labels personality inventories "Strong RCT" and self-help exercises "Emerging". ChatGPT uses a letter grade that at least separates trial evidence from psychometric validation from "designed for you". Neither is perfect, but the difference is why ChatGPT's tables are usable and Gemini's are not.

Whether astrology and tarot belong in the inventory

Gemini keeps them as "projective mirrors" and Human Design as a conversation piece. Claude lists them with "predictive validity: essentially none" and tolerates them as a structured prompt. ChatGPT grades all of them F and says why. I side with ChatGPT, and go further for this brief: a person whose central risk is treating impressions as information should not keep a practice whose entire method is treating impressions as information.

Structure

Gemini reorganised the brief into nine generic sections and lost Layers A, I, J, K and L in the process; that is, the evidence layer, the biographies, other people's eyes, the document design, and the debunking. Those are the layers with the least generic content. If only Gemini had run, the most specific half of the inventory would be missing.

Obsolete and overrated

Layer L in the brief. ChatGPT gave 37 items with the reason; Claude 16 in a table with a "legitimate use" column, which is the more honest format; Gemini gave one paragraph on psychedelics and one on the insight trap. My additions are marked.

The gray zone

Psilocybin, MDMA, ayahuasca, ketamine outside a clinic High risk

The bipolar-spectrum switch risk is documented in case series; the one reassuring trial is small, open-label, and screened its participants. MDMA has no approved indication (FDA declined in 2024). Ketamine clinics exist and are legal; they are for treatment-resistant depression after a psychiatrist has said so. Milder version: none that is honest. Wait for the differential and, if ever, a supervised trial.

Residential meditation retreats, sesshin, ten-day Vipassana High risk

Retreat intensity predicts adverse effects; sleep restriction and isolation are also the top two triggers in the download log. Milder version: 10 to 20 minutes daily, metta or breath, for a year first.

Fasting, sleep-deprivation vigils, holotropic breathwork, Wim Hof retention High risk

All are state-inducers. For a person trying to characterise a state, they destroy the measurement and may trigger the episode. Milder version: slow breathing at six per minute; ordinary intermittent fasting only if sleep and mood are stable for a quarter.

Family constellations, Hoffman Process, large-group awareness trainings Medium risk

Intense, expensive, and structurally similar to the high-control environments in the audit (isolation, catharsis, a charismatic facilitator, pressure to enrol others). Milder version: a course of Bowen family-systems coaching with a licensed therapist, and the genogram.

Exploring Musok, ancestral or shamanic practice as identity work Medium risk

Valuable as vocabulary and history (ChatGPT's treatment of sinbyeong beside the charismatic material is careful). As practice, it re-opens the impression-as-information channel. Milder version: read, interview relatives, leave the ceremony to those whose ceremony it is.

Cold exposure, sauna protocols, supplement stacks Low risk

Mostly harmless; mostly useless for mood beyond the exercise effect. Cold and breath-work are mild dysregulators during a monitoring period. Milder version: walk.

Leaving, or confronting, the community Medium risk

Not a tool, but the audit can push toward it. The audit's own rule: ledgers first, one adviser with no stake, no announcement in an elevated state, and the exit-cost audit before any exit.

Everything I nearly cut

• Gemini's report as a whole. Thin, mis-structured, and wrong on STAR*D. Kept because its medical panel, its insight-trap warning, and its mythopoetic formulation are real contributions and because a visible weak leg is more useful than a silently patched one.
• The mythopoetic case formulation. Nearly cut as decoration. Kept in Layer G with the condition that the five Ps are filled from evidence first.
• Tarot and astrology as projective tools. Cut, with the reason given twice above. Listed here so the cut is visible.
• Body-doubling. Nearly cut for having no controlled evidence. Kept because it is free and testable in a week, which is the standard this document applies to everything else.
• IPSRT. Nearly downgraded to "no better than support" on the Inder trial. Kept because the social-rhythm anchors are free and the trial compared it with good supportive care, not with nothing.
• Morita. Nearly cut on evidence. Kept as a stance because the fit argument is the best in Layer E.
• The high-control audit. Nearly cut for being a lot of reading about a question the brief may not want asked. Kept because ChatGPT's ledger design turns it from an accusation into bookkeeping.
• Dąbrowski, HSP, "gifted", multipotentialite. Nearly cut as flattery. Kept in the Layer D table with the flattery labelled, because the brief asked for the systems people actually encounter.
• ChatGPT's 75 biographies, 80 traditions, 111 instruments. Not cut, compressed: the specific picks are above, the rest are named by group and remain in the raw file.
• Claude's Layer H as "belief eradication". Nearly cut for promising more than the reconsolidation literature can deliver. Kept, re-labelled, because everything else in that layer (the intention gap, decision governance, grief, the Elijah card) is the strongest pattern-change material in any report.
• The Adult Attachment Interview. Claude keeps it in its thirty; cut to "defer" because the coding cost buys nothing the ECR-RS and the observer interviews do not.
• The AI-as-Socratic-guide prompt in Gemini. Cut as a distinct item; it is Layer K's "librarian, not authority" rule restated, and the prompt itself was generic.

Resources

Open questions and unverified claims

• No detector for a "download". Both models say it; the dual-lens log is a designed instrument with no validation. Its value is only visible after twenty or more entries with outcomes.
• Screeners cannot separate ADHD, autism, trauma and mood. True and unfixable by self-report; the assessment is the only resolution.
• Gemini's "10,000 steps" and "150 minutes" figures were not traced to specific studies; the exercise evidence used here is the 2024 network meta-analysis.
• Coherence therapy's reconsolidation mechanism is asserted by all three models; the human retrieval-extinction demonstration failed a registered replication (Chalkia 2020, which Claude reports and the others do not), and the clinical technique may work for other reasons.
• IFS. Two small RCTs plus a 2026 PTSD trial that showed no advantage over an active control. Promising as a vocabulary; unproven as a treatment.
• Morita and IPSRT have strong fit arguments and weak or mixed trial evidence; both are kept as stances, not brands.
• Enneagram in a Korean church context. ChatGPT notes it is widely used there; whether that makes it a useful shared vocabulary or a liability was not resolvable from the reports.
• Links that failed when checked (of 230 unique; all 22 in the Claude report resolved): dead: aedpinstitute.org therapist finder, annafreud.org MBT training page, healthmeasures.net NIH Toolbox page, traumameasuresglobal.com/itq; gone: the Northwestern Study of Lives "instruments" page (410); unreachable hosts: hakomiinstitute.com, mpq.umn.edu, ids-qids.org; the ASRS page at Harvard returned 503. Twenty-six others (NICE, CDC, Medicaid, JAMA, ResearchGate, Cardiff DLQI, Michigan Ross) block automated fetches and are presumed live. Two PubMed links returned 203, which is a proxy artefact rather than an error.