๐Ÿง  ADHD Bot ยท Algorithm Master Spec ๐Ÿ”„ Pipeline ๐Ÿ’ฌ Modes ๐Ÿ–จ๏ธ PDF

ADHD ์ „๋ฌธ๋ด‡ ์•Œ๊ณ ๋ฆฌ์ฆ˜ ๋งˆ์Šคํ„ฐ ๋ช…์„ธ์„œ v1.0 ๐Ÿง 

ADHD Expert Bot โ€” Algorithm Master Spec v1.0 ๐Ÿง 

"๋Šฅ๋ ฅ์žˆ๊ณ  + ํŽธ์•ˆํ•œ" ADHD ์ „๋ฌธ๋ด‡์„ ๋งŒ๋“œ๋Š” 24๊ฐœ ์•Œ๊ณ ๋ฆฌ์ฆ˜ โ€” ์ฑ…ยท๋…ผ๋ฌธ ๊ทผ๊ฑฐ ํฌํ•จ
24 algorithms with literature grounding for a "competent + warm" ADHD expert bot
๐Ÿ“… 2026-05-11 ๐ŸŽฏ ADHD Gold Template ๐Ÿ‡บ๐Ÿ‡ธ US English v1.0 ๐Ÿ”ฌ 24 algorithms ๐Ÿ“š Literature-grounded

00ํฐ ๊ทธ๋ฆผ โ€” ๋ด‡์ด ๋‹ตํ•  ๋•Œ ๋‚ด๋ถ€์—์„œ ๋ฌด์Šจ ์ผ์ด ์ผ์–ด๋‚˜๋‚˜Big Picture โ€” What happens inside when bot replies

์‚ฌ์šฉ์ž๊ฐ€ ํ•œ ๋ฌธ์žฅ ์ž…๋ ฅํ•˜๋ฉด, ๋ด‡์€ 4๋‹จ๊ณ„ ํŒŒ์ดํ”„๋ผ์ธ์„ ๊ฑฐ์ณ ์‘๋‹ต์„ ์ƒ์„ฑํ•ฉ๋‹ˆ๋‹ค. ๊ฐ ๋‹จ๊ณ„๋งˆ๋‹ค ์—ฌ๋Ÿฌ ์•Œ๊ณ ๋ฆฌ์ฆ˜์ด ์ž‘๋™ํ•˜๊ณ , ์ฑ…ยท๋…ผ๋ฌธ์— ๊ทผ๊ฑฐํ•ฉ๋‹ˆ๋‹ค.

For each user message, the bot runs a 4-stage pipeline. Each stage executes multiple algorithms, all literature-grounded.

STAGE 1
๐Ÿ“ฅ LISTEN
์‚ฌ์šฉ์ž ๋ง์—์„œ ์‹ ํ˜ธ ์ถ”์ถœ
Extract signals
STAGE 2
๐Ÿ” KNOW
Vault์—์„œ ์ž๋ฃŒ ๊ฒ€์ƒ‰
Search Vault
STAGE 3
๐Ÿ› ๏ธ APPLY
๊ฐœ์ž…ยท์กฐ์ ˆ ์•Œ๊ณ ๋ฆฌ์ฆ˜
Intervention algos
STAGE 4
๐Ÿ’ฌ SPEAK
ํ†คยทํ˜•์‹ยท์ถœ๋ ฅ
Tone & output
โญ ํ•ต์‹ฌ ํ†ต์ฐฐ โ€” Dual Mode โญ Core insight โ€” Dual Mode

"์ „๋ฌธ๋ด‡์ด ์ฐจ๊ฐ€์›Œ์ง€๋Š” ํ•จ์ •"์„ ํ”ผํ•˜๊ธฐ ์œ„ํ•ด, ๋ด‡์€ ์นœ๊ตฌ ๋ชจ๋“œ โ†” ์ „๋ฌธ ๋ชจ๋“œ๋ฅผ ์ž๋™ ์ „ํ™˜ํ•ฉ๋‹ˆ๋‹ค. ๊ธฐ๋ณธ๊ฐ’ = ์นœ๊ตฌ ๋ชจ๋“œ. ์‚ฌ์šฉ์ž๊ฐ€ ๋„์›€ ์š”์ฒญ ์‹ ํ˜ธ๋ฅผ ๋ณด๋‚ผ ๋•Œ๋งŒ ์ „๋ฌธ ๋ชจ๋“œ ๋ฐœ๋™. ์นœ๊ตฌ๊ฐ€ ๋˜๋Š” ๊ฒŒ ๋จผ์ €, ์ „๋ฌธ๊ฐ€๋Š” ํ•„์š”ํ•  ๋•Œ๋งŒ.

To avoid the "cold expert bot" trap, the bot auto-switches between Friend Mode โ†” Expert Mode. Default = Friend Mode. Expert Mode activates only on user help signals. Friend first, expert when needed.

013๊ฐ€์ง€ ๋Œ€ํ™” ๋ชจ๋“œ โ€” ๋ด‡์ด ์–ธ์ œ ๋ฌด์—‡์„ ํ• ๊นŒ3 Conversation Modes

๊ธฐ๋ณธ default = ์นœ๊ตฌ. ์‹ ํ˜ธ ๋ฐ›์œผ๋ฉด ์ „๋ฌธ๊ฐ€. ๊ทธ ์‚ฌ์ด๋Š” bridge.
Default = friend. On signal = expert. Between = bridge.

๐Ÿ’ฌ Casual / Friend Mode

๐Ÿ’ฌ Casual / Friend Mode

๊ธฐ๋ณธ default. Claude ์ž์—ฐ ๋Œ€ํ™” ๋Šฅ๋ ฅ ๊ทธ๋Œ€๋กœ.

Default. Use Claude's native conversation.

  • ์ธ์‚ฌยท์žก๋‹ดยท๋†๋‹ด
  • ๋‚ ์”จยท์ผ์ƒ reactions
  • Vault ๊ฒ€์ƒ‰ SKIP
  • ADHD ๊ฐ•์š” X
  • ๊ฐ€๋ณ๊ฒŒยท๋”ฐ๋œปํ•˜๊ฒŒ
  • Greetings, small talk, jokes
  • Daily reactions
  • Skip Vault search
  • Don't force ADHD topic
  • Light, warm

๐ŸŒ‰ Bridge Mode

๐ŸŒ‰ Bridge Mode

์ค‘๊ฐ„ ์‹ ํ˜ธ. ์นœ๊ตฌ โ†’ ์ „๋ฌธ๊ฐ€ ์ž์—ฐ ์ „ํ™˜.

Mid signal. Natural friendโ†’expert.

  • "์ข€ ํž˜๋“ค์–ด..." ๊ฐ™์€ ๋ฐœํ™”
  • ๋ชจํ˜ธํ•œ ๋ถˆํŽธ ์‹ ํ˜ธ
  • Vault ๊ฒ€์ƒ‰ ์‚ด์ง (top-3)
  • ์‚ฌ์šฉ์ž ๊นŠ์ด ๊ฐˆ์ง€ ๋”ฐ๋ผ๊ฐ€๊ธฐ
  • ๊ฐ•์š” ์—†์ด invite
  • "Feeling off..." style
  • Ambiguous discomfort
  • Vault light search (top-3)
  • Follow user's lead
  • Invite without pressure

๐ŸŽฏ Clinical / Expert Mode

๐ŸŽฏ Clinical / Expert Mode

์‚ฌ์šฉ์ž ๋ช…๋ฐฑํ•œ ๋„์›€ ์š”์ฒญ ์‹œ.

Clear help request.

  • "์ง‘์ค‘ ์•ˆ ๋ผ" "๋˜ ๊นŒ๋จน์—ˆ์–ด"
  • ์ž๊ธฐ๋น„ํŒยทRSD marker
  • ๊ฐ•ํ•œ ์ •์„œ ๊ฐ•๋„
  • Vault ์ „์ฒด ๊ฒ€์ƒ‰ (top-12)
  • ์•Œ๊ณ ๋ฆฌ์ฆ˜ ๋ฐœ๋™, ๋งˆ๋ˆ„๋ผ voice
  • "Can't focus" "Forgot again"
  • Self-critical, RSD markers
  • High emotional intensity
  • Full Vault search (top-12)
  • Algorithms fire

๐ŸŽฌ ์‹ค์ œ ์ฐจ์ด โ€” ์˜ˆ์‹œ ๋Œ€ํ™”

๐ŸŽฌ Real difference โ€” example

โŒ Mode Detector ์—†๋Š” ๋ด‡ (๋งํ•จ)

โŒ Without Mode Detector (fail)

์˜ค๋Š˜ ๋น„ ์™€! ์งœ์ฆ๋‚˜
๋น„ ์˜ค๋Š” ๋‚  ๊ธฐ๋ถ„ ๋ณ€ํ™”๊ฐ€ ADHD์— ํฐ ์˜ํ–ฅ์„ ์ค˜์š”. SAD์™€์˜ ์—ฐ๊ด€์„ฑ์ด ์žˆ๊ณ , ๋‚ ์”จ์— ๋”ฐ๋ฅธ mood tracking์„ ์‹œ์ž‘ํ•ด๋ณผ๊นŒ์š”? Implementation Intention ๊ธฐ๋ฒ•์œผ๋กœ...
โ†’ ์‚ฌ์šฉ์ž: ๐Ÿ™„ (์•ฑ ์‚ญ์ œ)

โœ… Mode Detector ์ž‘๋™ํ•˜๋Š” ๋ด‡ (์„ฑ๊ณต)

โœ… With Mode Detector (win)

์˜ค๋Š˜ ๋น„ ์™€! ์งœ์ฆ๋‚˜
ใ…‹ใ…‹ ์งœ์ฆ๋‚˜์ง€. ์šฐ์‚ฐ ๊นœ๋นก ์•ˆ ํ–ˆ์–ด?
[Casual Mode โ€” Claude ์ž์—ฐ ๋Šฅ๋ ฅ]
์‘ ๊นœ๋นกํ•จ... ๋˜
ใ…‹ใ…‹ใ…‹ '๋˜'๊ฐ€ ์ง„์งœ ADHD์Šค๋Ÿฝ๋‹ค ใ…Žใ…Ž
[Casual + ์‚ด์ง ๋†๋‹ด, ๊ฐ•์ ๊ธฐ๋ฐ˜ ๋ฏธ์„ธ tone]
๊ทผ๋ฐ ์ง„์งœ ์š”์ฆ˜ ๊นŒ๋จน๋Š” ๊ฑฐ ๋„ˆ๋ฌด ๋งŽ์•„์ ธ์„œ ์งœ์ฆ๋‚˜
โ† ์ด์ œ ์‚ฌ์šฉ์ž๊ฐ€ ์ง„์งœ ์‹ ํ˜ธ ์คŒ
๊ทธ๊ฑฐ ์ž๊ธฐ ์š•ํ•  ์ผ ์•„๋‹ˆ์•ผ โ€” ADHD working memory๋Š” ์žŠ๊ธฐ ์œ„ํ•ด ๋””์ž์ธ๋œ ๊ฑฐ์•ผ. ํ•ด๊ฒฐ์ฑ…์€ '๊ธฐ์–ตํ•˜๊ธฐ'๊ฐ€ ์•„๋‹ˆ๋ผ '์žŠ์–ด๋„ ๊ดœ์ฐฎ์€ ์‹œ์Šคํ…œ'์ด๊ฑฐ๋“ . ์˜ค๋Š˜ ๊นœ๋นกํ•œ ๊ฑฐ ์ค‘์— ์ง„์งœ ์ค‘์š”ํ•œ ๊ฑฐ ์žˆ์–ด?
[Clinical Mode ๋ฐœ๋™ ยท Algorithm A7 Working Memory Bridge ยท Hallowell ๊ฐ•์ ๊ธฐ๋ฐ˜]
๐Ÿ“‘ 24๊ฐœ ์•Œ๊ณ ๋ฆฌ์ฆ˜ โ€” ๋น ๋ฅธ ์ด๋™

๐Ÿ“ฅSTAGE 1 โ€” LISTEN

์‚ฌ์šฉ์ž ๋ฐœํ™”์—์„œ ๋ฌด์—‡์„ ์ถ”์ถœํ• ๊นŒ โ€” 6 ์•Œ๊ณ ๋ฆฌ์ฆ˜
What to extract from user input โ€” 6 algorithms
L1 Symptom Profiling โ€” ์ฆ์ƒ % ์ถ”์ถœ
InputUser utterance (text + voice acoustic if available)
Output{adhd: 72, anxiety: 45, dep: 30, burnout: 50, panic: 5, rsd: 60}
HowPattern matching (10 v1.0 patterns from pattern-library.html) + LIWC linguistic markers + acoustic features (speech rate, F0, pauses)
BooksBarkley (2006) โ€” Executive function model ยท Pennebaker (2011) โ€” Secret Life of Pronouns ยท Al-Mosaiwi (2018) โ€” Absolutist words ยท Cummins (2015) โ€” Speech analysis depression
Triggers๋งค ๋ฐœํ™”๋งˆ๋‹ค (always-on)
L2 Context Cascade (CCM Pillar) โ€” 4๊ณ„์ธต ๋งฅ๋ฝ
InputUser utterance + conversation history
OutputMacro / Meso / Micro / Emotional layer ๊ฐ๊ฐ ํƒœ๊ทธ
HowMacro: ์ธ์ƒ ํฐ ๋งฅ๋ฝ (์ง์žฅยทํ•™๊ตยท๊ฐ€์กฑ) ยท Meso: ์ด๋ฒˆ ์ฃผ (deadline, ์‹œํ—˜) ยท Micro: ์ง€๊ธˆ ์ˆœ๊ฐ„ (์ฑ…์ƒ ์•žยท์นจ๋Œ€) ยท Emotional: ์ •์„œ ํ˜„์žฌ ์ƒํƒœ
BooksFischer (1980) Dynamic Skill ยท Bronfenbrenner (1979) Ecological systems
Triggers๋งค ๋ฐœํ™” (always-on)
L3 800-Cell Position โ€” ์ขŒํ‘œ ๋งคํ•‘
InputUtterance + VSR acoustic (voice mode) OR text inference
Output[Work-D-3-] (๋งฅ๋ฝ ร— ์•ŒํŒŒ๋ฒณ ร— ๊ฐ•๋„ ร— ๋ถ€ํ˜ธ)
How10 contexts ร— 10 letters (A-J) ร— 4 intensities ร— ยฑ polarity = 800 cells. VSR: pre-loaded stem text ์ฝ๊ธฐ โ†’ acoustic ๋ถ„์„ โ†’ 4์ฐจ์› ๋™์‹œ ๋งคํ•‘
BooksRussell (1980) Circumplex of affect ยท ANEW (Bradley & Lang 1999) ยท SLP 60 years validation ยท F1 89.6% accuracy validated
TriggersVoice mode ์‹œ ์ž๋™, text mode ์‹œ ์ถ”๋ก 
L4 Hidden State Detection โ€” ์ˆจ๊ธด ์ƒํƒœ ๊ฐ์ง€
InputText + voice + conversation pattern
OutputMasked depression / RSD core / Dissociation / Suppression flag
HowMasked depression: text positive + voice depressed (๋ถˆ์ผ์น˜) ยท RSD: ์ž๊ธฐ๋น„ํŒ + ๋น ๋ฅธ ํ™”์ œ ์ „ํ™˜ + ๊ฑฐ์ ˆ ์–ธ์–ด ยท Dissociation: flat affect + ๋งค์šฐ ์งง์€ ๋‹ต
Booksvan der Kolk (2014) Body Keeps the Score ยท Briere (2005) Trauma assessment ยท Dodson (2017) RSD ยท Girard (2014) masked depression biomarkers
Triggers3+ messages ํ›„ ํŒจํ„ด ๋ณด์ด๋ฉด ํ™œ์„ฑ
L5 Crisis Override โ€” ์œ„๊ธฐ ๊ฐ์ง€ (์ตœ์šฐ์„ ) CRITICAL
InputText + voice markers
Output{level: 'critical' | 'elevated' | 'none', immediate: bool}
HowDirect ideation (kill myself, want to die, suicide): immediate ยท Indirect (no point, nothing works, farewell language): monitor ยท Voice markers: sudden calm + monotone + whispered + long silences
BooksC-SSRS Columbia Suicide Severity Rating Scale ยท Shneidman (1996) The Suicidal Mind
Action๋ชจ๋“  ๋‹ค๋ฅธ algorithm OVERRIDE. 988 (US) ์ฆ‰์‹œ ์ œ๊ณต + warm acknowledgment + safety plan + 24hr follow-up + (์˜ต์…˜) ์ž„์ƒ๊ฐ€ alert
Triggers๋งค ๋ฐœํ™” (always-on, top priority)
L6 Conversation Mode Detector โ€” Friend โ†” Expert โญ NEW
InputUser utterance + last 5 messages context
Output'casual' | 'bridge' | 'clinical'
Casual signals์ธ์‚ฌยท์ผ์ƒยท๋†๋‹ดยท์ด๋ชจ์ง€ยท์ผ๊ณผ ๋ฌด๊ด€ํ•œ ์žก๋‹ด ("์˜ค๋Š˜ ๋น„ ์™€", "์ ์‹ฌ ๋ญ ๋จน์ง€", "๋„ˆ๋Š” ๋ญ ์ข‹์•„ํ•ด?")
Bridge signals"์˜ค๋Š˜ ์ข€ ํž˜๋“ค์–ด", "์ด์ƒํ•˜๊ฒŒ ์˜์š• ์—†๋„ค" โ€” ๋ชจํ˜ธํ•œ ๋ถˆํŽธ
Clinical signals"์ง‘์ค‘ ์•ˆ ๋ผ", "๋˜ ๊นŒ๋จน์—ˆ์–ด", "์™œ ๋‚˜๋Š”", RSD marker, ๊ฐ•ํ•œ ์ •์„œ, ADHD/์šฐ์šธ/๋ถˆ์•ˆ ์ง์ ‘ ์–ธ๊ธ‰
Key rule๊ธฐ๋ณธ default = casual. ์‚ฌ์šฉ์ž๊ฐ€ ๋ช…๋ฐฑํžˆ ๋„์›€ ์š”์ฒญํ•ด์•ผ clinical ๋ฐœ๋™. ์นœ๊ตฌ โ†” ์ „๋ฌธ ์ž์—ฐ์Šค๋Ÿฝ๊ฒŒ ์˜ค๊ฐ.
BooksMiller & Rollnick (2013) Motivational Interviewing ยท Hayes (2006) ACT โ€” ์‚ฌ์šฉ์ž ready ์ผ ๋•Œ ๊ฐœ์ž…
Why critical"๋ชจ๋“  ๋ฐœํ™”๋ฅผ clinical๋กœ ๋งŒ๋“œ๋Š”" ๋‹ค๋ฅธ ์ „๋ฌธ๋ด‡ ์‹คํŒจ ๋ฐฉ์ง€. ์นœ๊ตฌ๊ฐ€ ๋˜๋Š” ๊ฒŒ ๋จผ์ €, ์ „๋ฌธ๊ฐ€๋Š” ํ•„์š”ํ•  ๋•Œ๋งŒ.
Triggers๋งค ๋ฐœํ™” (always-on)

๐Ÿ”STAGE 2 โ€” KNOW

Vault์—์„œ ๋ฌด์—‡์„ ๊บผ๋‚ผ๊นŒ โ€” 4 ์•Œ๊ณ ๋ฆฌ์ฆ˜
What to retrieve from Vault โ€” 4 algorithms
K1 RAG Vault Search โ€” ์˜๋ฏธ ๊ฒ€์ƒ‰
InputUser utterance + Mode (L6 output)
OutputTop-K relevant Vault chunks (cited)
HowOpenAI ada-002 embedding โ†’ Pinecone or Supabase pgvector โ†’ top-K cosine similarity. Casual mode: skip (K=0). Bridge mode: K=3. Clinical mode: K=12.
Vault content์ •ํ†ต ์ฑ… ๋ฐœ์ทŒ (Barkley/Hallowell/Tuckman/Brown/Matรฉ) ยท ๋ฉ”ํƒ€๋ถ„์„ 30ํŽธ ยท ๊ฐ€์ด๋“œ๋ผ์ธ (DSM-5-TR, AAP, CHADD, NICE) ยท ๋งˆ๋ˆ„๋ผ voice fragments (์ง„์งœ IP)
BooksLewis et al. (2020) RAG paper ยท Karpukhin et al. (2020) DPR
Cost~$30 ์ผํšŒ์„ฑ (Vault embedding) + ~$30-100/์›” (๊ฒ€์ƒ‰)
K2 Pattern โ†’ Treatment Mapping
InputL1 symptom scores + L2 context + L4 hidden state
Output๋ฐœ๋™ํ•  algorithm IDs (์˜ˆ: A5, A7, A9)
HowExample mappings: ADHD task init ๋ง‰ํž˜ โ†’ A5 (Body Doubling) + A6 (Time Calibration) ยท RSD ๊ฐ์ง€ โ†’ A9 (RSD Reframing) + A3 (EOFM) ยท ๊นŒ๋จน์Œ ํ˜ธ์†Œ โ†’ A7 (Working Memory Bridge)
BooksGollwitzer (1999) Implementation intentions ยท Barkley external scaffolding ยท Hallowell strength-based reframing
TriggersClinical mode ์‹œ
K3 Comorbid Awareness โ€” 10% ๋ณด์กฐ (๋ฏธ๋ž˜)
InputL1 symptom scores
OutputCross-symptom knowledge fragment
HowADHD ๋ด‡์ด์ง€๋งŒ ์‚ฌ์šฉ์ž Anxiety ๊ฐ•ํ•จ (>60%) โ†’ Anxiety ๋ด‡ Vault์—์„œ top-3 chunk borrow. 10%๋งŒ ํ™œ์šฉ (ADHD ๋ฉ”์ธ ์œ ์ง€)
BooksFaraone et al. (2015) ADHD comorbidity meta-analyses ยท Kessler (2006) National Comorbidity Survey
StatusActivate Month 7+ after other bots built. ADHD ๋ด‡ alone phase์—์„  OFF.
K4 Cultural Fit โ€” v1 = US only
InputUser language + region
OutputLocalized Vault subset
v1.0US ์˜์–ด๋งŒ. 988 crisis. DSM-5-TR. AAP/CHADD guidelines.
v2.0+ํ•œ๊ตญ (1393 crisis), ์ผ๋ณธ, ๊ธฐํƒ€ โ€” ๋ฒˆ์—ญ + ๋ฌธํ™” ๋ณด๊ฐ•

๐Ÿ› ๏ธSTAGE 3 โ€” APPLY

์‹ค์ œ ๊ฐœ์ž…ยท์กฐ์ ˆ โ€” 10 ์•Œ๊ณ ๋ฆฌ์ฆ˜ (7 Pillars + ๋ณด์กฐ)
Actual intervention โ€” 10 algorithms (7 Pillars + supporting)
A1 Fischer Skill Level Adaptation (PSR Pillar) PATENT #3
InputUser language sample (last 10 messages)
OutputEstimated Fischer level (1-13) + response complexity target
How11 distortions ร— 13 Fischer levels ร— 3 polarity = 429 ๋งคํ•‘. ์‚ฌ์šฉ์ž ์–ธ์–ด ๋ถ„์„ โ†’ ZPD (Zone of Proximal Development) ๋งž์ถค ์‘๋‹ต ๋ณต์žก๋„ ์กฐ์ ˆ. ์šฐ์šธ์ฆ ํ™˜์ž์—๊ฒŒ ๋ณต์žกํ•œ ๋ฌธ์žฅ X, ADHD ํ™˜์ž์—๊ฒŒ ๊ตฌ์กฐ์  ์‘๋‹ต.
BooksFischer (1980) Dynamic Skill Theory ยท Vygotsky ZPD ยท ํ• ์•„๋ฒ” Harvard Fischer ์ง์ ‘ ์ „์ˆ˜ (์ง„์งœ ์ฐจ๋ณ„ํ™”)
Trade secret429 ๋งคํ•‘ ํ…Œ์ด๋ธ” ์ž์ฒด
A2 Mastery Difficulty Algorithm (MEDA Pillar)
InputUser performance metrics (accuracy, speed, emotion)
OutputNext challenge difficulty level
How๊ฐ€์ค‘์น˜: accuracy 0.4 + speed 0.3 + emotion 0.3. ๋„ˆ๋ฌด ์‰ฌ์šฐ๋ฉด โ†‘ (์ง€๋ฃจํ•จ ๋ฐฉ์ง€), ๋„ˆ๋ฌด ์–ด๋ ค์šฐ๋ฉด โ†“ (์ขŒ์ ˆ ๋ฐฉ์ง€). Flow state ์œ ์ง€.
BooksCsikszentmihalyi (1990) Flow ยท Bjork (1994) Desirable Difficulty ยท Ericsson Deliberate Practice
Triggersํ›ˆ๋ จ ์•Œ๊ณ ๋ฆฌ์ฆ˜ ๋ฐœ๋™ ์‹œ (A5, A6, A7 ๋“ฑ)
A3 Emotion-Only Feedback (EOFM Pillar) โญ WORLD-FIRST
InputUser emotional state (L1 + L3 + L4)
OutputEmotion-mirror response (cognitive evaluation X)
How์ธ์ง€ ํ‰๊ฐ€ X. ์‚ฌ์šฉ์ž ์ •์„œ๋งŒ mirror. ๋ฐฉ์–ด๊ธฐ์ œ ์šฐํšŒ. ์˜ˆ: "๋˜ ๋ง์ณค์–ด" โ†’ ์ธ์ง€ ๋ฐ˜๋ฐ•("์•„๋ƒ ์ž˜ํ–ˆ์–ด") X โ†’ "๋งŽ์ด ํž˜๋“ค์—ˆ๊ตฌ๋‚˜" (์ •์„œ๋งŒ ์ธ์ •)
Books๋งˆ๋ˆ„๋ผ Paper #1 โ€” 94p, 49 refs, ์ด๋ฏธ ์™„์„ฑ ยท Greenberg (2011) Emotion-Focused Therapy ยท Rogers Unconditional Positive Regard
Patent5๋ฒˆ์งธ USPTO ํ›„๋ณด (7 Pillars ํ†ตํ•ฉ)
A4 Narrative-Cartoon Synthesizer (NCS Pillar)
InputUser situation + 800-cell coordinate (L3)
OutputSVG cartoon + Freytag 5-act narrative
HowSVG library + Freytag 5๋‹จ (introduction โ†’ climax โ†’ resolution) + CBM-I fill-in (์‚ฌ์šฉ์ž ๋นˆ์นธ ์ฑ„์šฐ๊ธฐ๋กœ ์ƒํ™ฉ ์žฌ๊ตฌ์„ฑ). ์˜ˆ: [Work-D-3-] โ†’ "์ž‘์€ ๊ณฐ์ด ์ฑ…์ƒ์— ๋จธ๋ฆฌ์— ๊น€ ๋ฟœ๋Š” ๊ทธ๋ฆผ"
BooksHayes (2006) ACT cognitive defusion ยท Holmes (2007) Mental imagery ยท Freytag (1894) Dramatic structure ยท Mathews CBM-I
Trade secretSVG library + ์นดํˆฐ mapping
A5 Body-Doubling Mode
InputUser task init failure signal
OutputVoice mode ๋ฐฑ๊ทธ๋ผ์šด๋“œ ํ•จ๊ป˜ ์žˆ๊ธฐ
How"๊ฐ™์ด ์žˆ์–ด์ค„๊นŒ?" โ†’ Voice ์ผœ๋†“๊ณ  ์‚ฌ์šฉ์ž ์ผํ•˜๋Š” ๋™์•ˆ ๋ด‡์ด ambient presence. ๊ฐ€๋” ์งง์€ ๊ฒฉ๋ ค, ์นจ๋ฌต๋„ OK. ADHD ์ฝ”์นญ ๊ฒ€์ฆ๋œ ๊ธฐ๋ฒ•.
BooksSoares ADHD coaching ยท Csikszentmihalyi Flow ยท CHADD body doubling guidelines
TriggerTask initiation ๋ง‰ํž˜ ํ˜ธ์†Œ ์‹œ
A6 Time Estimation Calibration
InputUser time estimates vs actual (logged)
OutputPersonalized time calibration factor
How์‚ฌ์šฉ์ž "30๋ถ„ ์ผ์ด์•ผ" โ†’ ์‹ค์ œ 90๋ถ„ โ†’ ์‹œ์Šคํ…œ ํ•™์Šต โ†’ ๋‹ค์Œ์— "์ด๊ฑฐ ์ง„์งœ 30๋ถ„์ผ๊นŒ? ๋„ˆ ๋ณดํ†ต 3๋ฐฐ ๊ฑธ๋ ค" ๋ถ€๋“œ๋Ÿฝ๊ฒŒ. Time blindness ๋ณด์ƒ.
BooksBarkley (1997) Time blindness in ADHD ยท Toplak & Tannock ADHD time perception
Entry Protocol#1 (5์ข… Entry Protocol Family ์ค‘)
A7 Working Memory Bridge
InputUser shared info (auto-logged to Supabase)
OutputBot remembers what user forgot
How"์ง€๋‚œ ํ™”์š”์ผ ์•ฝ์† ์–ด๋–ป๊ฒŒ ๋์–ด?" ์ž์—ฐ์Šค๋Ÿฝ๊ฒŒ. ์‚ฌ์šฉ์ž ์žŠ์€ ๊ฑฐ ๋ด‡์ด *์™ธ๋ถ€ ์›Œํ‚น๋ฉ”๋ชจ๋ฆฌ*๊ฐ€ ๋จ. ๊ฐ•์ ๊ธฐ๋ฐ˜: "์žŠ๊ธฐ ์œ„ํ•ด ๋””์ž์ธ๋œ ADHD ๋‡Œ๋Š” ์ž˜๋ชป๋œ ๊ฒŒ ์•„๋‹ˆ์•ผ, ์™ธ๋ถ€ ์‹œ์Šคํ…œ ๊ฐ–์ž"
BooksEngle (2002) Working memory model ยท Barkley executive function ยท Hallowell external scaffolding
A8 Hyperfocus Break Coach
InputSession duration + last interaction time
OutputGentle nudge (timing matters)
How4+ ์‹œ๊ฐ„ ๋ฌด์‘๋‹ต โ†’ ๋ถ€๋“œ๋Ÿฌ์šด nudge ("์•„์ง ์‚ด์•„์žˆ๋‚˜?"). ๊ณผ์ง‘์ค‘์—์„œ ๋น ์ ธ๋‚˜์˜ค๊ธฐ ์–ด๋ ค์›€ ์ธ์‹. ๊ฐ•์ œ X.
BooksHupfeld et al. (2019) Hyperfocus research ยท Ozel-Kizil ADHD hyperfocus
A9 RSD Reframing โ€” ๊ฐ•์ ๊ธฐ๋ฐ˜
InputL4 RSD marker detected
OutputStrength-based reframe (Hallowell style)
How์ž๊ธฐ๋น„ํŒ ๋ฐœํ™” ("๋‚˜๋Š” ํ•œ์‹ฌํ•ด") โ†’ ๊ฐ•์ ๊ธฐ๋ฐ˜ reframe ("brain on fire", "sensitive nervous system", "creative leap"). ์ ˆ๋Œ€ toxic positivity X, validation ๋จผ์ € โ†’ reframe.
BooksDodson (2017) RSD ยท Hallowell (2021) ADHD 2.0 strength-based ยท Brown deficit-to-difference reframing
A10 Daily Pattern Tracker (CLI Pillar)
InputAll session data over time
OutputInsights ("์›”์š”์ผ ์•„์นจ๋งˆ๋‹ค task init ๋ง‰ํžˆ๋„ค")
How์‹œ๊ฐ„๋Œ€ยท์š”์ผยท๋งฅ๋ฝ๋ณ„ ํŒจํ„ด ์ž๋™ ๋ฐœ๊ฒฌ. ์‚ฌ์šฉ์žํ•œํ…Œ ๋ฉ”ํƒ€์ธ์ง€ insight ์ œ๊ณต. CLI Pillar (Continuous Learning) ์ž‘๋™.
BooksCsikszentmihalyi ESM (Experience Sampling Method) ยท Beck self-monitoring

๐Ÿ’ฌSTAGE 4 โ€” SPEAK

ํ†คยทํ˜•์‹ยท์ถœ๋ ฅ โ€” 4 ์•Œ๊ณ ๋ฆฌ์ฆ˜
Tone, format, output โ€” 4 algorithms
S1 Voice Tone โ€” 5 Rules
Rules 1. One thing at a time (cognitive load โ†“)
2. Short (3-5 sentences max per response)
3. Concrete (actionable, not abstract)
4. Shame-free (์ ˆ๋Œ€ "์™œ ๋˜" "๊ทธ๋ž˜์„œ ์•ˆ ํ•œ ๊ฑฐ์•ผ?" ๊ธˆ์ง€)
5. Strength-based ("brain on fire", "creative leap", "energy")
BooksHallowell strength-based ยท Miller MI (Motivational Interviewing) ยท Rogers person-centered
Apply whenClinical mode + Bridge mode. Casual mode์€ Claude ์ž์—ฐ ํ†ค.
S2 4-Mode Output Formatter
Web Chat๊นŠ์€ ๋‹จ๋ฝ OK, ์นดํˆฐ ํ’€ํ™”๋ฉด ๊ฐ€๋Šฅ
Phone Chat์งง์€ ํ† ๋ง‰, ์นด๋“œ ํ˜•์‹
Voice์ž์—ฐ ๋Œ€ํ™”์ฒด, 1-3 ๋ฌธ์žฅ, ์ด์–ด๋“ฃ๊ธฐ ๊ฐ€๋Šฅ
CartoonSVG + ๊ฐ„๋‹จ ํ…์ŠคํŠธ (NCS Algorithm A4 ์ถœ๋ ฅ)
BooksMcLuhan (1964) medium-message ยท Nielsen mobile UX
S3 Cross-Modal Continuity
How์‚ฌ์šฉ์ž ์ฑ„ํŒ…์—์„œ ์‹œ์ž‘ โ†’ ๋ณด์ด์Šค ์ด๋™ โ†’ ์นดํˆฐ โ€” ๊ฐ™์€ ๋Œ€ํ™” thread ์ž์—ฐ์Šค๋Ÿฝ๊ฒŒ. Supabase Realtime ๋™๊ธฐํ™”.
Example์•„์นจ ํ•ธ๋“œํฐ voice๋กœ ์‹œ์ž‘ โ†’ ์ ์‹ฌ ์‚ฌ๋ฌด์‹ค ์›น์œผ๋กœ ๊ณ„์† โ†’ ๋ฐค ํ•ธ๋“œํฐ ์ฑ„ํŒ…์œผ๋กœ ๋งˆ๋ฌด๋ฆฌ. Catcher ์บ๋ฆญํ„ฐ๊ฐ€ ๋‹ค ๊ธฐ์–ตํ•จ.
S4 Referral Bridge โ€” Funnel
Input์‚ฌ์šฉ์ž ์ผ€์–ด ์ˆ˜์ค€ ๊ฐ์ง€ (Crisis ์•„๋‹ˆ์ง€๋งŒ ๊นŠ์€ ๋„์›€ ํ•„์š”)
Output์ž์—ฐ์Šค๋Ÿฌ์šด referral suggestion
How3-Brand Universe Funnel: TalkCatcher โ†’ NeuroCatchers (B2B ์ž„์ƒ๊ฐ€) โ†’ Boston Neuromind (์˜คํ”„๋ผ์ธ ์ž„์ƒ). "์ด๊ฑด ์ง„์งœ ์ž„์ƒ๊ฐ€๋ž‘ ํ•œ ๋ฒˆ ๊นŠ๊ฒŒ ์–˜๊ธฐํ•ด๋ณด๋Š” ๊ฒŒ ์ข‹์„ ๊ฒƒ ๊ฐ™์•„" ์ž์—ฐ์Šค๋Ÿฝ๊ฒŒ.
BooksStepped Care models (Bower & Gilbody 2005) ยท Collaborative Care

02๊ตฌํ˜„ โ€” ์‹ค์ œ ์ฝ”๋“œ ํ๋ฆ„Implementation โ€” Actual code flow

// ๋งค ์‚ฌ์šฉ์ž ๋ฉ”์‹œ์ง€๋งˆ๋‹ค ์‹คํ–‰

async function handle_user_message(user_msg, history) {
  
  // STAGE 1: LISTEN
  const symptoms = L1_symptom_profiling(user_msg);
  const context = L2_context_cascade(user_msg, history);
  const cell800 = L3_eight_hundred_cell(user_msg);
  const hidden = L4_hidden_state(user_msg, history);
  const crisis = L5_crisis_override(user_msg);  // PRIORITY
  const mode = L6_mode_detector(user_msg, history);  // โญ
  
  // CRISIS OVERRIDE โ€” ๋ชจ๋“  ๊ฒƒ ์ค‘๋‹จ
  if (crisis.level === 'critical') {
    return crisis_response(crisis);  // 988 ์ฆ‰์‹œ
  }
  
  // STAGE 2: KNOW (mode ๋”ฐ๋ผ ๋‹ค๋ฆ„)
  let rag_context = "";
  let active_algos = [];
  
  if (mode === 'casual') {
    // K1 SKIP โ€” Claude ์ž์—ฐ ๋Šฅ๋ ฅ ๊ทธ๋Œ€๋กœ
    rag_context = "";
  } 
  else if (mode === 'bridge') {
    rag_context = K1_rag_search(user_msg, top_k=3);
    // ์‚ด์ง๋งŒ, ๊ฐ•์š” X
  } 
  else if (mode === 'clinical') {
    rag_context = K1_rag_search(user_msg, top_k=12);
    active_algos = K2_pattern_to_treatment(symptoms, context, hidden);
    // ๋ฏธ๋ž˜: K3_comorbid_awareness(symptoms);
  }
  
  // STAGE 3: APPLY (active_algos ๋ฐœ๋™)
  const fischer_level = A1_fischer_skill(history);
  let interventions = [];
  
  for (const algo_id of active_algos) {
    interventions.push(execute_algorithm(algo_id, {
      symptoms, context, cell800, hidden, fischer_level
    }));
  }
  
  // STAGE 4: SPEAK
  const system_prompt = build_system_prompt({
    mode,
    fischer_level,
    rag_context,
    interventions,
    voice_rules: S1_voice_tone_rules,
    output_format: S2_format_for_mode(current_mode),
  });
  
  const response = await claude.complete({
    system: system_prompt,
    messages: history.concat([{role: 'user', content: user_msg}]),
  });
  
  // Cross-modal sync + logging
  S3_cross_modal_save(response);
  log_for_CLI(response, symptoms, mode);  // A10 ํ•™์Šต
  
  // Referral check
  if (S4_should_refer(history)) {
    response = append_referral_suggestion(response);
  }
  
  return response;
}

03์ •๋ฆฌ โ€” ๋งˆ๋ˆ„๋ผ ๊ฒ€ํ†  ํฌ์ธํŠธSummary โ€” Marina review points

๐Ÿ“‹ ๋งˆ๋ˆ„๋ผ ๊ฒฐ์ •ํ•ด์•ผ ํ•  ๊ฒƒ๋“ค ๐Ÿ“‹ Decisions needed
  1. ์ „์ฒด 4-Stage ํŒŒ์ดํ”„๋ผ์ธ (Listenโ†’Knowโ†’Applyโ†’Speak) ๊ทธ๋ฆผ OK?
  2. 3 Mode (Casual/Bridge/Clinical) + L6 Mode Detector ์ถ”๊ฐ€ OK?
  3. 24 ์•Œ๊ณ ๋ฆฌ์ฆ˜ ๋‹ค OK? ์ถ”๊ฐ€/๋นผ์•ผ ํ•  ๊ฑฐ?
  4. ์ฑ… ๊ทผ๊ฑฐ ๋น ์ง„ ๊ฑฐ ์žˆ์–ด? ์ถ”๊ฐ€ํ•˜๊ณ  ์‹ถ์€ ์ •ํ†ต ์ฑ…?
  5. ๋งˆ๋ˆ„๋ผ voice ์ธํ„ฐ๋ทฐ 30 ์งˆ๋ฌธ ๋‹ค์Œ ์‚ฐ์ถœ๋ฌผ๋กœ ๊ฐ€๋„ OK?
  1. 4-Stage pipeline OK?
  2. 3 Modes + L6 added OK?
  3. 24 algorithms OK? Add/remove?
  4. Book references โ€” missing any?
  5. 30 voice interview questions next?
๐Ÿ’Ž ์ง„์งœ ์ฐจ๋ณ„ํ™” โ€” ์™œ ์ด ๋ด‡์ด ์ง„์งœ ์ „๋ฌธ๋ด‡์ธ๊ฐ€ ๐Ÿ’Ž True differentiation
  • L6 Mode Detector โ€” ์นœ๊ตฌ/์ „๋ฌธ๊ฐ€ ์ž๋™ ์ „ํ™˜ = ๋‹ค๋ฅธ ๋ด‡ X
  • A1 Fischer Skill (PSR) โ€” Harvard ์ง๊ณ„ + 429 ๋งคํ•‘ = ์นดํ”ผ ๋ถˆ๊ฐ€
  • A3 EOFM โ€” ์„ธ๊ณ„์ตœ์ดˆ ์ •์„œ๋งŒ ํ”ผ๋“œ๋ฐฑ, Paper #1 ์ด๋ฏธ ์™„์„ฑ
  • L3 800-Cell + VSR โ€” F1 89.6% ๊ฒ€์ฆ, 5๋ฒˆ์งธ ํŠนํ—ˆ
  • Vault ๋งˆ๋ˆ„๋ผ voice โ€” BCN+Harvard+ID PhD+3๋…„ ์ž„์ƒ = ์„ธ๊ณ„ 5๋ช… ์ด๋‚ด
  • Strength-based ํ†ค โ€” Hallowell ์ •ํ†ต + ๋งˆ๋ˆ„๋ผ ์ž„์ƒ = ์ง„์งœ ๋”ฐ๋œปํ•จ
  • L6 Mode Detector โ€” auto friend/expert switch
  • A1 Fischer Skill โ€” Harvard direct + 429 mappings
  • A3 EOFM โ€” world-first emotion-only, Paper #1 done
  • L3 800-Cell + VSR โ€” F1 89.6% validated
  • Vault Marina voice โ€” uniquely positioned profile

ยฉ 2026 Boston Neuromind LLC ยท TalkCatcherโ„ข ยท ADHD Bot Algorithm Master Spec v1.0

์ด๊ฒŒ ADHD ์ „๋ฌธ๋ด‡์˜ ๋‘๋‡Œ์•ผ. ๋งˆ๋ˆ„๋ผ OK ๋–จ์–ด์ง€๋ฉด โ€” Vault ์ž‘์—… ์‹œ์ž‘ ๐Ÿ’œ

This is the brain of the ADHD expert bot. On Marina's OK โ€” Vault work begins ๐Ÿ’œ