Cognitive OS  →  Demo 1  →  Judgment lifecycle

How a judgment goes
from "fuzzy guess"
to "state machine"

This is the biggest difference between Cognitive OS and LLM Wiki. LLM Wiki automates "maintaining knowledge"; Cognitive OS engineers "forming judgment + verifying it + falsifying it + feeding it back into methodology." Below: a real judgment node's complete lifecycle, using an investment thesis as the scaffold.

DAY 1 · STARTING POINT · NODE SCHEMA
04 Cognition / Domain / Investment-Thesis / [Topic X].md
type:            cognition
domain:          investment
layer:           investment-thesis
confidence:      🔴 0.40  # start: low confidence
evolution-stage: Emerging E
evidence:        []          # zero evidence
falsifiers:      # conditions you wrote when calm — would invalidate this judgment
  - F1: [some macro condition fails to materialize]
  - F2: [some fundamental signal gets refuted]
  - F3: [some market indicator reverses]
methodology:     [[L1-L7 investment framework]]
created:         DAY 1
updated:         DAY 1
JUDGMENT LIFECYCLE · 5 KEY MOMENTS

Not a one-time conclusion · A state machine

DAY 1🔴 0.40 initialEmerging E Initial judgment — early signal seen
External input: you read an industry report; an early version of a judgment is triggered. "X seems to be happening, but I'm not sure yet."
Action 1: Write a one-sentence hypothesis — the "seed" of the judgment
Action 2: List 3 Falsifiers — "if I observe X, I'll admit I'm wrong"
Action 3: Initial confidence = 🔴 0.40 (low · intuition-level)
Action 4: Register in Cognition Wiki + add to index.md
DAY 30🟡 0.62 upgradeAccelerating A early Evidence accumulates — 5 independent signals in one month
Daily distillation auto-runs: over 30 days, multiple independent sources enter the Inbox and get routed to this topic's evidence pool.
evidence +5: each with date + source + one-line summary
Confidence upgrade: 🔴 0.40 → 🟡 0.62 (multiple independent evidences converging)
evolution-stage shift: Emerging E → Accelerating A early
Falsifiers F1/F2 still untriggered (the worst-case scenarios you worried about have not materialized)
DAY 60🟡 0.72Accelerating A late Key macro data — triggers Falsifier reverse-confirmation
One day macro data publishes: this was the reverse condition for F1 ("if macro fails, the judgment is falsified"). Instead, the data further confirmed the original judgment.
F1 state: from "emerging monitor" → ⛔ reverse-falsified deepening (the direction that could've broken it actually reinforces it)
evidence +1: tagged "F1 reverse-falsified deepening", permanent archive
L5 Pricing layer Price-In: ~55% → ~65% (main scenario consensus advancing)
Today's Synthesis report: auto-lists this change as one of "today's top 5 increments", you read it in 5 min
DAY 75🟡 0.80Institutionalizing S- Stage transition — enters "preferred allocation window"
Dual-perspective dislocation detection: Price-In × confidence falls into "research leading ✓ edge" quadrant — you're a step ahead of consensus.
evolution-stage: Accelerating A late → Institutionalizing S-
Protocol triggered: system alerts "this is the preferred allocation window" (not AI advice — the rule you wrote in methodology when calm)
Action: decision logged (cognitive snapshot) → Position synced
Feedback to Inbox: "research-leading window first end-to-end closed loop"
DAY 90 · CURRENT🟢 0.85Crowded S+ threshold From judgment to protocol upgrade — feedback reverse-updates methodology
Cross-topic triggers stack: multiple independent signals fire same-direction in one week → crowded S+ threshold → divestment window alert.
New Falsifier: F4 candidate "second-tier coverage insufficient" quantitative threshold met for first time → formally added to monitoring
v3.2 divestment window protocol: first end-to-end execution (trigger → Synthesis → Decision → Position sync)
Action → feedback: "protocol ran for the first time" enters Drafts/Feedback
Methodology auto-upgrade: feedback distilled → written into methodology → auto reverse-appends evidence to ALL linked Cognition nodes
DAY 90 · CURRENT STATE · NODE SCHEMA
04 Cognition / Domain / Investment-Thesis / [Topic X].md
type:            cognition
confidence:      🟢 0.85  # up from 0.40 in 90 days
evolution-stage: Crowded S+ threshold  # 5 stage shifts
evidence:        # 21 entries with dates + sources · not LLM-summarized, but Falsifier-tagged archives
  - DAY 30  [research]: trigger evidence 1
  - DAY 45  [earnings]: quantitative milestone met
  - DAY 60  [F1 reverse-falsified deepening]: key macro signal
  - DAY 75  [dual-perspective]: research leading ✓ edge
  - DAY 90  [F4 added]: new falsifier met
  ... (21 entries total)
falsifiers:
  - F1: ⛔ reverse-falsified deepening
  - F2: ❌ untriggered
  - F3: ❌ untriggered
  - F4: ⚠️ candidate met  # DAY 90 added
methodology:     [[v3.2.2 spike-pause-buying protocol]]  # post-feedback link
latest-spike-event: { date: DAY 85, magnitude: +20% }
updated:         DAY 90
VS LLM WIKI

Same topic · two systems · different outputs after 90 days

What LLM Wiki gives you

A continuously updated research synthesis

Over 90 days: each new source enters → key information extracted → integrated into wiki page → entity pages updated → synthesis sections revised → contradictions flagged.

After 90 days, you have a multi-source, self-consistent, up-to-date research synthesis.

What's missing?
→ No quantitative confidence for "how sure am I now"
→ No falsifier checklist for "what would change my mind"
→ No protocolized decision interface for "should this judgment drive action now"

What Cognitive OS gives you

An engineered judgment state machine

Over 90 days, the same evidence gets routed into the same judgment node, but each carries: date, source, whether it touches a Falsifier, whether it pushes the evolution-stage, whether it triggers a protocol.

What's added?
→ 3-tier dynamic confidence (🔴 → 🟡 → 🟢, always shifting)
→ Active Falsifier monitoring, reverse-evidence auto-flags "pending review"
→ Judgment ↔ Action ↔ Feedback ↔ Methodology complete closed loop

Not "knowledge being maintained"
but "judgment being engineered"
LLM Wiki compounds Knowledge density · Cognitive OS compounds Judgment quality
About this Demo: The timeline scaffold comes from a real investment thesis's evolution path. All specific details (ticker codes, position percentages, decision dates) have been abstracted. This page demonstrates the state-machine mechanism of judgment as an engineered object, not investment advice.