CLN Compare Series · Vol. 3
Cognitive Efficiency
= retention, recall speed, extraneous load reduction @ 3 topics · full scorecard · aggregate view
01 // Onboarding Documentation Memory
Prompt: "Write onboarding steps for a new backend engineer joining our team."
Standard LLM Prose Response

Welcome to the team! As a new backend engineer, your first few weeks will be focused on getting oriented with our codebase, tools, and team processes. We recommend beginning with a review of our system architecture documentation, which will give you a high-level understanding of how our services interact before you dive into any specific area.

Once you have a grasp of the overall architecture, you should set up your local development environment. Your team lead will share the relevant repository links and environment variables needed to get the stack running. Don't hesitate to ask for help — the setup process can be tricky the first time.

In your first week, try to get your first pull request merged, even if it's a small improvement or documentation fix. This helps you go through the full development cycle — branching, writing code, running tests, and getting a code review — in a low-stakes context.

During weeks two and three, you'll be gradually assigned more complex tasks as you become comfortable with the codebase. We hold sprint planning on Mondays and a team retrospective every two weeks. You'll also be assigned a buddy engineer who can answer questions during your first month.

CLN Format Structured Response
ONBOARDING: backend engineer // =goal: productive PR merged by end of week 1 @buddy: assigned on day 1 · check in daily for week 1 WEEK 1: orient + first PR // .day 1: read architecture overview doc (link in wiki) .day 1–2: set up local env → ask buddy for .env values .day 3: run full test suite locally → confirm green .day 4–5: pick good-first-issue → open PR → get review →outcome: full dev cycle done once in low-stakes context WEEK 2–3: ramp up // .take on real tasks from sprint backlog .attend sprint planning Mon · retro every 2 weeks .map 2 services you don't understand → ask for walkthrough BLOCKERS: common first-week issues // !env setup fails → ping #dev-env Slack channel (not buddy) !unclear ownership — check CODEOWNERS file first *asking questions is expected — silence is the only mistake
MetricLLM vs CLNΔ Delta
!Semantic Density
3.2
9.0
+5.8
.Scan Speed
3.0
9.5
+6.5
=Structure Clarity
1.0
10.0
+9.0
*Recall Ease
2.8
9.2
+6.4
~Tension Visibility
2.0
7.0
+5.0
→Action Clarity
4.5
9.8
+5.3
?Unknown Honesty
2.5
7.5
+5.0
TOTAL SCORE
avg of 7 metrics
LLM 2.9
CLN 8.9
2.9 / 10
8.9 / 10
+6.0 delta
→ Verdict: Onboarding documents are read under high cognitive load — the new hire is simultaneously processing a new environment, new people, and new codebase. CLN's day-by-day structure with explicit outcomes (→) and blocker flags (!) reduces extraneous cognitive load to near zero. The "ask questions" note is placed last as a * contextual note — not buried in paragraph 3 as a hedge.

02 // Study Notes — TCP/IP Model Learning
Prompt: "Explain the TCP/IP model layers for someone studying for a networking exam."
Standard LLM Prose Response

The TCP/IP model is a conceptual framework that describes how data is transmitted over a network. It consists of four layers, each responsible for a specific aspect of communication. Understanding these layers is essential for networking exams as questions frequently test your ability to map protocols to their respective layers.

The Application Layer is the topmost layer and is where user-facing protocols operate. This includes HTTP, HTTPS, FTP, SMTP, and DNS. When you browse a website, your browser interacts at this layer.

Below that is the Transport Layer, which handles end-to-end communication between processes. The two main protocols here are TCP, which provides reliable, ordered, and error-checked delivery, and UDP, which sacrifices reliability for speed and is used in applications like video streaming or gaming.

The Internet Layer (sometimes called the Network Layer) is responsible for logical addressing and routing. IP operates here, assigning source and destination addresses to packets and determining the best path through the network. ICMP also operates at this layer.

Finally, the Network Access Layer (also called the Link Layer) handles the physical transmission of data on the local network. Ethernet and Wi-Fi protocols operate here, dealing with MAC addresses and frame formatting.

CLN Format Structured Response
TCP/IP MODEL: 4 layers — top to bottom // =exam key: map each protocol → correct layer =mnemonic: "All The Intelligent Nerds" (Application → Network Access) LAYER 4: Application // =what: user-facing protocols · where humans interact .protocols: HTTP · HTTPS · FTP · SMTP · DNS *exam trap: DNS is here, NOT Internet layer LAYER 3: Transport // =what: end-to-end process communication +TCP — reliable · ordered · error-checked (slower) -UDP — unreliable · fast · streaming/gaming *uses ports: 80=HTTP · 443=HTTPS · 25=SMTP LAYER 2: Internet // =what: logical addressing + routing between networks .protocols: IP (v4/v6) · ICMP · ARP *IP addresses live here — NOT MAC addresses LAYER 1: Network Access // =what: physical transmission on local network .protocols: Ethernet · Wi-Fi · MAC addressing *also called Link Layer — both names appear on exams !OSI has 7 layers — TCP/IP has 4 — don't confuse them on exam
MetricLLM vs CLNΔ Delta
!Semantic Density
4.0
8.8
+4.8
.Scan Speed
3.8
9.4
+5.6
=Structure Clarity
1.0
10.0
+9.0
*Recall Ease
3.5
9.5
+6.0
~Tension Visibility
3.0
8.0
+5.0
→Action Clarity
3.5
8.5
+5.0
?Unknown Honesty
2.8
8.8
+6.0
TOTAL SCORE
avg of 7 metrics
LLM 3.1
CLN 9.0
3.1 / 10
9.0 / 10
+5.9 delta
→ Verdict: Study notes have a specific use case: rapid review before an exam. CLN's * notes surface the two highest-value exam traps ("DNS layer" and "OSI vs TCP/IP") as scannable lines — not sentences to parse at speed. The +/− on TCP/UDP lets the reader grasp the trade-off without constructing it mentally from prose. Recall Ease 9.5 vs 3.5 reflects this directly.

03 // Mental Model Capture Knowledge
Prompt: "Explain the mental model behind compound interest for someone who wants to internalize it, not just calculate it."
Standard LLM Prose Response

Compound interest is often called the eighth wonder of the world, and for good reason. Unlike simple interest, which grows linearly, compound interest grows exponentially — meaning the rate of growth itself increases over time as you earn returns on your returns.

The mental model to internalize is this: every unit of money you invest today is not just one unit — it is the seed of a tree whose branching grows faster and faster. In year one, a branch grows. In year two, every branch grows new branches. By year thirty, the tree is enormous compared to what a simple interest model would produce.

The key variables are rate, time, and frequency of compounding. Of these, time is the most powerful and the most overlooked. An investor who starts at 25 will significantly outperform one who starts at 35, even if the late starter saves more money per year, purely because of the extra decade of compounding.

The practical implication is that delay is the enemy. Every year you wait is not just one year of returns lost — it is one year of exponential branching removed from your future. This is why financial advisors emphasize starting early above all other advice.

CLN Format Structured Response
MENTAL MODEL: compound interest // =core insight: you earn returns ON your returns =growth = exponential, not linear — the curve bends upward =intuition: each dollar today is a seed, not a coin KEY VARIABLES: what drives it // *rate — how fast each branch grows *time — how many branching cycles occur *frequency — how often branches split !time = most powerful variable · most underestimated INTERNALISE THIS // !1 year delay ≠ 1 year lost — it removes entire branching cycle !start at 25 vs 35 → 10yr head start → often 2× outcome at 65 ~the model is invisible in year 1–5 · only visible in year 20+ ~human brains are bad at exponential intuition — requires deliberate re-framing →practical rule: start now, optimize rate later →frame every delay as removed branches, not missed months
MetricLLM vs CLNΔ Delta
!Semantic Density
4.2
8.5
+4.3
.Scan Speed
3.5
8.8
+5.3
=Structure Clarity
1.0
10.0
+9.0
*Recall Ease
4.5
9.0
+4.5
~Tension Visibility
3.0
9.2
+6.2
→Action Clarity
4.0
9.0
+5.0
?Unknown Honesty
3.0
8.5
+5.5
TOTAL SCORE
avg of 7 metrics
LLM 3.3
CLN 9.0
3.3 / 10
9.0 / 10
+5.7 delta
→ Verdict: Mental model capture is where CLN's tilde operator shows unique value. The two ~ lines ("invisible in years 1–5" and "brains are bad at exponential intuition") are the exact cognitive obstacles that the reader needs to overcome — they're not flaws in the model, they're friction points in internalizing it. Prose buries these as qualifications. CLN makes them explicit hurdles.

— // Aggregate Results — All 3 Volumes
Cross-Series Average by Metric
9 comparisons across 3 volumes · avg score per metric across all topics
! Semantic Density
LLM 3.7
CLN 8.8
+5.1
. Scan Speed
LLM 3.0
CLN 9.1
+6.1
= Structure Clarity
LLM 1.0
CLN 10.0
+9.0
* Recall Ease
LLM 3.4
CLN 9.0
+5.6
~ Tension Visibility
LLM 3.0
CLN 8.7
+5.7
→ Action Clarity
LLM 3.8
CLN 9.2
+5.4
? Unknown Honesty
LLM 3.1
CLN 8.3
+5.2
OVERALL AVERAGE — ALL VOLUMES
LLM 3.3 / 10 CLN 9.0 / 10
DELTA
+5.7
avg across all metrics