Context Window Quest

Tracks

1. 01🪟What a Context Window Really Is

   0/5 lessons

   Working memory, not memory

   Separate model weights, prompt, retrieved context, tool results, output budget, and the human steering that makes all of it usable.

   Lesson list (5)

   • 01
     The Window Is a Contract, Not a Soul~22 min · working-memory, input-output, state
   • 02
     Budget, Not Trash Bin~24 min · budget, safety-margin, output
   • 03
     Context Layers: Prompt, Memory, Retrieval, Tools~25 min · memory, retrieval, tools
   • 04
     Context Management Is Collaboration~28 min · human-steering, collaboration, meaning
   • 05
     The 1M-Token Illusion~26 min · long-context, cost, latency, attention

   ✓Quiz · 5 questions→

2. 02🧮Token Economics

   0/5 lessons

   Count what you spend

   Understand tokenization, input/output/reasoning budgets, multilingual cost, cache asymmetry, and why measuring beats vibes.

   Lesson list (5)

   • 01
     Tokens Are Not Words — BPE in Sixty Seconds~22 min · tokenization, bpe, measurement
   • 02
     Tokenizer Surprises: CJK and Code~22 min · multilingual, code, gotcha
   • 03
     Output and Reasoning Tokens Still Count~24 min · output, reasoning, limits
   • 04
     Cost, Latency, and the Cache Asymmetry~26 min · cost, latency, prefill, cache
   • 05
     The Measurement Loop~23 min · logging, telemetry, budget

   ✓Quiz · 5 questions→

3. 03🧭Attention Drift

   0/5 lessons

   Capacity is not equal quality

   Why advertised windows are not uniformly usable, why middle content gets lost, and why safety margin beats heroic stuffing.

   Lesson list (5)

   • 01
     Advertised Window vs Effective Window~27 min · effective-context, benchmarks, quality
   • 02
     Lost in the Middle~26 min · position-bias, lost-in-the-middle, retrieval
   • 03
     Attention Sinks and Sticky Starts~24 min · attention-sinks, system-prompt, prefix
   • 04
     Do Not Drive to the Wall~25 min · safety-margin, checkpoint, long-session
   • 05
     Tactics Against Position Bias~22 min · position, ordering, duplication

   ✓Quiz · 5 questions→

4. 04🏛️Prompt Architecture

   0/5 lessons

   Structure beats stuffing

   Design prompts in spine, body, and tail. Use section markers, role discipline, and self-documenting structure for long context.

   Lesson list (5)

   • 01
     The Spine: Stable Rules and Tool Schemas~24 min · system-prompt, tools, stability
   • 02
     The Body: Selective Sources and References~25 min · sources, retrieval, selectivity
   • 03
     The Tail: Live State and Next Action~22 min · state, next-action, tail
   • 04
     Section Markers That Help Attention~22 min · xml, markdown, delimiters
   • 05
     Role Discipline: System, User, Assistant~22 min · roles, system-prompt, discipline

   ✓Quiz · 5 questions→

5. 05🤝The Human in the Loop

   0/5 lessons

   Meaning needs an owner

   Turn the human collaborator into an explicit part of context management: steering, correction, authority, and stale-truth cleanup.

   Lesson list (5)

   • 01
     The Human Decides What Still Matters~28 min · human-steering, priority, truth
   • 02
     Authority Boundaries: Who Decides What~24 min · approval, permissions, authority
   • 03
     Interruptions Are Context Events~23 min · interruptions, priority, session
   • 04
     When to Ask, When to Continue~25 min · autonomy, confirmation, workflow
   • 05
     Steering Notes vs Reminders~22 min · steering, reminders, noise

   ✓Quiz · 5 questions→

6. 06📋Compaction and Handoff

   0/5 lessons

   Preserve state before it rots

   Build checkpoints, compact deliberately, and hand the next agent the truth instead of a nostalgic transcript.

   Lesson list (5)

   • 01
     Checkpoint Before Compact~26 min · checkpoint, compact, state
   • 02
     Compaction Is Useful and Lossy~24 min · compaction, loss, verification
   • 03
     Live Handoff Boards~27 min · handoff-board, source-of-truth, operations
   • 04
     Rehydrating a New Session~25 min · rehydration, new-session, sources
   • 05
     Cross-Session Letters~22 min · letters, voice, continuity

   ✓Quiz · 5 questions→

7. 07🗂️Library Strategy & RAG

   0/5 lessons

   Pointers beat piles

   Do not load everything. Index, point, retrieve, inspect, and unload like a careful operator.

   Lesson list (5)

   • 01
     Do Not Put the Whole Library on the Desk~27 min · index, pointers, selective-loading
   • 02
     RAG vs Long Context Is a Routing Problem~26 min · rag, long-context, routing
   • 03
     Tool Results Need Summaries, Not Dumps~24 min · tools, summaries, noise
   • 04
     Evict What Stopped Earning Attention~25 min · eviction, working-set, stale-context
   • 05
     Citations and Provenance~22 min · citations, provenance, trust

   ✓Quiz · 5 questions→

8. 08⚡Caching and Stable Prefixes

   0/5 lessons

   Speed and cost come from structure

   Learn prompt caching, stable prefix design, cache-busting mistakes, and provider-specific knobs.

   Lesson list (5)

   • 01
     Stable Prefix, Variable Tail~25 min · prompt-caching, prefix, structure
   • 02
     Cache Busters: The Top Five~23 min · cache-busting, timestamps, tool-schemas
   • 03
     Provider Differences Without Provider Worship~26 min · openai, anthropic, gemini, portability
   • 04
     Cache Is Not Memory~24 min · cache, memory, state
   • 05
     Cache Math: Worked Examples~24 min · cost, math, savings

   ✓Quiz · 5 questions→

9. 09🩺Meta-Dialog Debugging

   0/5 lessons

   Ask the model what the model sees

   Make the model introspect its own context: drift audits, re-prompt spirals, self-disagreement, tool-trace reading, and cache-hit probing.

   Lesson list (5)

   • 01
     Ask the Model to Introspect~24 min · introspection, drift-audit, evidence
   • 02
     The Re-Prompt Spiral~22 min · loop, rot, anti-pattern
   • 03
     When the Model Disagrees with Itself~22 min · self-disagreement, consistency, audit
   • 04
     Reading Tool-Call Traces for Context Clues~23 min · tools, traces, diagnosis
   • 05
     Probing Cache Hits and Hidden State~24 min · cache, telemetry, probing

   ✓Quiz · 5 questions→

10. 10🛡️Context Operations Playbook

    0/5 lessons

    Run long AI work safely

    Apply context hygiene to coding agents, quest remakes, reviews, debugging, and live collaborator workflows.

    Lesson list (5)

    • 01
      Coding Agents Need Context Hygiene~28 min · coding-agent, repo, workflow
    • 02
      Run Drift Audits~24 min · drift, audit, self-check
    • 03
      Recovering from Context Rot~25 min · recovery, fresh-session, rot
    • 04
      Anti-Patterns Catalog~24 min · anti-patterns, gotchas, checklist
    • 05
      The Final Playbook~27 min · playbook, rules, summary

    ✓Quiz · 5 questions→