// Third batch: usage, decision rule, habits, closing. function WeeklyBudget({ page, total }) { return (
Your weekly budget
A real ceiling — not a <br/>soft nudge. <em style={{ color: COLORS.rust }}>Respect it and it rarely bites.</em>
Weekly
Total cap for the week.
Often split across “All models” and a separate “Sonnet only” counter, depending on your plan.
Session
A rolling short-window cap.
Resets every few hours. Prevents you from burning the whole weekly allowance in one sitting.
); } function CostDrivers({ page, total }) { const drivers = [ { n: '01', h: 'The model you picked', b: 'Opus costs roughly 10–20× more per token than Haiku.' }, { n: '02', h: 'How much context is loaded', b: 'Long conversations, attached files, and active connectors all grow the bill.' }, { n: '03', h: 'How long Claude\u2019s reply is', b: 'A short answer costs fewer output tokens than a five-page essay.' }, ]; return (
Three cost drivers
Every bill comes from the same three levers.
{drivers.map((d, i) => (
{d.n}
{d.h}
{d.b}
))}
); } function UsageTab({ page, total }) { return (
Reading the usage tab
Three counters, three time horizons.
If you’re hitting session limits often, the model is too big or the chats are too long. If you’re hitting weekly limits, your overall pattern needs tuning.
{[ { label: 'Current session', sub: 'Rolling short-window limit. Resets every few hours.', bar: 0.42, color: COLORS.rust }, { label: 'Weekly limit', sub: 'Total cap across the week.', bar: 0.28, color: COLORS.ink }, { label: 'Daily scheduled runs', sub: 'Routine runs used today vs. your daily allowance.', bar: 0.6, color: COLORS.sage }, ].map((row, i) => (
{row.label}
{Math.round(row.bar * 100)}%
{row.sub}
))}
); } function TwoQuestions({ page, total }) { return (
The decision rule
Two questions. Every time.
{[ { n: '01', q: 'How hard is this task?', a: ['Easy, one-step → Haiku', 'Moderate judgment → Sonnet', 'Ambiguous, real nuance → Opus'] }, { n: '02', q: 'How often will I run it?', a: ['One-time or rare → pick by hardness', 'High-volume / scheduled → drop one tier', 'Sonnet → Haiku; Opus → Sonnet'] }, ].map((b, i) => (
{b.n}
{b.q}
    {b.a.map((x, j) => (
  • — {x}
    • ))}
))}
); } function WorkedExamples({ page, total }) { const rows = [ { task: '"Summarize this email in one sentence."', h: 'Easy', v: 'One-time', pick: 'Haiku', vehicle: 'bike', }, { task: '"Every weekday at 7am, triage my inbox and draft replies."', h: 'Moderate', v: 'Daily, automated', pick: 'Sonnet', note: 'Dropped one tier because it runs every day.', vehicle: 'sedan', }, { task: '"Help me decide: rebuild our onboarding flow or patch it?"', h: 'Hard, ambiguous', v: 'One-time', pick: 'Opus', vehicle: 'truck', }, ]; return (
Worked examples
From task to the right key on the hook.
{rows.map((r, i) => (
{r.task}
{r.note && (
)} {r.note && (
{r.note}
)}
Hardness
{r.h}
Frequency
{r.v}
Pick
{r.pick}
))}
); } function HabitsAll({ page, total }) { const habits = [ ['01', 'Match the model to the task.', 'Default to Sonnet. Drop to Haiku for obvious things. Reach for Opus only when you need it.'], ['02', 'Start fresh chats for new topics.', 'Don\u2019t let one thread balloon to 50 turns across three unrelated things.'], ['03', 'Use Projects to scope context.', 'Pinned files only load when referenced — cheaper than re-attaching.'], ['04', 'In Cowork, keep folders lean.', 'Every file counts toward context. Archive what a task doesn\u2019t need.'], ['05', 'Load tools only when needed.', 'Settings → Capabilities. The other setting pre-loads every connector into every chat.'], ['06', 'Right-size scheduled tasks.', 'Daily runs rarely need Opus. Haiku or Sonnet saves enormous usage over a week.'], ['07', 'Don\u2019t attach what you don\u2019t need.', 'Every file uploaded stays in context for the rest of the conversation.'], ]; return (
Seven ways to preserve tokens
Small habits. Much longer weekly ceiling.
{habits.map(([n, h, b], i) => (
{n}
{h}
{b}
))}
); } function DecisionRecap({ page, total }) { return (
Closing · the decision rule
{String(page).padStart(2, '0')} / {String(total).padStart(2, '0')}
If you remember one slide
{[ { kind: 'bike', name: 'Haiku', line: 'fast and cheap.' }, { kind: 'sedan', name: 'Sonnet', line: 'balanced.' }, { kind: 'truck', name: 'Opus', line: 'for the hard stuff.' }, ].map((c, i) => (
{c.name}
is {c.line}
))}
Match the model to the task, and the weekly limit rarely bites.
); } function TryThisWeek({ page, total }) { const challenges = [ { tag: 'Monday', line: 'Audit your last five chats. Which should have been on a smaller model?' }, { tag: 'Wednesday', line: 'Start a fresh chat the moment one feels slow. Notice the difference.' }, { tag: 'Friday', line: 'Pick one scheduled task. If it\u2019s on Opus, drop it to Sonnet.' }, ]; return (
Try it this week
Three small moves. One full week.
{challenges.map((c, i) => (
{c.tag}
{c.line}
))}
Come back next session with one number: how much of your weekly budget did you use?
); } function Landmark({ page, total }) { const lines = [ 'Haiku is fast and cheap. Sonnet is balanced. Opus is for the hard stuff.', 'A token is roughly ¾ of a word.', 'Context is everything Claude can see at once.', 'Every turn re-sends the whole conversation.', 'Long chats get exponentially expensive.', 'If a chat feels slow, start a new one.', 'Scheduled tasks should almost always run on Haiku or Sonnet.', 'Tokens are finite. The weekly limit is real.', ]; return (
Landmark language
The phrases we say — and keep saying.
    {lines.map((l, i) => (
  • {l}
    • ))}
); } function EndSlide() { return (
End of session
Questions?
Claude · Models & Tokens · Foundations 02
); } Object.assign(window, { WeeklyBudget, CostDrivers, UsageTab, TwoQuestions, WorkedExamples, HabitsAll, DecisionRecap, TryThisWeek, Landmark, EndSlide, });