// Second batch of slide components: tokens, context, usage, decision rule, habits, closing. function TokenDefinition({ page, total }) { return (
What a token is
A small unit of text — <br/>roughly <em style={{ color: COLORS.rust }}>¾ of a word</em>.
About four characters each. Every message you send costs tokens. Every message Claude sends back costs tokens. Both count toward your weekly limit.
A rough conversion
{[ ['"Hello world"', '2 tokens'], ['One sentence', '15 – 30'], ['Short email', '100 – 300'], ['One page', '500 – 750'], ['Ten-page doc', '5,000 – 7,500'], ['Book chapter', '20,000 – 50,000'], ].map(([a, b], i) => (
{a}
{b}
))}
); } function BothSidesCount({ page, total }) { return (
Both sides count
Input tokens. Output tokens. <br/><em style={{ color: COLORS.rust, fontStyle: 'italic' }}>Both pull from the same bucket.</em>
Input
What you send
Your prompt, attachments, project instructions, every prior turn of the chat.
Output
What Claude writes
Reply text, artifacts, code blocks. A five-page essay costs more than a one-line answer.
); } function ContextContents({ page, total }) { const items = [ 'Every message you\u2019ve sent in the current chat', 'Every message Claude has sent back', 'Any files you\u2019ve attached', 'Project-level instructions and pinned files', 'Active connectors\u2019 relevant data', 'The current CLAUDE.md skill (if you\u2019re in Cowork)', ]; return (
What context holds
Everything Claude can see <br/>at once — <em style={{ color: COLORS.rust }}>not just your prompt</em>.
    {items.map((it, i) => (
  • {String(i + 1).padStart(2, '0')} {it}
    • ))}
); } function BigNumber({ page, total }) { return (
Context · the window
{String(page).padStart(2, '0')} / {String(total).padStart(2, '0')}
The context window
200,000
tokens and up — roughly 150,000 words, the length of a short novel.
That sounds unlimited. It is not.
); } function HiddenMechanic({ page, total }) { return (
The hidden mechanic
Every turn re-sends
the entire conversation.
So Claude can “remember” what was said, the full history travels back with every message. A 20-turn chat sends the whole history 20 times. Token cost compounds.
); } function TurnComparison({ page, total }) { const turns = [ { n: 1, tokens: 100, label: 'Fresh chat', cost: 'Tiny' }, { n: 15, tokens: 1500, label: 'Mid-session', cost: 'Noticeable' }, { n: 30, tokens: 3000, label: 'Long thread', cost: 'Heavy' }, ]; const max = 3000; return (
Turn 1 vs. turn 30
The same chat gets <br/><em style={{ color: COLORS.rust }}>exponentially expensive</em>.
{turns.map((t, i) => (
Turn
{String(t.n).padStart(2, '0')}
~{t.tokens.toLocaleString()} tokens of history sent
{t.label}
{t.cost}
))}
); } function ThirtyTurnOpus({ page, total }) { return (
A worked example
{String(page).padStart(2, '0')} / {String(total).padStart(2, '0')}
A 30-turn Opus chat
can easily cost
50 – 100×
what a new, short Opus chat costs.
What drives it
{[ ['01', 'The history gets longer each turn.'], ['02', 'Opus tokens cost 10 – 20× more than Haiku.'], ['03', 'Output tokens grow with longer, more considered replies.'], ['04', 'Attached files stay in context for the life of the chat.'], ].map(([n, t], i) => (
{n}
{t}
))}
); } function SlowdownSignal({ page, total }) { return (
If the chat feels slow —
Start a new one.
Slowness is the visible tell that context has grown heavy. A fresh chat costs nothing and gives Claude room to think again.
); } Object.assign(window, { TokenDefinition, BothSidesCount, ContextContents, BigNumber, HiddenMechanic, TurnComparison, ThirtyTurnOpus, SlowdownSignal, });