Make a context budget before sending

WHAT YOU WILL BUILD

A context selection that fits a supplied budget while keeping the current task and constraints.

A context window is a capacity limit, not a target to fill. Budget the fixed instructions, tool definitions, current task, evidence, history, and expected output. The exercise uses supplied counts so you can inspect the selection arithmetic without a tokenizer.

FOLLOW ALONG

Work through the example.

1. 01

   Reserve fixed costs

   Start with a 1,000-token allowance, reserve 200 for output and 150 for fixed input. Calculate the remaining 650 before choosing history or evidence.

2. 02

   Select by task importance

   The sample order is constraints, current task, evidence, then old chatter. Run the script to retain the first three entries, totaling 550 tokens, and leave 100 unallocated.

3. 03

   Test an impossible budget

   Increase constraints to 800. Observe that this simple selector skips them. A real system must fail or request a shorter task when required material cannot fit; it must not silently discard required constraints.

4. 04

   Measure a real request later

   Use the selected provider's tokenizer or counting endpoint with the exact messages and tools. Recheck the total after formatting; include any provider-specific accounting for images or reasoning before making cost claims.

THE COMPLETE LAB

Run it locally.

Run this command from the folder containing your downloaded file:

node prompt-context-budget.mjs

// SaveMyToken local lab. Run with Node.js 22 or newer.
import assert from "node:assert/strict";

// Supplied token counts, not a tokenizer or measured provider usage.
const windowSize = 1000, outputReserve = 200, fixedInput = 150;
const available = windowSize - outputReserve - fixedInput;
const items = [
  { id: "constraints", tokens: 120 },
  { id: "current-task", tokens: 180 },
  { id: "evidence", tokens: 250 },
  { id: "old-chatter", tokens: 400 },
];
let used = 0;
const kept = items.filter(item => {
  if (used + item.tokens > available) return false;
  used += item.tokens;
  return true;
});
assert.ok(used + fixedInput + outputReserve <= windowSize);
assert.equal(used, 550);
console.log("Available input budget:", available);
console.log("Kept:", kept.map(item => item.id).join(", "));
console.log("Unallocated tokens:", available - used);

Expected output for the unchanged example

Available input budget: 650
Kept: constraints, current-task, evidence
Unallocated tokens: 100

The assertions also check the baseline behavior. After changing an input, predict the result and update the relevant assertion.

EXPLAIN WHAT YOU LEARNED

Interview practice

Try answering aloud before opening the reference answer. These are original learning questions, not a record of any employer's interviews.