Your usage ran out by lunchtime for the third day running. Nothing about your plan has changed. This week you started using Claude in your terminal alongside the web app, and you have been feeding whole 200-page manuals in rather than the extracts you used before. What is the most complete explanation?

The principle — when an allowance runs out sooner, there are only two things that can have moved: how much you are asking for, and how many places you are asking from. Diagnosing well means checking both rather than stopping at the first plausible cause.

Why the key is correct — two documented mechanisms are in play and they multiply rather than compete. Anthropic states that usage of all the different Claude product surfaces, naming claude.ai, Claude Code and Claude Desktop, counts toward the same usage limit — so adding terminal work does not open a second budget, it spends the existing one faster. Separately, usage is affected by the length and complexity of your conversations, the features used, the model chosen and the effort level. Swapping extracts for 200-page manuals raises length and complexity together. Either change alone would bring the limit forward; both at once explain a lunchtime cut-off comfortably, and the fix follows directly — go back to extracts for routine work, and reserve whole manuals for when the whole manual is genuinely the question.

Why the others are wrong — the two explanations that isolate a single cause each ignore a documented factor, and the surfaces-are-separate reading contradicts the shared-limit rule outright. Nothing suggests differential metering between surfaces. And three consistent days is what a changed habit looks like, not what a fault looks like; reporting it delays the change that would actually help.

Remember this — one allowance across every surface, spent faster by bigger requests. Check both before concluding anything.

Sources — Anthropic's usage and length limits article.