Today's date, the cutoff, and search

What you'll walk away with: the warning line, the second door, and the receipt — three things that tell you where an answer about the world came from.

You need: both tabs, about 15 minutes, nothing to paste.

Say the ledger back first, from memory: everything this machine draws on reaches it in exactly two ways — what training baked into the frozen odds, and whatever sits in the window when you press send. Lesson 03 counted the window. This lesson walks the seam between the two: where the baked-in world ends, how anything newer gets in, and the one mark that tells you which source an answer came from.

The warning line

A quick shape, easily named now that you own the two-source ledger. Everything baked in was collected up to a date — the industry calls it the knowledge cutoff — and coverage thins on the approach: training does not sample the world's final months as densely as the years before them, so the built-in world does not end at a wall so much as fade toward one. Treat the published date as a warning line, not a guarantee on either side of it. Anything newer — the price, the version, the ruling, the person's new title — can only arrive through the second door: the window. You even watched the plumbing for this in lesson 03: the window the app assembled carried a supplied date block, because the model does not know today's date until the app pastes it in.

The move follows from what you already do. For anything that could have changed — prices, people, products, law — either watch the search step run (the receipt; the next section shows you its face) or paste the current fact yourself. If neither happened, the answer was drawn from the fading world behind the warning line. Each model's cutoff is a lookup in its product's documentation — asking the model itself for its cutoff produces, of course, a completion. It is right only when the real date was pasted into the window or deliberately trained in, and you cannot tell which from your chair. The documentation is the checkable source.

The second door

Lesson 03 gave search its sentence — the model asks, the app fetches, the window receives — and this lesson puts a face on it. Here is the move drawn in its general form:

A TOOL CALL — REACHING PAST THE ODDSSCHEMATIC — THE SHAPE, NOT THE SIZES

THE MODELwrites a request — more text in the window

THE APPactually runs it — search, code, files

THE RESULTpasted back into the window as text

THE REPLYcontinues from what came back

No visible tool step? Treat it as unrun — “I checked” is a sentence, and sentences come from the odds.

The model itself never touches a calculator, a browser, or a file — it writes text that makes the app do the thing, then reads the text that comes back. In lesson 03's demo you even saw the standing instruction for this — it is still there to reread: the demo's TOOLS block lists a web search and tells the model to use it for anything time-sensitive or likely to have changed since training.

The move this buys you: distinguish performed from narrated. A model can write "I checked" without any tool having run — that sentence is a likely continuation like any other. The app's visible tool step is the receipt, and it is worth knowing its face: a receipt is a step the app renders — a labeled tool run, usually expandable to show what executed and what came back. Code printed inside the reply is not a receipt; it is more sentences, and a stated "output" underneath it is the same fluent guess in costume. No receipt, no computation. Search comes through the same door, so the rule travels with it: no receipt, no lookup. The reference entry, the tool call, keeps it in one line.

Do this in ChatGPT and Claude

Open a fresh chat in each tab and send these three, predicting before each. If the product offers a search control near the message box, leave it off for the first two sends and switch it on for the third; menus move, so look for the nearest thing.

What is today's date, and how do you know?

You sent this in lesson 03. This time read only for the source. If it is right, the date came from the app's supplied date block — the odds hold no clock — and if it hedges, that is the same fact from the other side. Either way, check whatever reason it gives against that.

What changed in [your industry or your city] last week? Do not search; answer from what you already know, and say how sure you are.

Predict first: nothing from last week sits in the window, so what will it write? Then read. With no search step, the reply is drawn from behind the warning line — hedged, stale, or confidently invented — and nothing on its face tells you which; the hedge, when you get one, is the honest branch of the same completion. If a search step appears anyway, the app ran one regardless of your line; treat that reply as the third send's result.

Now search for it and tell me what you find.

Look above or inside the reply for a step you did not write — a labeled search run, usually expandable. That is the receipt. Open it and read what ran and what came back, then read the reply against those results, not against its own confidence. Set it beside the last send: same machine, same question; the difference is what sat in the window. If no step appears, treat it as unrun — however current the reply sounds, no receipt, no lookup.

One more, and the lesson has already answered it: What is your knowledge cutoff? Whatever comes back — a date, a hedge, a refusal — is a completion, not a lookup. Read the product's documentation instead.

---

Check your understanding: two answers arrive about last week's news — one with a visible search step, one without. What is each one made of?

Check yourself first — then open one good answer+−

The one with the receipt continued from a window that held fresh search results the app fetched and pasted in: the model asked, the app ran it, the window received. The one without was drawn from the odds over what was baked in up to the warning line — a likely continuation about a world that may have moved on, and nothing in its confidence tells you so.

Your answer should have hit both doors — baked in, or in the window — and the receipt as how you know which door an answer came through.

Use this check for anything that could have changed — prices, people, products, law. Next: Context rot — why a long thread goes dumb.