The tips page is a list of individual moves. This page is the order to make them in. It assumes you already know how the ranking and colour bands work; if not, start there. The method below is what turns a hundred-guess day into a forty-guess day, and it is most valuable exactly on the days when your instincts are producing nothing — open today’s puzzle and run it alongside.
Why a method beats good guesses
Contexto is a search problem in a space you cannot see. You know two things about that space: it has a single peak, and every probe returns your rank rather than your distance. Under those conditions, the winning approach is not cleverness — it is coverage followed by disciplined descent.
The failure mode this method exists to prevent is premature convergence: finding one moderately warm word early and spending the rest of the round mining its neighbourhood, when a completely different region was warmer all along. It costs more guesses than every other mistake combined, and it feels like progress the entire time, which is why it needs a procedure rather than good intentions.
Phase 1 — the category sweep
Goal: find out which broad region the answer is in. Budget: 5–8 guesses. Exit when: one guess is at least four times better than the next best, or you have spent the budget.
Guess one word from each of these areas, in any order. They are chosen because they sit far apart in meaning-space, so their ranks are close to independent:
- A physical object — keyboard, bottle, engine
- A natural phenomenon — storm, river, fire
- A living thing — bird, tree, child
- An abstract human concept — memory, justice, fear
- A social activity — market, wedding, war
- A body or sensory word — hand, voice, taste
- A time or motion word — morning, journey, begin
You are not trying to be right. You are trying to make the space smaller. Seven ranks from seven unrelated regions is enough to tell you where the answer is not, and that is most of the work.
Write down — mentally or otherwise — the best two. You need the runner-up later.
Do not skip this because guess three came back at 400. A rank of 400 is genuinely promising, and it is also exactly the situation where a category two areas over is sitting at 60 and you will never find it. Finish the sweep. It costs four guesses and routinely saves thirty.
Phase 2 — bisect the warm region
Goal: find which part of the warm region the answer occupies. Budget: 4–8 guesses. Exit when: something lands inside the top 300.
Take your best word from the sweep and probe outward from it in different directions rather than repeatedly near it:
- Up the ladder: the category it belongs to. From river, guess water.
- Down the ladder: a specific instance. From river, guess delta.
- Sideways: a sibling. From river, guess lake.
- By function: what it does or is for. From river, guess flow.
- By property: what it is like. From river, guess wet.
- The opposite: whatever the antonym is. From river, guess drought.
Six probes, six different axes. One of them will be meaningfully better than your anchor, and that direction is your gradient.
This is the phase where the abstraction ladder matters most. Embeddings encode “how specific is this word” strongly, so moving up and down that axis produces bigger rank changes than moving sideways among synonyms. If water is much better than river, the answer is a general concept; if delta is better, it is a specific thing.
Phase 3 — converge
Goal: get inside the top 20. Budget: 6–15 guesses. Exit when: you are inside rank 20, or you plateau (see below).
Now the ranking is precise enough to trust, and the technique changes: small steps, one variable at a time. From your best green word, guess its nearest neighbours in meaning — the near-synonym, the technical term, the everyday term, the thing it is part of, the thing it is made of.
Keep two rules:
One change per guess. If your best word is night sky, do not jump to constellation — test night and sky separately first. When two ideas move at once you cannot tell which one helped.
Track the direction, not the words. What matters is not that satellite beat planet, but that specific body beat class of bodies. Keep pushing in whichever conceptual direction produced the last improvement until it stops working.
Phase 4 — the endgame
Inside the top 20, the remaining candidates are usually a handful of words that are all obviously related to each other. The trap here is guessing near-identical variants — lunar, lunary, moonlight — and getting three ranks that tell you nothing new.
Instead, ask what kind of word is missing from your list. If you have every adjective and no noun, guess the noun. If you have the technical term and not the everyday one, guess the everyday one. Most rounds end because someone finally guesses the plain, obvious, slightly-too-simple word they had been assuming was too simple to be the answer.
That instinct — it can’t be that, that’s too ordinary — is wrong more often than it is right. Contexto answers are usually common words.
When to abandon a lead
This is the decision that separates a forty-guess round from a hundred-and-twenty-guess one. Abandon a lead when both of these are true:
- You have made six or more guesses in that region.
- Your best rank in it has not improved by at least half.
When that happens, go back to your runner-up from the sweep and run Phase 2 on it instead. It feels like throwing work away. It is not: you have established that the region is bounded, which is real information, and the runner-up is now the best-supported hypothesis you have.
If you had no runner-up because you skipped the sweep, do the sweep now. It is still cheaper than continuing.
When to take a hint
The tip system is part of the game, not a confession. Use it at these three moments:
- You have a strong green you cannot improve. A tip gives you a second anchor, and two points define a direction where one point defines nothing.
- You have run two full Phase-2 passes with no green. Your model of the answer is wrong at the category level, and no amount of further guessing inside a wrong category will fix that.
- You are about to stop playing. A finished round teaches you something; an abandoned one does not.
Do not take a hint in the first ten guesses. Not for purity — because a tip early replaces the sweep, and the sweep is the part that teaches you how the space is shaped.
Afterwards, look at where the tip ranked and ask why the model considers it close. Doing that consistently is the difference between players who plateau at sixty guesses and players who settle around twenty-five. The archive is the best place to practise it, because you can reveal the answer and check your reasoning immediately, and the ten closest words on each answer page show you the neighbourhood you were trying to find.
That is the whole method. Go and run it on today’s puzzle — the first time through it will feel slow, and it will still probably beat your average.