How Catan Lens works

Catan Lens answers one question: where should I put my first two settlements? It scores every legal corner with a small set of rules that strong players agree on and explains the result in plain sentences. The same position and settings give the same answer in the same app version. Scoring runs in your browser, with no account or analysis service. Links contain the board data you open or share.

If you opened this page from the app in this tab, Back to the app returns to that browser entry, keeping its Drill return board too, even after refreshing this page. A direct help link or a new tab opens the board and placements carried in its link.

Both videos are recorded from the app itself and narrated by a computer voice.

Catan Lens: plan your opening

Video could not be loaded or played. Try the transcript below, or use the written guide further down.

How Catan Lens works: the scoring, your opening, Drill and Your board

Video could not be loaded or played. Try the transcript below, or use the written guide further down.

Use it at a table

1. Choose Players and Your seat in Table settings before entering tiles.
2. Open Your board and Start blank. Paint resources, set numbers, then finish the harbours or explicitly exclude their effects.
3. Mark opponents in placement order. When your turn comes, choose a corner and use Keep for your own settlement.
4. Tap the side of that settlement where your starting road went. With Show best on, the card says how that direction compares with the others.
5. Compare second-settlement candidates before committing. On a base-game table, check what you cannot produce and which first build needs the fewest new cards.

For practice, choose Drill, set your table and commit before seeing the answer. Replay a cleared level to try that situation again. For a group review, Copy link from Your board saves explicit tiles, harbours and placement order. Generated links use a seed and the current setup rules; scoring can change as the app is updated.

Weak connection? Open the app once while connected. Most browsers then keep an offline copy, so the board page and this help page open and refresh without a connection. Videos still need one.

Install it on your phone. On Android, open the app in Chrome, then choose Install app or Add to Home screen from the browser menu. On an iPhone or iPad, open it in Safari, tap Share, then Add to Home Screen. It opens full screen like an app, with the same board links and no account.

If the board cannot finish loading, check your connection and use Reload this board. It retries with your current board link and keeps this tab's Drill return point. The loading screen also keeps How it works available.

Enter, recover and practise

Settlement advice waits until every hex has a resource or desert and every resource hex has a number token. Then finish the standard harbour set, choose Ignore harbour effects, or confirm a custom set with Use these … harbours. Excluding harbours keeps your entered coast intact while removing harbour effects from every recommendation. Overlapping ports must be repaired or excluded. Removing a tile or token hides advice until it is repaired; your placed pieces stay on the board.

Read a board from a picture. In Your board, choose Read from a picture and pick a screenshot or a photo taken from straight above. Tap the number tokens on the four highlighted corner tiles; Catan Lens then guesses every tile and number, and nothing changes until you choose Use these tiles. Tiles it was unsure about are outlined so you can check them. Harbours are not read: add them with Ports. The picture never leaves your device. On real photos the tile types read well but numbers are often unsure, so compare every tile with your table.

Opening Your board by keyboard focuses the first tool. Activate a paint, Numbers or Ports with Enter or Space to focus the first hex or coastal edge. Tab moves through the board; Enter or Space edits it, and Escape returns to the selected tool. The number picker focuses the current token; Tab to another and press Enter to set it. Escape returns to the hex, then another Escape returns to Numbers. Choosing a blank or desert hex brings an explanation into view instead of number choices; Escape returns to that hex. Your place stays focused after each edit. Choosing Players or Your seat keeps focus on that button too, so Tab continues from your choice.

Changing between the 19- and 30-hex layouts starts a blank board of the new shape. Restore previous board or the browser's Back button recovers the board you had before, including its placements, even after a refresh. Start blank, Random board and leaving Your board for Explore keep the same recovery in this tab. Recovery focuses the restored panel heading and brings it into view. A copied link shares only the current board.

In Explore, New board, a changed seed and the balanced-numbers switch also keep the previous board. Clear removes settlement marks and offers Restore cleared pieces, with the original placement order. Clearing an empty board or entering the same seed keeps the useful recovery point.

If the browser refuses a link update, editing still works and Copy link includes the board shown here. Keep the tab open, wait a moment and use Retry link update before refreshing. A board replacement or Drill mode change is not applied if its recovery point cannot be saved.

An unreadable board link shows an error and keeps its original data in the address bar. Advice stays hidden until you recover the board or explicitly start again. If the saved board matches the other board size, the app offers the matching player counts so you can recover its tiles, harbours and settlements.

In Drill, Tab to an unlocked level and press Enter or Space to replay it. Locked levels become available as you clear the ladder.

Scores show hundredths, and score percentages round down so only an exact tie displays 100%. A negative score or a best score of zero or below uses a point gap instead. Drill misses also show the point gap; a close model score can still describe a good real-game choice. The exercise checks the app's ranking, and score shares are not winning chances.

Drill progress is stored in this browser, separately for each table. If reading or saving fails, a warning offers Retry saving. You can keep practising in this tab, but unsaved progress may be lost if you reload or close it. Retrying preserves other tables' saved progress.

A shared seed or an unsuccessful search can put a Drill board outside its usual difficulty band. A notice follows that board through refreshes without revealing the answer; the pass rule stays the same. New board tries again.

Leaving Drill returns to your previous board, settings, placements and placement tool, including after a refresh during setup or an exercise. That return point stays with this tab's current history entry. A copied link shares the Drill table setup or exercise; it does not include the board you were exploring before it.

Your plan, and a look-ahead that was withdrawn

Your plan. By seat follows the classic rule: early seats chase ore, wheat and sheep for cities and development cards, and later seats lean on wood, brick and harbours. Choose Cities or Roads under Your plan in Table settings to rank your own corners for that plan instead. The other seats keep the seat rule in every projection. Cities & Knights keeps its own weighting.

Look-ahead (withdrawn). A second ranking, tested on 12 October 2026, valued each corner by the whole opening it leads to, with the other seats picking by the classic model. It won fewer games than the classic ranking in the test below, so the app no longer offers it. A link that asks for it opens the classic ranking and says so.

Understand your opening

After your second settlement, Your opening shows average cards per 36 rolls, the chance of any income and the five-resource breakdown. Repeated numbers can pay several cards together without improving that chance. These averages assume available resources and omit the robber and trades; actual rolls vary.

Missing-resource notes connect your production to the things you can build. The summary also lists your starting hand and owned-harbour trade rates. Plan your first builds shows the extra cards each purchase would need from that hand; each row is a separate option, and building still needs a legal place or purchase on your turn. Which rolls pay? lists every dice total.

When previewing a second corner, open What this opening gives you to see the same figures before Keep. This preview stays hidden during blind practice and unanswered drills. Confirming a choice focuses its result and brings it into view on a phone.

Ignoring harbour effects also removes their trade rates from the summary. Cities & Knights figures are labelled as a two-settlement comparison: the app does not model its starting city or commodities, so it withholds base-game income and build planning in that mode.

Try it: same corners, different starting hand

These base-game examples use the same two locations in opposite placement orders. Both average 22 cards per 36 rolls, but collect on only 14 of the 36 dice outcomes. Repeated numbers pay several cards together. Neither opening produces wood.

Open either example and expand Plan your first builds under Your opening. Starting cards come from the second settlement, so its order changes your immediate options even though production stays the same. Each row assumes that starting hand before rolls or trades; a purchase still needs your turn. In a real draft, mark opponents first to check which locations remain available.

The rules in one table

Every rule traces to a published source. The source table, the exact weights and what was left out are in RESEARCH.md; the full explanation is in HOW-IT-WORKS.md.

Projected placements shows this model's choices for each unmarked turn. People may choose differently. Road advice considers the remaining setup; if its target is also your projected second settlement, the card explains that the site would be occupied when setup ends. When no two-road destination is legal, the card states that limit; the hint does not check longer routes. Show best reveals road hints in the map and verdicts.

Your starting road. Placing it is optional. After you keep a settlement, click or tap one of its sides (or Tab to a side and press Enter); choose another side to move it, and Undo puts it back. The road is judged as of that settlement's turn, so later placements do not change its grade: Great for the best direction, Fair when the best open corner two roads that way scores at least 90% of the best direction's (a teaching convention the author chose, not a measured cutoff), Weak otherwise. If no direction has an open settlement site two roads away, there is no grade. Opponents' roads are not entered, and longer routes are not checked. With Show best off, the grade appears only once both of your settlements are down, and the reasons wait for Show best. Drill practises settlements only.

Does following it win games?

The classic ranking is the app's ranking. An experimental look-ahead ranking lost to it in the newest test below and was withdrawn. The classic policy counts legal inland approaches to harbours, excludes trading access you already own, and rounds the same physical board consistently. Each earlier test keeps its own frozen policy. No weights were tuned on these outcomes.

Catanatron's value-function bot plays every seat. One seat's opening is chosen by each strategy, then it plays on like everyone else. Strategies face the same boards and seat orders. An evenly matched seat would win 25%.

Loading benchmark results…

Look-ahead ranking · 12 October 2026

This test finished 32,000 games on 8,000 fresh boards. It compares an experimental look-ahead ranking with the classic ranking, the bot's own opening and taking the most pips. Every policy was fixed before its games began.

The look-ahead won fewer games than the classic ranking: −1.3 percentage points (95% interval: −2.1 to −0.5; exact p = 0.0047). Against the bot's own opening the difference was −1.3 percentage points (95% interval: −2.7 to +0.1; exact p = 0.046), not a significant difference either way. The exact test alone would call it significant, but the interval includes zero, so under the plan it is not. The look-ahead is withdrawn from the page and kept only for study, as the plan required; the owner makes the final call.

Complete harbour reachability · 9 October 2026

This separate test finished 24,000 games on 8,000 fresh boards. It compares the correction for previously missed inland harbour routes with the policy immediately before it. Both versions were fixed before their games began.

The correction had one more win: 1,981 versus 1,980 out of 8,000. The difference is +0.0125 percentage points, with an approximate 95% interval from −0.77 to +0.80. This does not establish an advantage or prove equal performance. Of the matched games, 488 changed from a loss to a win and 487 from a win to a loss.

Against the bot's own opening, the corrected policy's difference was −0.425 percentage points (95% interval: −1.72 to +0.87). That comparison also remains uncertain. The fix follows legal harbour access; its correctness does not depend on winning more simulated games.

Owned-harbour correction · 9 October 2026

This separate test finished 24,000 games on another 8,000 fresh boards. Both policies were fixed before their games began. The correction removes credit for trading access you already have; it was not chosen to improve these results.

The correction had four fewer wins than the older policy: 1,972 versus 1,976 out of 8,000. That is −0.05 percentage points, with the planned approximate 95% interval from −0.098 to −0.002. Only four win/loss pairs changed, all against the correction, so the narrow estimate rests on very few changed outcomes. This test shows no win-rate gain from the correction and does not prove equal performance.

Correction, 12 October 2026: that interval excludes zero only because so few games changed. An exact test on the four changed games gives p = 0.125, so the two policies show no significant difference.

Against the bot's own opening, the corrected policy's difference was −0.16 percentage points (95% interval: −1.67 to +1.35). That comparison remains uncertain.

Earlier policy on fresh boards · 9 October 2026

This test used 40,000 complete games on fresh board and bot tie-breaking seeds. The scoring policy was fixed before the test; these outcomes were not used to tune it.

Across matched runs, Lens's advantage over “most pips” was 7.9 percentage points, with a 95% interval from 6.5 to 9.3 points. Against the bot's own opening, the difference was −0.2 points, with an interval from −1.8 to +1.3. That comparison shows no clear advantage and does not prove equal performance.

These results cover bot play in the four-player base game. They describe the frozen policy tested in that setting; they do not predict your personal win rate or validate three-, five- or six-player games, Cities & Knights, or play against people.

Intervals use the variation across 20 runs, each with its own bot tie-breaking seed. See the benchmark methods and records for the raw games, paired comparisons and reproduction instructions.