Reviewing a TAMNA game — reading the game log and the replay after you lose

Start with what is even reviewable. Only games that ran to a finish become material. Those are the ones listed on My record, and those are the ones that get a replay link. A room where everybody stopped moving is swept away automatically once more than thirty minutes pass with no progress at all, and the record of that game goes with it. Rooms that reached a result are excluded from that sweep, so what you are left with is exactly the games you finished.

Leaving mid-game is harsher still. The seat you vacate is converted to a bot on the spot, and your connection to that room is cut. If the remaining players carry the game to a finish, it still shows up as a completed game in your list. But when you try to open the replay, you are no longer part of that room, so it will not open. The result survives; the content does not.

One practical conclusion follows. The moment you feel "this one is lost" is the moment the game holds the most information — everything that broke is still sitting on the board. Quitting at that moment throws away the richest material you had. Staying seated through a loss is not a matter of etiquette; it is self-interest, paid out in the next game.

Know where the entrance is, too. In the lobby, next to your name, there is a line like "3 wins / 12 games · 25% win rate", and that line is the link to My record. If you have not finished a single game, the line does not appear at all. My record has columns for Date, Room, Result and Players, and every row carries a "🎬 Replay" link. Every review starts from that table.

And one thing that is easy to forget: the table only lists games you played in. Other people's games never appear. Reviewing is, by design, reading your own games — not browsing someone else's. The final section returns to that boundary.

During play, the Game log panel stacks up one line per event. The shape of a line is fixed: an icon, the name of whoever acted, and a short description. The descriptions come from a fixed set of phrasings, so with a little practice you stop reading words and start reading shapes.

These are the lines you will actually see: "🏠 Placed starting settlement & road", "🎲 Dice 4+3 = 7 · 7! Dokkaebi", "🗑️ Discarded 5 card(s)", "👹 Moved the Dokkaebi · stole 1 card", "🏙️ Built a city (+2 VP)", "🔄 Trade: Basalt → Tangerine", "🤝 Trade offer: Clay1 → Tangerine1", "🃏 Bought a dev card", "⚔️ Played a General card · moved the Dokkaebi", "🌾 Year of Plenty · got Tangerine1 Basalt1", "🪙 Monopoly card · all Basalt", "⏭️ Ended the turn". Resource movements are written as a name glued to a count, like "Clay2 Basalt1", so you can recount the exact quantities afterwards.

Under a dice line sits a row of cards showing who received what on that roll. This is the strongest thing in the log: it lets you say who is eating on this board as a measurement rather than a guess. You do not need to stare at the board numbers and do probability in your head to describe an opponent's income — scroll back and count how many times their name appears under those rolls.

Information about the victim of a steal is deliberately fenced off. Only when you are the thief or the victim does the line name the other player — "Stole from ◯◯" — with the stolen card shown alongside. From any uninvolved seat, the same event reads only as "👹 Moved the Dokkaebi · stole 1 card". Other people's hands are inferred, never observed. What you infer from is their "🤝 Trade offer" lines: what somebody asks for is public information about what they are short of.

The tabs above the log matter as well. Beside All there is a tab per seat, so you can isolate one player. When somebody feels like they are running away with it, switch to their tab and follow only their building lines; it usually takes a handful of rows to see which turn they accelerated on. Note that the log holds roughly the last eighty lines, so the opening of a long game scrolls away. To look at the opening, you need the replay, not the live log.

The replay rewinds the board. It does not reconstruct the whole game, and confusing the two is how reviews go in circles. What comes back exactly is where settlements, cities and roads were built and which tile the Dokkaebi sits on; those are recorded deterministically. What does not come back is who held which resources and how many, what was drawn from the development deck, and which card a steal took.

So the division of labour is clean. Questions about space — where did I build, where did I get blocked — are answered by the replay. Questions about hands and timing — why could I not build — are answered by the game log. When you cannot remember why you lost, the temptation is to scrub the replay over and over, but with resources invisible the answer is simply not in there.

The player itself is plain. One slider, a "◀ Prev" / "▶ Play" / "Next ▶" set of buttons, and playback speeds of 0.5×, 1× and 2×. Your position shows as something like "12 / 148" — current step over total steps. Autoplay advances roughly every 0.65 seconds, so 2× halves that. The board is not clickable; it is a board for looking at.

Step 0 is the board with nothing built on it and the Dokkaebi parked on the desert. Tile resources, numbers and Naru harbours never change during a game, so step 0 is literally the board you were looking at when you chose your first settlement. In a four-player game, steps 1 through 8 are the opening placements, with each settlement and its road recorded as a single entry. The order runs forward once and then back in reverse, so the last four placements mirror the first four. Only the second settlement pays out the adjacent tiles the moment it goes down.

There is exactly one place where the replay shows less than the live log did. In a replay, the name of a steal victim is never displayed — even for a move where you were the thief, the line reads only "👹 Moved the Dokkaebi · stole 1 card". If you want to know who a card came from, that was information available only while the game was live. For review purposes, treat that as a reminder that some things can only be collected in the moment.

Replaying every step with equal attention is a habit nobody keeps. Fix three places instead: just after the opening placements, the first Dokkaebi, and the last three turns.

First, just after the opening (step 8 in a four-player game). Do not grade the placement on feel — grade it against that game's log. If bank "🔄 Trade" lines show up repeatedly, you were buying missing resources at a bad price all game, which narrows the cause to either the placement itself or a harbour you failed to claim. If instead there are almost no trade lines and also very little building, the placement probably was not the problem — you were sitting on cards. Two identical losses, two opposite prescriptions.

Second, the first Dokkaebi. Jump to the step where "· 7! Dokkaebi" first appears and advance a handful of steps. Look at two things: whether it landed on your main production tile, and how many steps it stayed there. Then count how many lines are yours during that stretch. That separates "I was blocked" from "I stalled". The Dokkaebi only moves on somebody else's turn, so if you neither traded nor built while it sat there, half the damage is your own.

Third, the last three turns. Endings always look abrupt. Step back three "⏭️ Ended the turn" markers from the winner's final one and follow only their lines. A run of "🃏 Bought a dev card" suggests points you could not see were accumulating; a run of "🏙️ Built a city (+2 VP)" means they were simply jumping two at a time. Which one it was decides what you watch for against that opponent next time.

All three places share a single question: with the information you had at that moment, could you have chosen differently? If you could not, that was variance, not a mistake. Reviews turn painful when you judge past moves with knowledge of the result — that produces regret rather than learning. A move made with too little information is not a fault to punish; it is a prompt to find a way to have more information next time.

At the top of My record sits a line like "3 wins / 12 games · 25% win rate". This is where most people misread the screen. The denominator of that rate is not the number of rows below it. The list stops at your thirty most recent games, but the summary counts every completed game you have ever played, and the percentage is that value rounded to a whole number.

That structure has a practical consequence. The more games you accumulate, the less the win rate responds to anything recent. Losses from your first week stay in the denominator forever, so a strong current run barely moves the number. Using the win rate as your improvement metric therefore makes improvement invisible, which is a good way to quit caring. If you want to see progress, count the top ten rows of the list by hand — it is faster and more honest.

The summary also mixes bot games and games against people at equal weight. This is where the Players column earns its place: it shows how many humans were in that game. A 1 means a bot game; 2 or more means people were involved. Split the same thirty rows on that column and recount, and you get two separate win rates instead of one blended figure. Which of the two you actually want to raise is a decision the screen cannot make for you.

The Date and Room columns read too. Dates clustered on a single day mean a session, and watching how results move across that session tells you at which game number your concentration goes. Repeated room names mean repeated opponents. If losses pile up in the back half of a session, the fix is rest, not tactics.

Finally, going from a row into "🎬 Replay" is the standard route for a review. Skimming the Result column and closing the page wastes the table. Pick one loss, look only at the three places from the previous section, and close it. A single game takes a few minutes at most.

At the top of the replay there is a "🔗 Share this scene" button, which produces a link to the moment you are on. What matters is what actually travels. The link opens a public scene card carrying two things: the icon for that move and its description text, capped at eighty characters. No board, no full log, nobody's hand. The social preview image is a single emoji.

The card is headed "🎲 A moment from TAMNA — (the description of that move)" and carries a "Watch this scene" entrance below it, but only someone who was in that room can open it. A person who merely received the link and clicks it does not get to see the game. Sharing is for bragging and for keepsakes; it is not a way to show your game to other people.

That boundary shapes how you review with others. If you want somebody to look at a game with you, that somebody has to have been at the table. With a friend who was, you can each open the same replay and talk in step numbers — the numbering is shared, so "look at step 62" is enough. With anybody who was not there, you have to describe the situation in words, because the board itself will not travel to them.

Close every review the same way. Pick one of the three places, and decide on exactly one behaviour to change in the next game. "The turn after the Dokkaebi lands on me, I open with a trade before anything else" works. So does "my second settlement gets chosen by the resource I am missing, not by the number." Change two things and the next result cannot tell you which one mattered, and you will end up running the same review again.

When that next game ends, check only one thing in the log: did the line you were aiming for appear? If it did, the review succeeded regardless of whether you won. If it did not, the thing to fix was never the tactic — it was your trigger for noticing the moment. Which is, in the end, why it is worth staying in your seat through a loss.