The first three minutes and the first weeks: the introduction, the narrator and the achievements

The first minutes of a game decide whether a player comes back. According to a GameAnalytics report (data from 2024), on mobile, in the best quarter of games, about 27% of players return on the second day, and after a week the median is about 3.4 to 3.9% (the details are in the article When the game was finished). The introduction of the original game is not mentioned in the sources used. Audamir has ten screens, and if a newcomer saw them all at once, they would not know where to click.

This article is about what was done about that, and about what is done with the player after the first week.

The first screen: a die that asks

A player who has played nothing does not see ten tiles. They see the welcome card from the Avatar^[1], the die that accompanies the game: Show me how to play or Skip – I know how to play. Whoever skips it never sees it again. Whoever lets themselves be guided gets six short steps on tiny maps (about three minutes in all), which are played on the core of the real game (with real or lab dice, as in the puzzles): the attack, the odds, what happens after a win, the order of attacks, the end of the turn and reinforcements, and finally the first real game against the Weak AI.

The welcome card of the game, "Welcome to Audamir!", at the first start: a die, a short text, the button Show me how to play and the option Skip – I know how to play — A fresh profile sees a card that offers the introduction instead of the tiles; the option to skip it is under the button that starts it.

Every step can be skipped or repeated, accepts only the clicks that belong to it, and the next tile pulses. Nothing is written to the profile, except that the introduction took place.

A lesson that loses on purpose

The second lesson is the most interesting. On the map 03 13 03 (your stack of three dice against the opponent's stack of three dice) the player learns from the number on the border that the chance is 45.4%. And then that first attack is lost. The roll is fixed in advance for this lesson (in the code a fixed "roll salt", which the tool lessons.js checks for every lesson).

The second lesson of the introduction, The odds: three tiles in a row with three dice on each, the chance of winning of 45% on the border and, under the map, the status line with an attack of three dice on three — The second lesson of the introduction: three dice against three have a chance of only 45%, because on a tie the defender wins.

It is not a swindle, it is a teaching aid. The lesson is called The odds and it teaches one thing that no text can say: a chance is not a certainty. Only the second attack (the second stack) wins. The first loss is fixed in advance in the lesson, but what it teaches holds in the real game too: even fair dice sometimes lose.

The first game: it has to be winnable

Lesson number six is the first real game. It is a game with 19 tiles (10 against 9) and the Weak AI, recorded as a normal game in Free play. A win ends it. Numbers from the measurement: an AI of level 3 wins such a game in 93 to 96% of cases, the weakest AI in 69%, and the game lasts 7 to 11 rounds. It is a game that a newcomer wins with great probability, but not with certainty.

The map of the first game of the introduction as the game draws it: 19 tiles, 10 of them yours (red, 31 dice in all) and 9 the opponent's (blue, 28 dice in all); the number is the number of dice on a tile. The introduction uses a fixed game number (FIRST1).

A screen that is revealed

Even after the introduction there are ten tiles. That is why they are revealed gradually:

at the start only Principalities^[2], Minigames and the Daily challenge are visible,
Puzzle Labs and the Campaign appear after two minigames or five games,
the Collection after the first Golden Dice,
Statistics, Achievements and Free play after the first finished game,
a new tile has a NEW badge until it is opened.

The profiles of players who played before this change see everything: nobody has anything taken away. The rule that guards this is in the list of the shared rules of the project: a new feature behind an unlock must be visible to a newcomer (what it is and how to get it), and the migration must retroactively grant the claim to those who met the condition earlier. The companions of the Campaign once did not meet this: according to the development records, a campaign that the player had finished before they were introduced did not show the companions anywhere. Since then it is a rule.

To this belong one-time tips: when a player plays for the first time with a terrain, a bonus or a situation that has not occurred yet, a card appears with an explanation and a button "Got it". On a large window it lies above the map as a notice, on a small one the map goes around it (if the map moved under the hand during fast clicking, the player would click a tile that is no longer there; this was measured in the browser: the top left corner of the map moved from (196, 16) to (280, 164) pixels).

A narrator that changes nothing

The Avatar is not just a die in the corner of the screen. It is a narrator: it comments on the game in a bubble.

Bad odds. I'll pretend I didn't see them. (before a risky attack)

Hey, that was mine! (when you lose a tile)

The dice hate me today. It's personal. (after a series of unlucky rolls)

Even I can't roll sixes forever. (after a loss)

In the first version there were 52 lines in every language in 18 situations, today there are over a hundred (the start of a game, a risky or a safe attack, the loss of a tile, the elimination of an opponent, the closeness of a win or a loss, the luck of the dice, the end of the game, an achievement, a new level and tips), in the first person, in a dry tone, at most 90 characters. In the settings it can be switched off, kept brief, turned on fully (Chatty) or left to tips only.

More interesting is the technique. A line is not chosen at random, but by a fingerprint of the game number (seed) and an event counter. This has two consequences:

A game with the narrator and without it is the same game. The narrator never reaches for the random generators of the game, and a test compares the snapshots of both games bit by bit. The narrator is only an observer.
The same game is commented on in the same way. That matters too, because the record of a game (the next article) can be replayed and the commentary fits.

The same game twice: on the left with the narrator, on the right without it. The rolls, the moves and the result are the same, only the bubble differs. The events are an example, not a record of a game; the lines are from the catalogue of the game.

The rules are concrete: for every event there is a probability (for example for a risky attack 50% in the Brief mode and 85% in the Chatty mode), a gap (how many events must pass before the same event speaks again) and a ban on repeating the last five lines.

And one more thing about the text. In Slovak, Czech, Serbian and Russian the past tense has a gender ("vyhral som" and "vyhrala som": "I won", masculine and feminine), and the game does not know who is playing. That is why everything is written in genderless forms (the present tense, commands, infinitives). Where that was not possible, gender groups {gender#…|…|…} were made. An independent review found eight lines that broke this rule, and they were rewritten. The rule is built into a check in a test.

Achievements: from 138 to 87

After two days of development (19 Oct) Audamir had 138 achievements. Many of them were one achievement with a different number: Veteran, General, Legend (10, 50, 200 wins), Enthusiast and Addict, the steps of the three dice, and Firm defence with Untouchable for each of the seven numbers of players (2 to 8). The brief was clear: as in World of Warcraft, one panel per family, which always shows the step you are working on.

The result is 87 panels and 272 steps. A panel shows the name of the rank with stars ("Veteran ★★"), a description, a reward and a bar "62 / 100". When a step is completed, the next one replaces it. A click opens the completed steps (the date and the reward), the steps ahead of you stay hidden. Instead of a description for every step there is one template description for the whole chain. The intention that much less would have to be translated did not come true, however: the number of texts fell by only four, because 13 new achievements and 40 names of ranks used up everything that the templates saved.

Each square is one item: the flat list of 19 Oct 2026 had 138 achievements, after the change there are 87 panels with 272 steps. Example: from the achievements Veteran, General and Legend there is one panel with the name of a rank, a description and a progress bar.

Why chains and not 138 single achievements? The brief spoke of a shorter and clearer list and of less text to translate. To that belongs my own reasoning: a panel that shows only the next step tells the player what to do now, and does not burden them with the rest.

Numbers worth noticing

The chain of wins ends at 10,000. At 50,000 wins a player, at ten minutes a game and half of the games won, would need about 16,000 hours. It was left out with a note that it can be added when an online game exists.
Untouchable (win without losing a single tile) was originally meant to be earned three times or ten times. The AI of level 3, however, did not win even once without losing a single tile in 56,000 games. When nobody has reached the goal, the counts have nothing to rest on. That is why for every number of players (2 to 8) a ladder of allowed losses was made, built from the measured distribution. The first step is reached by 21 to 27% of the games of the AI of level 3, the second by 4 to 6%, the third by 0.4 to 0.6% and the fourth (no tile lost) never. The fourth step is therefore meant for people, not for the AI.
The Unclimbed Wall (win a boss mission with Brynja in the party and with as few tiles lost as possible): the original condition (at most two losses) was unattainable for the AI, so it became a ladder of 40, 20, 10 and 5 losses. According to the measurement from the documentation the AI reaches a median of 38 to 115 losses in boss missions, and at most five losses in only 0.3% of games against one of the bosses.
The sum of the rewards rose from 18,020 Golden Dice to 48,067. It is a deliberate change: the table of rewards is in the documentation, and the tool achievements.js --check guards that the sum stays in the band of 40,000 to 56,000.

Secrets that are revealed slowly

The most interesting decision is the secret achievements. They are not encrypted: their texts are in the catalogues and their conditions in the code. Encryption would cost more than it would bring. Instead there is slow revealing: the secrets are in a fixed order and one is revealed every seven days from the day of the first win. It is counted from dates, not from visits: a month offline reveals four at once. A revealed secret moves to the category Feats of Strength, the others are only a line ("Secrets still hidden: 7. The next is revealed in 5 days.").

Revealing changes only the visibility. Unlocking, the counters and the rewards do not depend on it.

It is a soft reason to come back, without punishment: if the player does not return, nothing is lost. It is exactly the compromise that is written about in the article on attention: the industry relies on the daily login, and the game tries to have its effect without its pressure. Whether that succeeds, only the players will show.

A promise: nothing is lost and nothing is paid twice

The change from 138 achievements to 87 chains is a change of what the player already has saved. The rule is: nothing unlocked may be lost, nothing may be paid out twice and nothing may be taken back. The old IDs became the IDs of steps wherever a step kept its goal. Where a goal moved (200 wins became 100), a migration table holds the paid state. It was verified with the data of the old code (profiles from git archive), because hand-written data once did not catch four achievements that were paid out twice (more in the article How to test a game we will never stop changing).

What I take from it

The first three minutes are design, not text. Teach with what the player does, not with what they read.
Let the player lose in the introduction. One attack lost on purpose teaches more about chance than a paragraph.
Reveal gradually. Ten tiles at once is a bad first screen.
The narrator must not change the game. A test that guards it is cheaper than an argument about whether the dice lie.
A panel with one step is better than a list. The player needs to know what comes next.
A secret does not have to be locked. It is enough that it arrives drop by drop.

Sources and notes

The introduction, the lessons, the revealing of the tiles and the narrator: docs/DESIGN.md §23 and §30; the introduction in the ROADMAP, chapter 3. The lessons are verified with the tool node tools/lessons.js --check.
The achievements: docs/ROADMAP.md, chapter 18; docs/DESIGN.md §33; docs/TUNING.md §20 and §25; the tools achievements.js and untouchable.js. The ladder of losses: a measurement with the AI of level 3, 8,000 games per step (according to the documentation).
Pictures: the welcome card and the second lesson of the introduction are screenshots of the game (assets/shots/, made on 4 November 2026, the welcome card shows version 0.108.3; the same ones are in the wiki on the page The first game). The map of the first game is drawn by the real renderer of the game (Scenario.svg, the fixed game number FIRST1, 6 November 2026). The scheme of the narrator and the picture of the change of achievements are drawn by hand in tools/charts.js, they are not a measurement: the events in the scheme are an example, the lines are from the catalogue of the game (js/i18n/en.js, 6 November 2026), the numbers 138, 87 and 272 from docs/DESIGN.md §33.3 of the game (the table of totals).
Mobile retention: GameAnalytics, benchmarks 2025 (data from 2024), via GameDevReports.