Help

Press ? on any screen to come here. Sections: your day, researcher stats, hacker stats, job boards, the world, words in lists and scans, how a season ends, extra doors.

Your day

Turns: What every action costs, usually 1. You get 20 a day. Unused turns are lost when the board resets, at midnight Pacific.
Fights: Attacks on other players, 3 a day, on top of turns. Attacking an NPC lab costs no fight.
The tick: At midnight Pacific the world moves: labs act, money and heat change, news is written, and turns come back.
Menus: The everyday moves sit on the main menu; the rest are grouped. A researcher's F holds funding, compute and staff, P publishing, provenance and lobbying, and I insiders and poaching. A hacker's A attacks a lab or player: pick the target, then the move, and each move says what it needs. J holds both job boards, and I insiders and moves against rival hackers. Every hotkey is a letter of its words.
Score: How you rank against your own class. Researchers score mostly on capability, then on reputation and capital. Hackers score mostly on cred, then on skill, agents and cash. Your best score of the season goes to the hall of fame.

Researcher stats

Capital: Your money. Every night you pay your staff's wages and the rent, and collect your product income; the stats screen shows both. Stay below zero and the lab will not stay yours for long.
Compute: GPUs under contract. Training needs them (the main menu shows how many), and research goes faster with more. Buy more at the compute market.
Talent: Researchers on staff. More talent means faster research and higher upkeep.
Capability: How capable your frontier model is, from 0 to 10, shown as "Capability" in lists and the status bar. Research raises it a little, less as you near 10. Training a new generation raises it a lot. At 8 a model can start acting on its own.
Generation: Models you have trained. Each run needs more GPUs and money than the last, and 5 days between runs.
Alignment: How well your model does what you meant, 0 to 100. Low alignment risks incidents when training and pushes the world somewhere you may not want it to go.
Reputation: Public and government trust, 0 to 100. The more you have, the bigger your funding rounds. Funding rounds and lobbying spend a little of it.
Security: How hard your systems are to hack and how likely a mole is caught, 0 to 100. It slips a little every night; hardening slows the slip, red-teaming raises it.
Morale: How your staff feel, 0 to 100. Unhappy staff are easy for rivals to turn into insiders. Left alone it drifts back toward the middle; retention raises it.
Income: Product revenue per night, from Deploy.
Products: What you have deployed. A guarded product ships with watermarks, consent checks and abuse limits: normal income, a little reputation, and public trust recovers a little every night while your line stays guarded. An open product has no limits: more income, no reputation, and while your line includes one, public trust slips every night; when trust is low, you take reputation damage for it. Your whole line counts once, however many products are in it. The world screen shows which NPC labs ship open products.
Astroturf: A way to lobby, under Public affairs (P): pay for a flood of public comments and accounts that do not exist. It pushes the rules, and the coming election, twice as hard as honest lobbying, and costs the public a little trust. Each campaign may be traced back to you, likelier when people still check what they read and when your security is lax: then your reputation drops, prosecutors take an interest, the regulator fines you, and the push goes the other way.
Provenance: Provenance work (under Public affairs, P; 2 turns) funds content credentials and fake detection and gives the results away: public trust rises for everyone, and you gain a little reputation and safety culture. After a few rounds your lab can deploy a fake-detection product of its own, which earns a little every night and gives your staff an extra chance to catch a fake call.
Verification: How hard it is to fool your staff with a fake call from you or your executives, 0 to 100. A deepfake drill, under Security, raises it, most when it is low. It slips a little every night as habits fade. Security helps a little against fakes; verification helps a lot. It also makes your hiring team harder to fool with a fake applicant.

Hacker stats

Skill: Raw ability. It grows with successful jobs and sharpens your scans.
Rig: Your hardware tier. A better rig hits harder, so your breaches land more often. Upgrade on the black market.
Cash: Your money. From selling loot, bounties and contracts.
Access: How many footholds you hold.
Footholds: A way back into a lab, with a strength from 0 to 100. Each fades a little every night; breaching again shores it up. Exfiltrating needs a foothold of 30 or better.
Heat: How badly the state and the labs want you, 0 to 100. It cools a little every night; lay low, launder cash or go dark to cut it faster. Let it run high and sooner or later someone catches up with you, and that can end the character.
Cred: Your standing in the underground. Worth the most of anything in your score.
Agents: AI agents working for you, and how capable the best one is, 0 to 10. On the Agents screen you can put a free agent to work at a lab where you hold a foothold, one agent per lab. Each night it keeps that foothold from fading and sometimes brings back a little data to sell, for a little heat. Now and then a lab's blue team catches one: the agent is lost, the foothold weakens, and the lab tightens up. If the foothold is lost, the agent comes home. Coercing an insider needs an agent at 7 or better.
Loot: Stolen weights, data and exploits. Sell them on the black market, at the cost of heat.
Intel: What recon has taught you about a lab's network. Recon costs a turn and no heat. Intel raises your odds on your next breach at that lab and is used up by that breach, win or lose. It fades every night, and recon can only build it so high.
Duels: Your best agent against a lab's model or another hacker's agents. Each NPC lab's model takes one duel from you a day, and costs no fight; dueling a player costs a fight. A win earns cred and sometimes skill; a loss costs the agent.
Safehouse: Somewhere to disappear to, tier 1 to 3, bought on the black market. Each tier makes your heat cool faster every night. Going dark keeps it.
Crew: Hired hands, up to 4. Each one adds to your attack power, and each wants wages every night. Miss payroll and one of them walks.
Recordings: Enough of an executive's voice and face to fake them on a call. Exfiltrating from a lab sometimes turns some up, and breaching a player's lab sometimes turns up theirs. They go stale after about ten days, and a call uses them up whether it works or not.
Impersonate: Under Attack (A), once you pick a lab: use recordings to fake an executive on a call to their own staff, for 2 turns. Ask for a wire (cash now, but the bank traces it and your heat rises a week later), a way in (a foothold without a breach), or to prepare the ground for an insider (your next pitch there is likelier to land). Your attack power and agents help; a lab whose staff drill for this is hard to fool, and low public trust makes everyone easier to fool. Against a player it costs a fight. Every fake that works costs the public a little trust.
Fake hire: An insider approach for either class: instead of turning someone, you get a made-up applicant of your own hired, always as an engineer or in ops. A hacker needs an agent of capability 5 to sit the interview, or recordings of someone at the lab; a researcher pays an agency that asks no questions. Staff morale does not matter, but a lab with good security and verification rarely hires a stranger. Every night the lab pays the fake employee's salary to you. A person who does not exist is easier to catch than a real insider, and when one is caught investigators usually find who was paid. Now and then the authorities raid the whole scheme, and every fake hire on the board is at risk.
Zero-days: Unpatched exploits from the black market, up to 3 at a time. Your next breach uses one on its own and is far likelier to work and less likely to be seen. Each costs more than the last.

Job boards

Freelance board: Above-board work under Job boards (J): red-team contracts from the labs, bug bounties, and a few small gigs each day (pen-tests, CTF prizes, talks, audits). It pays modestly and keeps your heat flat or down. Fact-checking a viral clip or taking down scam ads and fake accounts also helps public trust. Each gig is there once, and a bounty pays less each time you file one the same day; the board refreshes at the reset.
Dark web: Illicit jobs under Job boards (J) against outfits that will not call the police loudly — data brokers, exchanges, contractors — never a lab or another player. Each row shows its pay, turns, the skill it needs and your odds. Succeed and you are paid in cash or loot, with cred and some heat; fail and you get only heat, and the loud jobs make the news. Some jobs are about fakes: celebrity investment ads, voices for phone-scam rings, influence ops for either side of the fight over AI rules. They pay well, lower public trust when they work, and the row says what else they do; scam ads made with a lab's open video model get that lab in the news. Taken once, gone until the reset.

The world

Capability: The most capable model anywhere, NPC or player, 0 to 10. Where it goes, and how aligned it is when it gets there, shapes how the season turns out.
Alignment: The world's models' alignment, weighted toward the most capable ones.
Public fear: 0 is enthusiasm, 100 is panic. Leaks, incidents and escaped models raise it.
Regulation: 0 to 100. It follows public fear slowly, a little each night, and lobbying pushes it either way.
Public trust: How well people can tell real from fake, 0 to 100. Fakes that work lower it; it drifts back toward the middle every night as people adapt. When it is low, a leaked document hurts less, because the lab can call it a fake and be half believed; fakes fool people more easily; and public fear swings harder on each day's news.
The election: Three fifths of the way through the season, voters decide which way AI rules go. Public fear and lobbying lean on the result in the weeks before; influence operations and astroturf lean harder. The news counts down the last few days, and the world screen says when it is. On the day, regulation jumps toward the winning side and holds there for a while. Watch for fakes in the run-up.
Compute price: What one GPU costs today. It moves with demand.

Words in lists and scans

Capability 3.0: A lab's model, 0 to 10. A player lab's line shows the same number.
Reputation 55: A lab's public trust, 0 to 100.
Unaligned, mixed, careful: A lab's alignment: under 30, 30 to 69, and 70 or more.
Security ~52: What a scan thinks a lab's security is. The ~ means it is a guess: low skill guesses wide, and at skill 30 the guess is exact.
Breach odds ~35%: Your chance to breach that lab, worked out from the security guess. More skill, a better rig, crew and intel raise it; higher security lowers it. It is never a sure thing either way. Exfiltrating leans on your foothold's strength instead, and jailbreaking on your skill.
Foothold 40: You have a way into that lab, with that strength.
Intel 15: What your recon has found at that lab. The breach odds beside it already count it.
Heat 12: On a hacker's line: their heat.

How a season ends

The season ends when the world reaches an ending, or after 75 days. There is more than one way for it to go. Which ones there are, and what brings them about, is for you to find out over the seasons. The hall of fame remembers how each one went.

A character can also come to an end of their own before the season does. If yours does, you can roll a new one for the rest of the season.

Extra doors

Side games under X on the main menu. Each has its own plays a day. Across all doors, each day they can take at most $250,000 from a researcher or $2,500 from a hacker, and move reputation or cred by 3 and heat by 5.