These words come from the product, so this page is the place they are defined.
01
The Atlas
The Atlas is ARK's occupation record: every job scored on the same skill axes so AI pressure can be read job by job.
The Atlas holds three levels: 1,016 occupations, 18,797 task statements covering 923 of them, and 35 transferable skill axes that all 1,016 occupations are scored against. A profile shows which of a person's skills sit under pressure, which hold, and which tasks move first. The Org Simulator reads the same record to show how a team's professions change as models advance.
- HOW IT WORKS
- Occupation and task records come from the O*NET database, exposure from the published Felten AI exposure index. Importance weighted ratings across the 35 skill axes give every occupation a comparable score, 35,560 ratings in total.
- NOT THIS
- The Atlas is not a prediction of who loses a job. It reads where pressure sits today, at the task level.
02
Rep
A rep is one short piece of real work a person does with AI on purpose, logged so the result can be compared later.
A rep runs on the person's own task, not a sample exercise. It takes a few minutes and ends with something they can keep: a draft, a rewritten process, a decision they can defend. The log is what makes it useful a month later, because two runs of the same task can be put side by side.
- HOW IT WORKS
- One rep a day is the unit of practice. Reps carry points toward a rung, and revisits return on a fixed schedule of 1, 2, 4, 8, 16 and 30 days, reset on a miss.
- NOT THIS
- A rep is not a lesson or a quiz. Nobody scores it out of ten.
03
Dread task
A dread task is the specific piece of work a person puts off, named in their own words, and the place ARK starts.
Capability work fails when it begins with a syllabus. It holds when it begins with the thing someone is avoiding this week: the board update, the tender response, the fourteenth candidate rejection. ARK asks for that task first and hands back a first draft of it inside about a minute.
- HOW IT WORKS
- The dread task is captured in the first run and stored on the profile. Later reps, EDGE moves and daily cards are grounded in it.
- NOT THIS
- A dread task is not a learning objective. It is a real item on a real week.
04
The arc
The arc is the twelve week shape of a person's capability climb, positioned by what they have actually done rather than by the calendar.
An arc holds six rungs, L1 to L6. Position comes from work on record, so a quiet fortnight moves nobody backwards out of guilt and a burst of real rebuilds moves someone up. The arc names what the next rung opens, which is the only honest reason to climb it.
- HOW IT WORKS
- Points accrue from daily cards, logged reps and verified rebuilds. Rung thresholds live in the running configuration and are published on /standards.
- NOT THIS
- The arc is not a course schedule and not a streak. Missing a day costs momentum, never position.
05
EDGE
EDGE is a non-obvious move from another profession, handed to a person one rung above where they stand, with a sixty second way to try it.
The point of EDGE is transfer. A recruiter borrows a move from a trial lawyer, an accountant borrows one from a screenwriter. It arrives after the rep, as the reward, and it never repeats for the same person.
- HOW IT WORKS
- Moves are drawn from a fixed library, filtered by level and diversity. A model may reframe a move in the person's language and never invents one.
- NOT THIS
- EDGE is not a prompt of the day. Every move is attached to a profession that actually uses it.
06
Wild use
A wild use is a time someone reached for AI on their own, with nothing in ARK asking them to.
This is the number that matters. Practice inside a system only proves the system works, while practice outside it proves the person changed. ARK counts wild uses from a capture toggle and one question at the weekly close, and shows the count only once there is a first entry.
- HOW IT WORKS
- Wild uses are self reported and labelled as such. A season certificate carries a line built from wild uses only, and zero is a valid, stated result.
- NOT THIS
- A wild use is not a logged rep, and it is never inferred from activity.
07
Verified proof
A verified proof is a rebuilt workflow with before and after minutes, confirmed by a teammate who did not build it.
Self reported gains are kept and labelled, never totalled. A proof becomes verified when someone else confirms the two timings and how often the task runs. At that point the figures lock, and a correction becomes a new row rather than an edit.
- HOW IT WORKS
- Reclaimed hours shown anywhere in ARK count confirmed proofs only. The wording is measured, not modelled, and it is literal.
- NOT THIS
- A proof is not an estimate, a projection or a vendor calculation.
08
Capability record
A capability record is the durable account of what a person can actually do with AI, carried by the person rather than the employer.
It holds logged reps, concepts held, confirmed proofs and credentials, plus pursuits from outside work if the person chooses to show them. It travels between jobs, because ability does.
- HOW IT WORKS
- Records are owner scoped in the database. A public profile shows only the fields the person switched on, read through security definer functions.
- NOT THIS
- A capability record is not a transcript of courses and not a CV claim.
09
Software tourism
Software tourism is when people visit a tool, look around, and change nothing about how they work on Monday.
It explains most stalled AI programmes better than budget or model choice does. Licences get issued, a pilot runs, everyone agrees it was interesting, and the work stays exactly as it was. The cure is a change to one real task, kept.
- HOW IT WORKS
- ARK treats a first artifact as the entry point, so the first session ends with work the person keeps rather than a tour of features.
- NOT THIS
- Software tourism is not a skills problem. Capable people do it too, when nothing asks them to change the work.
The longer answer →10
System of record for AI capability
A system of record for AI capability is the single trusted place an organisation's real AI ability is measured and kept, the way a CRM holds customers.
Scattered evidence is the normal state: a spreadsheet of course completions, a channel of clever prompts, a manager's impression. A system of record replaces that with one account of demonstrated ability, at person, team and company level, that stays current as the models move.
- HOW IT WORKS
- Capability aggregates upward through owned scope only. A manager sees their own department, a company roll up needs at least three people per group, and cross company patterns need three distinct organisations.
- NOT THIS
- It is not a content library and not a reporting layer over a course catalogue.
The longer answer →11
The daily loop
The daily loop is ARK's one action a day: a card that is ready when the person arrives, one rep on their real work, then the day closes.
The card is generated when the previous day closes, so nobody waits on a model. From Friday the close hands off into the weekly review. Weekends default to a private rep on something outside work, which still counts toward the chain.
- HOW IT WORKS
- Cards are cached ahead with a deterministic fallback, so the loop works even when the day's generator is offline.
- NOT THIS
- The loop is not a feed and not a notification programme. One honest nudge a day at most, and silence is the default.