11.5 · Scoring
How scoring works.
One number a week: 155 points this week. When the bar is full, the week is done.
This is the rulebook behind the bar on the homepage. It is public so that anyone watching can check the arithmetic. The rules live in code, not in anything I edit day to day, and they apply the same way to everything.
The week
- One point is 15 focused minutes.
- The bar is set at the start of each week and holds until Sunday, whatever comes on or off the plate. Weeks run Monday to Sunday, Dubai time. This week it is 155 points, 38.75 hours.
- Full means done. Nothing more is owed that week. Anything past full shows as extra, and extra is a choice.
- Points only come from finishing something. Starting, planning and backlog earn nothing.
- Scoring began on 19 Oct 2026. Earlier work shows as progress but earns no points.
Two lanes
One question sorts everything: is there a commitment outside my head? That means a date someone else holds me to, a person waiting on it, or money already spent or at stake. Good for me someday does not count; nearly everything I would choose to do is.
- Quest. Yes. A booked exam, a deadline, work, anything someone else depends on. Quests are uncapped.
- Field Atlas outings. The one exception: a quest by rule, though nobody holds me to it. A day out counts in full, fills the bar and raises next week’s like any quest, and is never crowded out by study.
- Curiosity. No, and it does not have to. Reading, courses taken for interest, The List, and anything I write about them. Curiosity fills whatever the week has left after dated commitments: the bar less the floor (70 points this week, 17.5 hours). Past that it still counts toward the lifetime total, never the week.
A textbook or reading takes the lane of whatever it belongs to. On its own, it is curiosity.
The lane changes on an event, not a mood. An exam I mean to sit but have not booked is curiosity; the day it is booked, paid for or needed by someone, it becomes a quest. Work logged before a change keeps the lane it was done in, so a change never rewrites a week already under way.
How things score
- Parts. Work with a fixed end: chapters, modules, sections, pages, milestones. Each finished part is one tap. The parts of an item always add up to exactly its total.
- Repeatables. Work that can be done any number of times: past papers, practice exams, revision passes. Every one earns the same points. No limit.
- Sitting an exam. The exam itself, sat, pass or fail: its official length at the retrieval rate (1.5), with no marking, since someone else marks it. A 3-hour paper is 18 points, paid to the week it was sat, as a quest. Never corrected: the length is the exam board’s.
- Outcomes. The result itself: an exam passed, an offer, a paid project. Worth the item's full estimate, set in advance and paid once. Outcomes go to the lifetime total and the level, never the week. Taking longer or retaking earns points for the extra work, but the outcome never grows.
Estimates
Every price is set before the work starts, from the first of these that exists:
- An official timing. A past paper's set length. Used as it is, and never corrected.
- The content, counted. Pages or words at a reading speed (below), video at its running time, a walk from its distance and climb. Times 1.25 to start with, for breaks and setup.
- Published hours. A course's own figure. Times 1 to start with: published hours tend to run long, not short.
- My guess, for the whole thing, written down raw. Times 2 to start with. The factor does the correcting, so a guess is never padded by hand.
- Prices are set once and never changed. A part that took twice as long as expected earns its price, and so does one that took half. Over a month the swings cancel; judging one part at a time would only measure the noise.
- No part is worth more than 16 points, or 8 for a guess. Bigger work is split into parts, so one bad estimate can only be so wrong. An official timing is one unit, and exempt.
- Every part pays. A part is worth at least one point, so a button that finishes a piece of work never pays nothing; a group priced below that is raised to it. Pages are the one exception, since a book is logged by the page reached rather than a tap a page. Counted activities (cards, questions) pay by count, so a small batch can round to nothing until the next one carries it over; otherwise logging one card at a time would pay forty times the rate.
- One re-measure, never a re-price. Once the first part of something priced from published hours or a guess is done, what is left is priced again from what it holds: pages, words and video, counted. Up or down, once only, and whatever that first part took, so going slowly buys nothing. Otherwise a price reopens only if the work itself changes or a clerical error is found, never for speed or skill.
- Anything worked on before it is priced is set up at half rate, because a price set after a look at the work leans on it.
- Points are frozen when something is set up. Points already earned are never rewritten.
Calibration
My timing never changes a price. It corrects the factors, slowly, so the next estimate of the same kind is better. A rule that let my own speed set my future prices would pay slow work, so none is allowed.
- One factor per kind of estimate. Guesses, published hours and counted content each learn from their own record, never from each other. Official timings and the activity rates never learn: getting faster at them is the point.
- Once a month. A review on the first Monday of each month reads the last 26 weeks. A new factor applies only to things priced after it.
- Items, not minutes. Each item counts once, however many parts it has; none carries more than a quarter of the weight, and each is held between a third and three times its price. Trust in the record grows with the number of items: halfway at five, two thirds at ten, four fifths at twenty.
- Small steps. A review moves a factor at most a tenth, to the nearest 0.05. One that would end up under half or over double where it started is held, and the whole kind is checked by hand.
- Clean time only. Only time run on the timer counts, never time typed in afterwards. Every session counts its own minutes; overlaps between sessions are never trimmed, since pausing, deleting and typing time in leave overlaps that mean nothing. Only time up to an item's latest finished part counts, and an item with any finished part untimed, or a tenth or more of it typed in, is left out, so timing only the slow parts cannot steer a factor. An item needs at least 20 timed minutes.
| Estimate | Starts at | In force | On 7 Dec, so far |
|---|---|---|---|
| My guess | ×2 | ×1.8 | ×1.65, from 1 item |
| Published hours | ×1 | ×1 | ×1 |
| Counted content | ×1.25 | ×1.25 | ×1.25 |
The factors in force were set by the review of 2 November 2026. The next is on 7 December 2026.
Notes
Any finished part that took under half or over double its price on the timer gets a line here, with how it was priced. A part with a tenth or more of its time typed in afterwards gets none, since the timer saw only some of it. Points never change.
- 6 Nov · ICS Shipping Business, module: 34 min against 15 min priced (2.27×). Priced from a content measure: 6,000 words at normal reading, x 1.25; at least a point a module.
- 2 Nov · ICS Shipping Business, module: 45 min against 15 min priced (3×). Priced from a content measure: 6,000 words at normal reading, x 1.25; at least a point a module.
- 27 Oct · Frontier AI Governance, Section 2: 2 h 56 min against 10 h priced (0.29×). Priced from a guess: about 5 hours a section from BlueDot's guide, doubled to 10 after section one took about 10 hours; paid 11 points, not 40, what its content measures (8 for 9,565 technical and 9,910 normal words x 1.25, 3 for BlueDot's stated 45-minute exercise), a one-time exception.
Rates
| Reading tier | Speed | Pages per point | For |
|---|---|---|---|
| Light | 2 pages a min | 24 | self-help, popular non-fiction |
| Normal | 1 min a page | 12 | most books, essays and articles |
| Technical | 4 min a page | 3 | textbooks with mathematics, policy and legal reports |
| Dense | 10 min a page | 1.2 | research papers |
| Thing | How it scores |
|---|---|
| Essay or writeup | Once, on publish: 75 finished words a point (300 an hour). Edits never score. |
| Field Atlas outing | Gate to gate: 3 hours of drive and setup, plus 1 hour per 5 km and per 600 m of climb, times 1.25. Without route data, a typical day: hiking 6 h, diving 8 h, fishing 6 h. |
| Task with no estimate | A gut figure by scope, times 2: today 1 h, this week 2 h, this month 4 h, this quarter 8 h, this year 12 h, lifetime 8 h. |
| Recurring routine | Per occurrence, times 2: daily 0.25 h, weekly 1 h, monthly 2 h, quarterly 3 h, yearly 6 h. |
| A normal 300-page book | 25 points. |
Text counted in words, like an article, becomes pages at 300 words a page for light and normal reading and 600 for technical and dense.
Levels
One level per 1,000 lifetime points, about five full weeks. Levels never go down.
The bar
A bar I miss week after week only teaches me to miss it. So it moves with what I actually do, and a few rules keep it from drifting where it should not.
- Halfway. Each Monday the bar moves halfway from last week's bar to what last week counted, to the nearest 5. Score 100 against 200 and next week's bar is 150; beat it and it rises the same way.
- A floor. It never drops below what work with a date someone else fixed needs that week: every booked exam, and any task whose date was set for me, on the plate or not. That need is what each one had left at the start of the week, over the weeks to its date: unfinished parts, plus practice still to do, priced as they stood that Monday. A plan or re-measure made during a week counts from the next one.
- A ceiling. It never goes above 200 points, 50 hours.
- Curiosity gets the rest. What dated work needs (the floor) is the quests' share of the week; curiosity can fill everything else, and nothing more. There is no fixed split: an exam week can leave curiosity an hour or none, and a week with nothing dated leaves it the whole bar. Before 2 November 2026 curiosity had a flat 40%. The plate does not follow the bar: it holds 25 hours whatever the bar is, because the bar says when a week is full, not how much there is to do.
- Nothing by hand. It is worked out from the record every time the score is, and changing what is on the plate midweek changes none of it.
This week: a bar of 155, with room for 70 of curiosity, with a floor of 85. Next week, as it stands today: 90, with a floor of 90, and room for 0 of curiosity. Until 2 November 2026 the bar was 200 every week.
Load
Only what is active counts. Every active thing says how many hours a week it takes, and together they cannot pass 25: how many hours a week I can realistically take on. That figure is mine, set and reviewed by hand whenever I see fit, and in force as soon as it changes; it limits what goes on the plate and changes no points. It was 50 until 10 November, when the timer showed about 22 to 26 hours in a real week. Anything with a deadline works its figure out on its own: the hours its unfinished parts and what is left of its activity plans are worth, over the weeks left. Fall behind and the pace rises; get ahead and it drops. Anything without one carries the hours a week I choose to give it. Curiosity has a second limit: this week’s curiosity, the hours already logged plus what the plate still asks for, cannot pass its room in hours (17.5 hours this week), since hours past it could not fill the bar. Something refused for that can go on at the hours that still fit, or wait for Monday, when the logged hours start again. What cannot be given hours yet goes in the backlog, which counts for nothing until it is brought onto the plate with a date or hours a week. Anything that would push the plate over is refused until something comes off: a swap, so the load stays level. A task with no figure of its own takes its estimate spread over its scope (all of it in a this-week task, a thirteenth in a this-quarter one). Backlog counts for nothing until it is swapped in.
Activities
Understanding the material is half of passing. The other half is the work around it, and it scores by count against a fixed rate: realistic minutes per unit, turned into points. Retrieval practice earns 1.5×: recalling from memory is what turns understanding into marks, and it is the work least likely to happen on its own. Making flashcards never scores; they are made by AI. Something with a date can carry practice hours: how long all its activities together should take. From 16 Nov, an exam’s practice hours are worked out, not guessed: the hours its parts are priced at, one to one. Part prices learn from timed sessions every month, so practice follows them; how much practice I actually do never feeds back, or skipping it would pay. A figure of my own still overrides the rule, from the Monday after. What the logged activities have not covered, at their own rates, counts toward its hours a week and, for a fixed date, the floor (at four points an hour, so the floor errs low). Without practice hours, each activity can carry a plan instead: how many I mean to do. Neither ever pays; only logged units do.
| Activity | Points | Basis |
|---|---|---|
| Flashcard reviews | 1 per 40 cards | 15 seconds a card, including failed cards relearned, 1.5× as retrieval. |
| Practice questions, multiple choice | 1 per 3.7 questions | 1.7 minutes to answer (the SQE pace) and 1 to review, 1.5× as retrieval. |
| Practice questions, written | 4.5 per question | 36 minutes to answer (five to a 3-hour paper) and 9 to review, 1.5× as retrieval. |
| Past paper | 27 for a 3-hour paper | its official length, and half again to mark it, 1.5× as retrieval. |
| Learning statutes | 1 per 1.5 statutes | 10 minutes to read, understand and note one. |
| Learning cases | 1 per 1.88 cases | 8 minutes for the facts, the ratio and the principle. |
| Practice box, easy | 12 per easy box | 2 hours to own an easy box, user and root, 1.5× as retrieval. |
| Practice box, medium | 24 per medium box | 4 hours to own a medium box, user and root, 1.5× as retrieval. |
| Practice box, hard | 48 per hard box | 8 hours to own a hard box, user and root, 1.5× as retrieval. |
| Something else (your guess per unit) | worked out | your gut minutes per unit, times the guess factor in force, at most 8 points a unit. |
Time
Time spent on something that already has its own button (a chapter, flashcard reviews) is a record only; the button scores the work. Time on work with no button of its own, like supplementary reading or a one-off, scores at half rate: 2 points an hour. Hours are input, not output, and a timer cannot tell focus from drift. At the end of a session the focused minutes are confirmed, never more than the timer ran, with the breaks taken out by me. Its points are frozen into the log. Before 2 November 2026, one session counted for at most 3 hours. Time typed in afterwards is marked as such: it stays in the record, but never calibrates an estimate.
Finishing
An item reaches 100% when its last part is done. A task then closes as done and a course as completed, and both leave the plate. A book closes on its last page. An exam stays until it is sat; the result pays its outcome. Anything can also be marked done at any point, with one tap: it leaves the plate for good, earns nothing, and its unfinished parts stay unfinished. If it is done, it is done. A new week starts at midnight on Monday, Dubai time, on its own; past weeks are never stored separately, they are read back from the dated log in each item, so they cannot drift.
Changing the rules
A change to these rules goes in the changed-my-mind log before it applies, takes effect the Monday after at the earliest, and never touches points already earned. Every proposal has to pass one test: an extra minute on the clock must never buy a minute or more of future price.