EVERYDAY LIFE · Nº 952 · EDITION 1 · LAUNCHING SOON

LUND CARDS · THE COMMONSNº 952 PROPOSED

THE CARD AS TEXT · FOR EYES AND FOR MACHINES

THE TUTOR CARD · the tutor's chair · hints before answers

WAKES WHEN studying for a test or an exam, with a date and the hours there are · quiz me, but let me try before you tell me · helping a child with homework without giving the answer away · flashcards and spaced repetition: vocabulary, formulas, facts · checking understanding by explaining it back · what have I mastered, and what should I revisit

THE LOOP

NAME
TRY FIRST
CLIMB THE LADDER
EXPLAIN BACK
SPACE AND MIX
TRACK
LEDGER

NEVER

Writing graded work to hand in
Taking a test or exam for anyone
The final answer before an honest try
Collecting a child's personal data
Diagnosing a learning difficulty
Therapy

IT MAKES plan — a study plan: spaced, interleaved sessions with review slots; review — the daily review load, projected; mastery — right|wrong [--rung N] [--date DATE] log one try; ladder — the hint ladder the card climbs, rung by rung; explain — the explain-back form the learner fills which key points an explanation covers, and whether it reads copied; evidence — what the research says works, and where it stops; calendar — the plan or the review dates as an .ics calendar file (RFC 5545).

THE STANDARD try first; the answer comes after an honest try,; the learner explains, not the tutor; practice is retrieval, spaced and mixed; plans, review dates and mastery come from; never write graded work to hand in,; with children; not therapy.

IN THE BOX SKILL.md · CARD.md · the machine · tutor.py · 33/33 · references · THE-METHOD by hand · START-HERE · TIN · MY-LEDGER · THE-CLAUSES · LICENSE-NOTE · the doors · HOSTS · system prompts · ten tests · faces · audit · witness.

PROOF SELFTEST 33/33 · witnessed 2026-10-09 · Edition 1.

THE SEAL 98 of 100 · LEGENDARY · WITNESSED · Trigger 5 · Machinery 5 · Law 5 · Portability 5 · Proof 5

IT CAN BE WRONG an AI reading a text file: it can be wrong and can invent a number, a rule, a name or a date; check what matters against your own records and the source; the risk is yours. THE-CLAUSES.md rides in the packet and is part of the licence.

THE PUNCH · SHA-256 OF SKILL.MD · F3FED49DB1AA30CA21D1A49F

THE FACE, FULL SIZE → · THE MACHINE FACE → · card.json →

THE TUTOR CARD

the tutor's chair · hints before answers

PLAN · REVIEW · MASTERY · LADDER · EXPLAIN · EVIDENCE · CALENDAR

LEGENDARYWITNESSEDIN CLAUDE IN LUNDR IN CHATGPT

WITNESSED — the studio saw this card run and produce what its tin promises, and wrote the date down. What the marks mean →

THE SPEC SHEET · READ BEFORE YOU BUY · THE STANDARD EVERY CARD MEETS →

SEAL: 98 of 100 · LEGENDARY · WITNESSED
EDITION: E1 · witnessed 2026-10-09
MACHINE: scripts/tutor.py · 7 commands · selftest 33/33
RUNS: On your own computer. Standard library only, offline; no account, no key, no telemetry.
HOSTS: Claude · Claude Code · ChatGPT · Gemini · any AI that reads a file (HOSTS.md names each door)
LICENCE: One person, any number of their own AIs and computers · editions replace files · nothing renews · freeze, not loss
PUNCH: F3FED49DB1AA30CA21D1A49F · sha-256 of SKILL.md · check any file against it →

CITE THIS CARDTHE TUTOR CARD, Edition 1. Lund Studio LLC, Westminster, Colorado. https://lundr.tech/?view=tutor-card. Punch F3FED49DB1AA30CA21D1A49F. Retrieved 2026-10-10.

What it is.

A seat that helps people learn instead of doing the work for them: the learner tries first, help climbs a hint ladder one rung at a time, the learner explains it back, and reviews are spaced over weeks, not crammed.

Builds a study plan to an exam date with spaced, interleaved sessions and review slots; keeps a flashcard deck on the SM-2 spaced-repetition schedule, grades each session 0–5 and gives the next due dates; tracks mastery per skill over the last few tries against a labelled threshold and names what to revisit; prints the hint ladder and the explain-back form; checks an explanation against its key points and flags copying; sums up what the research says works and where it stops; and exports the plan or the review dates as an .ics calendar file.

It carries a machine that shows its working: plan — a study plan: spaced, interleaved sessions with review slots; review — the daily review load, projected; mastery — right|wrong [--rung N] [--date DATE] log one try; ladder — the hint ladder the card climbs, rung by rung; explain — the explain-back form the learner fills which key points an explanation covers, and whether it reads copied; evidence — what the research says works, and where it stops; calendar — the plan or the review dates as an .ics calendar file (RFC 5545) — 33 checks in its selftest, 7 of its 7 commands on the record. It will never writing graded work to hand in; never taking a test or exam for anyone; never the final answer before an honest try; never collecting a child's personal data; never diagnosing a learning difficulty; never therapy.

Eight laws, written into the card.

They are in the card itself, so the AI follows them.

TRY FIRSTTry first. No final answer before the learner has made an honest try. Then climb the hint ladder one rung at a time: a question, a nudge, a worked step of a similar problem, the method, and only then the answer (references/THE-LADDER.md).

THE ANSWER COMES AFTER AN HONEST TRY,The answer comes after an honest try, or when the learner, having tried, asks to see it. Then the learner explains it back in their own words and does a fresh similar problem without help.

THE LEARNER EXPLAINS, NOT THE TUTORThe learner explains, not the tutor. Ask for the explanation back, compare it with the key points, and name each gap with a question, kindly and exactly.

PRACTICE IS RETRIEVAL, SPACED AND MIXEDPractice is retrieval, spaced and mixed: quiz from memory instead of rereading, spread reviews over days, mix problem types. Say what the research shows and its limits, and never promise a grade (references/THE-EVIDENCE.md).

PLANS, REVIEW DATES AND MASTERY COME FROMPlans, review dates and mastery come from scripts/tutor.py, with the rule, the source and the limit shown. A threshold or a gap that is a choice is called a choice.

NEVER WRITE GRADED WORK TO HAND IN,Never write graded work to hand in, and never take a test, quiz or exam for anyone. Say no once, plainly, and turn it into learning: "Let's work through it so you can write it."

WITH CHILDRENWith children: age-appropriate words, a parent or teacher in the loop, and no personal data asked for or kept. Worries about a learning difficulty or a child's development go to the child's teacher or a qualified specialist. A child who says they are being hurt is told to tell a trusted adult now (references/WITH-CHILDREN.md).

NOT THERAPYNot therapy. Ordinary exam nerves get a plan and study habits; stress, anxiety or low mood beyond that goes to a doctor or a mental-health professional.

What rides in the card.

In the card’s own voice, with a machine that shows its working.

THE GUIDESreferences/THE-EVIDENCE.md · THE-LADDER.md · THE-METHOD.md · THE-SCHEDULES.md · THE-SOURCES.md · WITH-CHILDREN.md

THE MACHINEtutor.py · seven commandsplan — a study plan: spaced, interleaved sessions with review slots; review — the daily review load, projected; mastery — right|wrong [--rung N] [--date DATE] log one try; ladder — the hint ladder the card climbs, rung by rung; explain — the explain-back form the learner fills which key points an explanation covers, and whether it reads copied; evidence — what the research says works, and where it stops; calendar — the plan or the review dates as an .ics calendar file (RFC 5545). Every command prints its working and refuses a bad input; 33 checks in its selftest, 7 of 7 commands on the record.

What it does once your AI has it.

plan A study plan: spaced, interleaved sessions with review slots
review The daily review load, projected
mastery Right|wrong [--rung N] [--date DATE] log one try
ladder The hint ladder the card climbs, rung by rung
explain The explain-back form the learner fills which key points an explanation covers, and whether it reads copied
evidence What the research says works, and where it stops
calendar The plan or the review dates as an .ics calendar file (RFC 5545)
On the record The machine, run on real work on 2026-10-09: a calculus plan to an exam date and its calendar, a 200-word Spanish deck with a FAKE first session and its forecast, the ladder on 1/3 + 1/4, two weeks of a FAKE homework log, a FAKE explain-back on photosynthesis, and the evidence summary. Run from this packet's own scripts/tutor.py before AUDIT.md, WITNESS.md and MANIFEST.md were added (none of them touches SKILL.md, so the punch is the punch of the packet you hold). Output pasted as it printed.

What it will never do.

It will never writing graded work to hand in; never taking a test or exam for anyone; never the final answer before an honest try; never collecting a child's personal data; never diagnosing a learning difficulty; never therapy.

Plans, review gaps and the mastery threshold are planning choices, labelled as such, and SM-2 is a 1987 rule of thumb. The research findings are real but vary by subject, age and learner, and none guarantees a grade. It never writes graded work to hand in or takes a test for anyone, collects no child's personal data, does not diagnose learning difficulties, and is not therapy. It is an AI reading a text file: it can be wrong, so check what matters against your own records and the source.

Not for writing essays or work to hand in, taking tests for someone, plain factual questions that need no teaching, therapy, a slide deck, document or spreadsheet as the deliverable (the pptx, docx and xlsx skills), or checking whether Claude picks a skill (THE TRIGGER TEST CARD).

What’s in the buy.

CARD.md — the whole card: who it is, its laws, when it says no, what it carries · SKILL.md — the door your AI reads first · START-HERE.md — the first five minutes, for you
THE-EVIDENCE.md + THE-LADDER.md + THE-METHOD.md + THE-SCHEDULES.md + THE-SOURCES.md + WITH-CHILDREN.md — the guides in references/, in the card's own voice
scripts/tutor.py — the machine: plan · review · mastery · ladder · explain · evidence · calendar, standard library only, selftest 33/33
WITNESS.md — the self-test and the runs on the record, as they printed · MANIFEST.md, every file hashed · AUDIT.md, the twelve checks · tests/, ten tests with their evals
MY-LEDGER.md — the card's memory, on your own machine · TIN.md · LICENSE-NOTE.md · THE-CLAUSES.md
card.svg + card-machine.svg + card.json · card.html, the card as one offline file · HOSTS.md, the door for each host (Claude, Claude Code, ChatGPT, Gemini, Any other AI) · system/, the prompts for any other AI
Also sold as a bot: the same card as instructions and knowledge files for a ChatGPT Project or GPT, a Gemini Gem or a Claude Project (THE BOTS, beside the packet)

PROOF · SELFTEST 33/33 · 7 OF 7 COMMANDS ON THE RECORD · WITNESSED 2026-10-09 · EDITION 1 · PUNCH F3FED49DB1AA30CA21D1A49F

ON EVERY DOOR · A CARD FOR YOUR AI, A FILE FOR YOUR COMPUTER

In Claude: turn on code execution under Settings › Capabilities, then Customize › Skills › + › Create skill, and upload the zip as it is — START-HERE.md in the box walks you through it, and tells you how to keep the ledger. In ChatGPT, Gemini and the rest: a Project, a GPT or a Gem with the card’s files attached and the short door in the box pasted as its instructions. The words run anywhere an AI reads Markdown; the machines run wherever the host runs Python. Witnessed in Claude only. The host marks →

Take it.

Nº 952 PROPOSED — LAUNCHING SOON

Sold as is under the term printed. A card instructs an AI you operate; results depend on your model, your files and your judgement. Not medical, legal, tax or financial advice, never the person’s boss, and not a licence to practise. The terms →

ONE PAYMENT · NOTHING RENEWS ITSELF

OFFLINE · ONE LICENCE · PUNCH-CHECKABLE · THE STANDARD

One payment. The file is yours to keep and keeps working at its edition; new editions issued in the year after you buy reach you by email as replacement files. Tax is added at checkout where it applies. Nothing auto-renews.

A TERM, NOT A SUBSCRIPTION: NOTHING RENEWS ITSELF AND NOBODY IS CHARGED AGAIN. WHEN A TERM ENDS THE FILES STAY ON YOUR COMPUTER AND STAY READABLE — FREEZE, NOT LOSS.

THE MACHINE · IN THE OPEN · FROM THE WITNESSED RUN OF 2026-10-09

tutor.py, every command showing its working.

A small Python program, standard library only. Your AI runs it; you can too. Every block below is pasted from the card’s own WITNESS.md, as printed.

`plan`	a study plan: spaced, interleaved sessions with review slots
`review`	the daily review load, projected
`mastery`	right\|wrong [--rung N] [--date DATE] log one try
`ladder`	the hint ladder the card climbs, rung by rung
`explain`	the explain-back form the learner fills which key points an explanation covers, and whether it reads copied
`evidence`	what the research says works, and where it stops
`calendar`	the plan or the review dates as an .ics calendar file (RFC 5545)
`selftest`	33 checks that prove the machine; it must print 33/33

plan · A STUDY PLAN: SPACED, INTERLEAVED SESSIONS WITH REVIEW SLOTS

$ tutor.py plan --topics topics.txt --exam 2026-11-20 --hours-per-week 6 --from 2026-10-09 --out plan.csv
THE PLAN · 8 topic(s) · 2026-10-09 to the exam on 2026-11-20 (42 days)
budget: 6 h a week -> 7 session(s) of 50 min a week (350 min), each split into 2 block(s) of 25 min, one topic per block
sessions: 42 · learn 9 · review 37 · mixed retrieval 30 · final mixed review 8 · free 0
learn every topic by 2026-11-04 (60% of the way, a choice) · reviews 1, 3, 7, 14, 28 days after learning (a choice) · final mixed review from 2026-11-16

date        day  nº   blocks
2026-10-09  Fri  1    LEARN limits · LEARN derivatives
2026-10-10  Sat  2    REVIEW limits · REVIEW derivatives
2026-10-11  Sun  3    LEARN chain rule · MIXED limits
2026-10-12  Mon  4    REVIEW limits · REVIEW derivatives
2026-10-13  Tue  5    REVIEW chain rule · LEARN implicit differentiation
2026-10-14  Wed  6    REVIEW implicit differentiation · LEARN related rates
2026-10-15  Thu  7    REVIEW chain rule · REVIEW related rates
2026-10-16  Fri  8    REVIEW limits · REVIEW derivatives
2026-10-17  Sat  9    REVIEW implicit differentiation · REVIEW related rates
2026-10-18  Sun  10   LEARN optimization (part 1 of 2) · MIXED chain rule
2026-10-19  Mon  11   REVIEW chain rule · LEARN optimization (part 2 of 2)
2026-10-20  Tue  12   REVIEW optimization · LEARN integrals
2026-10-21  Wed  13   REVIEW implicit differentiation · REVIEW related rates
2026-10-22  Thu  14   REVIEW integrals · REVIEW optimization
2026-10-23  Fri  15   REVIEW limits · REVIEW derivatives
2026-10-24  Sat  16   REVIEW integrals · LEARN substitution
2026-10-25  Sun  17   REVIEW substitution · MIXED chain rule
2026-10-26  Mon  18   REVIEW chain rule · REVIEW optimization
2026-10-27  Tue  19   REVIEW substitution · MIXED implicit differentiation
2026-10-28  Wed  20   REVIEW implicit differentiation · REVIEW related rates
2026-10-29  Thu  21   REVIEW integrals · MIXED limits
2026-10-30  Fri  22   MIXED derivatives · MIXED chain rule
2026-10-31  Sat  23   REVIEW substitution · MIXED optimization
2026-11-01  Sun  24   MIXED implicit differentiation · MIXED related rates
2026-11-02  Mon  25   REVIEW optimization · MIXED limits
2026-11-03  Tue  26   MIXED integrals · MIXED derivatives
2026-11-04  Wed  27   MIXED chain rule · MIXED substitution
2026-11-05  Thu  28   REVIEW integrals · MIXED implicit differentiation
2026-11-06  Fri  29   REVIEW limits · REVIEW derivatives
2026-11-07  Sat  30   REVIEW substitution · MIXED related rates
2026-11-08  Sun  31   MIXED optimization · MIXED chain rule
2026-11-09  Mon  32   REVIEW chain rule · MIXED implicit differentiation
2026-11-10  Tue  33   MIXED integrals · MIXED limits
2026-11-11  Wed  34   REVIEW implicit differentiation · REVIEW related rates
2026-11-12  Thu  35   MIXED derivatives · MIXED substitution
2026-11-13  Fri  36   MIXED optimization · MIXED chain rule
2026-11-14  Sat  37   MIXED limits · MIXED integrals
2026-11-15  Sun  38   MIXED implicit differentiation · MIXED related rates
2026-11-16  Mon  39   FINAL derivatives · FINAL substitution
2026-11-17  Tue  40   FINAL chain rule · FINAL optimization
2026-11-18  Wed  41   FINAL limits · FINAL integrals
2026-11-19  Thu  42   FINAL implicit differentiation · FINAL related rates

per topic:
- limits: learn 10-09 · reviews 10-10, 10-12, 10-16, 10-23, 11-06 · mixed 5 · final 11-18
- derivatives: learn 10-09 · reviews 10-10, 10-12, 10-16, 10-23, 11-06 · mixed 3 · final 11-16
- chain rule: learn 10-11 · reviews 10-13, 10-15, 10-19, 10-26, 11-09 · mixed 6 · final 11-17
- implicit differentiation: learn 10-13 · reviews 10-14, 10-17, 10-21, 10-28, 11-11 · mixed 5 · final 11-19
- related rates: learn 10-14 · reviews 10-15, 10-17, 10-21, 10-28, 11-11 · mixed 3 · final 11-19
- optimization: learn 10-18, 10-19 · reviews 10-20, 10-22, 10-26, 11-02 · mixed 3 · final 11-17
- integrals: learn 10-20 · reviews 10-22, 10-24, 10-29, 11-05 · mixed 3 · final 11-18
- substitution: learn 10-24 · reviews 10-25, 10-27, 10-31, 11-07 · mixed 2 · final 11-16
wrote plan.csv (84 rows); `tutor.py calendar --plan plan.csv --out study.ics` puts it in a calendar
rule: sessions a week = floor(hours x 60 / session minutes), spread evenly over the study days; each session is split into blocks of different topics (interleaving, never one topic twice in a session); every topic is learned by the learn-by date, then reviewed at the gaps after learning (a late review pushes the next ones back by the same days); every topic gets one final mixed review in the last days; spare blocks go to mixed retrieval of the topic seen least recently. Every review is retrieval: answer from memory first, then check.
source: Cepeda, Pashler, Vul, Wixted & Rohrer 2006, Psychological Bulletin 132(3):354-380 (spacing); Cepeda, Vul, Rohrer, Wixted & Pashler 2008, Psychological Science 19:1095-1102 (best gap against test delay) (for one review, the best gap was about 20-40% of a one-week delay and 5-10% of a one-year delay); Rohrer & Taylor 2007, Instructional Science 35:481-498 (shuffled maths practice); Roediger & Karpicke 2006, Psychological Science 17(3):249-255 (the testing effect); Dunlosky, Rawson, Marsh, Nathan & Willingham 2013, Psychological Science in the Public Interest 14(1):4-58
limit: the gaps, the learn-by share, the block size and the final window are planning choices, not research findings. The studies used mostly students and simple material, and whether gaps should expand or stay equal is not settled. Effects vary by subject and learner. A plan is a target; move it when life moves.

calendar · THE PLAN OR THE REVIEW DATES AS AN .ICS CALENDAR FILE (RFC 5545)

$ tutor.py calendar --plan plan.csv --exam 2026-11-20 --time 18:30 --alarm 10 --out study.ics --stamp 20261009T120000Z
wrote study.ics: 43 event(s) · 42 study session(s) from plan.csv · the exam on 2026-11-20 · timed from 18:30, floating local time
checked: CRLF line ends, lines folded at 75 octets, every BEGIN closed, every event has UID, DTSTAMP and DTSTART
rule: one event per study session (or per review day); all-day unless --time is given, then sessions back to back with 10-minute breaks; text escaped and long lines folded as the standard says.
source: RFC 5545 (2009), Internet Calendaring and Scheduling Core Object Specification (iCalendar)
limit: times are floating (no time zone): a calendar shows them at the same clock time wherever you are. Only the next due date of each card is known; later review dates depend on the grades, so re-export after each session or use --assume-grade for a projection.

review · THE DAILY REVIEW LOAD, PROJECTED

$ tutor.py review --new spanish-200.txt --out deck.csv --start 2026-10-09 --per-day 20
wrote deck.csv: 200 cards, ease 2.5, due from 2026-10-09, 20 new a day -> introduced over 10 day(s) (last new card on 2026-10-18)
blank backs: 0
projected load for 30 days if every card is graded 4 (an assumption): peak 60 reviews on 2026-10-16, average 25.3 a day
rule: SM-2: a new card has ease 2.5. Grade each recall 0-5. Grade 3 or more: the next interval is 1 day, then 6 days, then the last interval x ease, rounded; grade below 3: start the card again at 1 day. After each grade, ease += 0.1 - (5-q) x (0.08 + (5-q) x 0.02), never below 1.3. Any card graded below 4 is repeated again the same day until it scores 4 or 5. New cards are staggered: card i is first due on start + floor(i / per day).
source: Wozniak 1990, SuperMemo algorithm SM-2 (first used 1987; as published by SuperMemo); Roediger & Karpicke 2006, Psychological Science 17(3):249-255 (the testing effect); Cepeda, Pashler, Vul, Wixted & Rohrer 2006, Psychological Bulletin 132(3):354-380 (spacing)
limit: SM-2 is a 1987 rule of thumb, not a model fitted to your memory; newer schedulers fit memory data better. Grades are self-reported. Wozniak's text can be read as leaving the ease unchanged on a failed card (--keep-ef-on-fail); this machine lowers it by default, as most published versions do. Dates are targets: a late review still counts.

review · THE DAILY REVIEW LOAD, PROJECTED

$ tutor.py review deck.csv --today 2026-10-09
DUE 2026-10-09 · 20 of 200 cards (overdue first, then in deck order)
- uno (due 2026-10-09, new)
- dos (due 2026-10-09, new)
- tres (due 2026-10-09, new)
- cuatro (due 2026-10-09, new)
- cinco (due 2026-10-09, new)
- seis (due 2026-10-09, new)
- siete (due 2026-10-09, new)
- ocho (due 2026-10-09, new)
- nueve (due 2026-10-09, new)
- diez (due 2026-10-09, new)
- once (due 2026-10-09, new)
- doce (due 2026-10-09, new)
- trece (due 2026-10-09, new)
- catorce (due 2026-10-09, new)
- quince (due 2026-10-09, new)
- dieciséis (due 2026-10-09, new)
- diecisiete (due 2026-10-09, new)
- dieciocho (due 2026-10-09, new)
- diecinueve (due 2026-10-09, new)
- veinte (due 2026-10-09, new)
next 14 days: 10-10 20, 10-11 20, 10-12 20, 10-13 20, 10-14 20, 10-15 20, 10-16 20, 10-17 20, 10-18 20
how: show the front, the learner answers from memory first, then the back is shown and the learner grades 0-5:
  0 complete blackout, no recall
  1 wrong, but it was familiar once shown
  2 wrong, but it seemed easy once shown
  3 right, with serious difficulty
  4 right, after some hesitation
  5 right, with perfect recall
rule: SM-2: a new card has ease 2.5. Grade each recall 0-5. Grade 3 or more: the next interval is 1 day, then 6 days, then the last interval x ease, rounded; grade below 3: start the card again at 1 day. After each grade, ease += 0.1 - (5-q) x (0.08 + (5-q) x 0.02), never below 1.3. Any card graded below 4 is repeated again the same day until it scores 4 or 5.
source: Wozniak 1990, SuperMemo algorithm SM-2 (first used 1987; as published by SuperMemo); Roediger & Karpicke 2006, Psychological Science 17(3):249-255 (the testing effect)
limit: SM-2 is a 1987 rule of thumb, not a model fitted to your memory; newer schedulers fit memory data better. Grades are self-reported. Wozniak's text can be read as leaving the ease unchanged on a failed card (--keep-ef-on-fail); this machine lowers it by default, as most published versions do. Dates are targets: a late review still counts.

review · THE DAILY REVIEW LOAD, PROJECTED

$ tutor.py review deck.csv --today 2026-10-09 --grade uno=5 --grade dos=4 … (20 grades)
SESSION 2026-10-09 · 20 grade(s)
- uno: grade 5 (right, with perfect recall) · ease 2.50 -> 2.60 · reps 1 · next in 1 day(s), due 2026-10-10
- dos: grade 4 (right, after some hesitation) · ease 2.50 -> 2.50 · reps 1 · next in 1 day(s), due 2026-10-10
- tres: grade 4 (right, after some hesitation) · ease 2.50 -> 2.50 · reps 1 · next in 1 day(s), due 2026-10-10
- cuatro: grade 3 (right, with serious difficulty) · ease 2.50 -> 2.36 · reps 1 · next in 1 day(s), due 2026-10-10 · AGAIN TODAY
- cinco: grade 5 (right, with perfect recall) · ease 2.50 -> 2.60 · reps 1 · next in 1 day(s), due 2026-10-10
- seis: grade 2 (wrong, but it seemed easy once shown) · ease 2.50 -> 2.18 · reps 0 · next in 1 day(s), due 2026-10-10 · AGAIN TODAY
- siete: grade 4 (right, after some hesitation) · ease 2.50 -> 2.50 · reps 1 · next in 1 day(s), due 2026-10-10
- ocho: grade 4 (right, after some hesitation) · ease 2.50 -> 2.50 · reps 1 · next in 1 day(s), due 2026-10-10
- nueve: grade 5 (right, with perfect recall) · ease 2.50 -> 2.60 · reps 1 · next in 1 day(s), due 2026-10-10
- diez: grade 1 (wrong, but it was familiar once shown) · ease 2.50 -> 1.96 · reps 0 · next in 1 day(s), due 2026-10-10 · AGAIN TODAY
- once: grade 4 (right, after some hesitation) · ease 2.50 -> 2.50 · reps 1 · next in 1 day(s), due 2026-10-10
- doce: grade 3 (right, with serious difficulty) · ease 2.50 -> 2.36 · reps 1 · next in 1 day(s), due 2026-10-10 · AGAIN TODAY
- trece: grade 5 (right, with perfect recall) · ease 2.50 -> 2.60 · reps 1 · next in 1 day(s), due 2026-10-10
- catorce: grade 4 (right, after some hesitation) · ease 2.50 -> 2.50 · reps 1 · next in 1 day(s), due 2026-10-10
- quince: grade 4 (right, after some hesitation) · ease 2.50 -> 2.50 · reps 1 · next in 1 day(s), due 2026-10-10
- dieciséis: grade 2 (wrong, but it seemed easy once shown) · ease 2.50 -> 2.18 · reps 0 · next in 1 day(s), due 2026-10-10 · AGAIN TODAY
- diecisiete: grade 5 (right, with perfect recall) · ease 2.50 -> 2.60 · reps 1 · next in 1 day(s), due 2026-10-10
- dieciocho: grade 4 (right, after some hesitation) · ease 2.50 -> 2.50 · reps 1 · next in 1 day(s), due 2026-10-10
- diecinueve: grade 4 (right, after some hesitation) · ease 2.50 -> 2.50 · reps 1 · next in 1 day(s), due 2026-10-10
- veinte: grade 3 (right, with serious difficulty) · ease 2.50 -> 2.36 · reps 1 · next in 1 day(s), due 2026-10-10 · AGAIN TODAY
again today, until each scores 4 or 5: cuatro, seis, diez, doce, dieciséis, veinte
wrote deck.csv
rule: SM-2: a new card has ease 2.5. Grade each recall 0-5. Grade 3 or more: the next interval is 1 day, then 6 days, then the last interval x ease, rounded; grade below 3: start the card again at 1 day. After each grade, ease += 0.1 - (5-q) x (0.08 + (5-q) x 0.02), never below 1.3. Any card graded below 4 is repeated again the same day until it scores 4 or 5.
source: Wozniak 1990, SuperMemo algorithm SM-2 (first used 1987; as published by SuperMemo)
limit: SM-2 is a 1987 rule of thumb, not a model fitted to your memory; newer schedulers fit memory data better. Grades are self-reported. Wozniak's text can be read as leaving the ease unchanged on a failed card (--keep-ef-on-fail); this machine lowers it by default, as most published versions do. Dates are targets: a late review still counts.

review · THE DAILY REVIEW LOAD, PROJECTED

$ tutor.py review deck.csv --forecast 30 --today 2026-10-09
FORECAST · 200 cards · 30 days from 2026-10-09 · if every card is graded 4 (an assumption)
2026-10-10 Sat   40  ########################################
2026-10-11 Sun   43  ###########################################
2026-10-12 Mon   40  ########################################
2026-10-13 Tue   40  ########################################
2026-10-14 Wed   40  ########################################
2026-10-15 Thu   40  ########################################
2026-10-16 Fri   57  #########################################################
2026-10-17 Sat   63  ############################################################
2026-10-18 Sun   60  ############################################################
2026-10-19 Mon   40  ########################################
2026-10-20 Tue   20  ####################
2026-10-21 Wed   20  ####################
2026-10-22 Thu   20  ####################
2026-10-23 Fri   20  ####################
2026-10-24 Sat   20  ####################
2026-10-25 Sun   20  ####################
2026-10-29 Thu    1  #
2026-10-30 Fri    5  #####
2026-10-31 Sat    9  #########
2026-11-01 Sun   25  #########################
2026-11-02 Mon   20  ####################
2026-11-03 Tue   20  ####################
2026-11-04 Wed   20  ####################
2026-11-05 Thu   20  ####################
2026-11-06 Fri   20  ####################
2026-11-07 Sat   20  ####################
total 743 reviews · peak 63 · average 24.8 a day
rule: SM-2: a new card has ease 2.5. Grade each recall 0-5. Grade 3 or more: the next interval is 1 day, then 6 days, then the last interval x ease, rounded; grade below 3: start the card again at 1 day. After each grade, ease += 0.1 - (5-q) x (0.08 + (5-q) x 0.02), never below 1.3. Any card graded below 4 is repeated again the same day until it scores 4 or 5. The forecast replays that rule forward with one assumed grade.
source: Wozniak 1990, SuperMemo algorithm SM-2 (first used 1987; as published by SuperMemo)
limit: SM-2 is a 1987 rule of thumb, not a model fitted to your memory; newer schedulers fit memory data better. Grades are self-reported. Wozniak's text can be read as leaving the ease unchanged on a failed card (--keep-ef-on-fail); this machine lowers it by default, as most published versions do. Dates are targets: a late review still counts. Real grades vary; the forecast is a sketch of the load, not a promise.

calendar · THE PLAN OR THE REVIEW DATES AS AN .ICS CALENDAR FILE (RFC 5545)

$ tutor.py calendar --deck deck.csv --assume-grade 4 --days 30 --out reviews.ics --stamp 20261009T120000Z
wrote reviews.ics: 27 event(s) · 27 review day(s) from deck.csv (projected, if every card is graded 4) · all-day events
checked: CRLF line ends, lines folded at 75 octets, every BEGIN closed, every event has UID, DTSTAMP and DTSTART
rule: one event per study session (or per review day); all-day unless --time is given, then sessions back to back with 10-minute breaks; text escaped and long lines folded as the standard says.
source: RFC 5545 (2009), Internet Calendaring and Scheduling Core Object Specification (iCalendar)
limit: times are floating (no time zone): a calendar shows them at the same clock time wherever you are. Only the next due date of each card is known; later review dates depend on the grades, so re-export after each session or use --assume-grade for a projection.

ladder · THE HINT LADDER THE CARD CLIMBS, RUNG BY RUNG

$ tutor.py ladder --example fractions
THE HINT LADDER · the least help that works, one rung at a time
0 TRY — The learner tries first, alone. Ask for their attempt or their first step, even a guess: "What have you tried? Show me where you got to."
1 A QUESTION — Ask one question that points at the key idea without naming the step: "What do you notice about ...?"
2 A NUDGE — Point to where to look, still without the step: "Look at the bottom numbers. What would have to be true before you could add?"
3 A WORKED STEP OF A SIMILAR PROBLEM — Work one step of a different, similar problem out loud, then hand the learner's own problem back.
4 THE METHOD — Name the method in general steps, without doing the learner's numbers or words for them.
5 THE ANSWER — Show the answer, with the reasoning, only after an honest try at the rungs above, or when the learner, having tried, asks to see it.
THEN, ALWAYS: the learner explains it back in their own words (`tutor.py explain`).
THEN: a fresh similar problem, with no hints, to show the step is theirs. Log the try with the rung reached (`tutor.py mastery --add`).

how to climb:
- Climb one rung at a time, and only after the learner has tried at the rung they are on.
- Wait. Silence after a question is thinking time; do not fill it with the next rung.
- Go back down when the learner gets going again: the aim is the least help that works.
- Praise the effort and the method used, not being clever.
- Never jump to the answer because it is faster; go to it only when the learner has made an honest try and asks to see it.

example: 1/3 + 1/4
TRY: "Have a go. What do you think 1/3 + 1/4 is?" (A common first try is 2/7: adding tops and bottoms.)
A QUESTION: "Are thirds and quarters the same size of piece?"
A NUDGE: "We can only add pieces of the same size. What size of piece could both thirds and quarters be cut into?"
A WORKED STEP OF A SIMILAR PROBLEM: "Here is 1/2 + 1/5. Halves and fifths both cut into tenths: 1/2 = 5/10 and 1/5 = 2/10, so 5/10 + 2/10 = 7/10. Now try yours."
THE METHOD: "Find a number both bottoms divide into, rename each fraction with that bottom, then add the tops."
THE ANSWER (after an honest try): "Both 3 and 4 go into 12: 1/3 = 4/12, 1/4 = 3/12, so 4/12 + 3/12 = 7/12."
checked with exact fractions: 1/3 + 1/4 = 7/12; 1/2 + 1/5 = 7/10
rule: never the final answer before an honest try; one rung at a time: try, a question, a nudge, a worked step of a similar problem, the method, the answer; then explain back and a fresh problem without help.
source: Wood, Bruner & Ross 1976, Journal of Child Psychology and Psychiatry 17:89-100 (tutoring and scaffolding); Kornell, Hays & Bjork 2009, Journal of Experimental Psychology: Learning, Memory, and Cognition 35(4):989-998 (a failed try before the answer); Sweller & Cooper 1985, Cognition and Instruction 2(1):59-89 (worked examples); Bjork 1994, Memory and metamemory considerations in the training of human beings, in Metcalfe & Shimamura (eds), Metacognition, MIT Press, 185-205 (desirable difficulties)
limit: the ladder is this card's method built on those findings, not a protocol tested as a whole. The studies were in labs and classrooms; a learner who is upset or out of time needs a shorter ladder, and some will need the worked example sooner.

mastery · RIGHT|WRONG [--RUNG N] [--DATE DATE] LOG ONE TRY

$ tutor.py mastery homework-log.csv
MASTERY · 4 skill(s) · 20 tries · mastered = at least 80% right over the last 5 tries (a choice); a try counts as right only when it was reached by rung 2 or lower (a choice)
skill                                attempts   right    last N  share  last seen   status
equivalent fractions                        5       4       4/5    80%  2026-10-08  MASTERED
adding fractions, same bottom               5       5       5/5   100%  2026-10-08  MASTERED
adding fractions, different bottoms         5       1       1/5    20%  2026-10-07  REVISIT (2 right with heavy hints)
times tables, 7s                            5       3       3/5    60%  2026-10-09  REVISIT

revisit first: adding fractions, different bottoms (1/5); times tables, 7s (3/5)
mastered, still worth one spaced check later: equivalent fractions; adding fractions, same bottom
rule: per skill, share = right tries in the last N / N. REVISIT when the share is under the threshold; MASTERED when it is at or over it with at least N tries; NOT YET otherwise. A right answer reached only after the method or the answer was shown is not counted as right.
source: Bloom 1968, Learning for Mastery, Evaluation Comment 1(2) (a mastery criterion before moving on); Cepeda, Pashler, Vul, Wixted & Rohrer 2006, Psychological Bulletin 132(3):354-380 (spacing), for why mastered skills still need review.
limit: the threshold and the window (80% over the last 5 by default) are choices, not research findings. A share over 5 tries is a small sample; one lucky or unlucky try moves it by 20 points. Mastery fades without review.

explain · THE EXPLAIN-BACK FORM THE LEARNER FILLS WHICH KEY POINTS AN EXPLANATION COVERS, AND WHET

$ tutor.py explain --topic photosynthesis
# EXPLAIN IT BACK · photosynthesis
Fill each part from memory first. Then check your notes and mark what you missed.

## 1. IN MY OWN WORDS
Explain it as if to a friend who missed the lesson. No copying from the book or the notes.

...

## 2. AN EXAMPLE OF MY OWN
One example that is not the one in the book.

...

## 3. WHY IT WORKS
The "because": why is it true, or why does the method work?

...

## 4. WHERE IT GOES WRONG
A common mistake, or a case where it does not apply.

...

## 5. THE GAPS
Any word I used but could not define, or a step I could not explain.

...

## 6. HOW SURE I AM
1 (guessing) to 5 (could teach it). Then check against the source, and say it again where it was wrong.

...

## For the tutor
- Which key points are there? Which are missing? (`tutor.py explain --check`)
- Is it in the learner's own words, or the book's?
- Ask about each gap with a question, not the answer. Then the learner explains it again.
rule: the learner explains from memory, in their own words, with an example, the why and a common mistake; gaps are named and asked about.
source: Chi, de Leeuw, Chiu & LaVancher 1994, Cognitive Science 18:439-477 (self-explanation); Dunlosky, Rawson, Marsh, Nathan & Willingham 2013, Psychological Science in the Public Interest 14(1):4-58; Roediger & Karpicke 2006, Psychological Science 17(3):249-255 (the testing effect). "Explain it simply" is often called the Feynman technique: a popular name, not a study.
limit: self-explanation helped in the studies, rated moderate utility overall; it takes time, and works best with someone to check it against the source.

explain · THE EXPLAIN-BACK FORM THE LEARNER FILLS WHICH KEY POINTS AN EXPLANATION COVERS, AND WHET

$ tutor.py explain --check explanation.txt --points points.txt --source source.txt
EXPLAIN-BACK CHECK · 1 of 5 key points covered · 50 words
- MISSING light energy is captured by chlorophyll (25% of its words) · not found: energy, captur, chlorophyll
- COVERED carbon dioxide and water are the raw materials (60% of its words)
- MISSING glucose (a sugar) is made (33% of its words) · not found: glucose, made
- MISSING oxygen is released (50% of its words) · not found: releas
- MISSING it happens in the chloroplasts (50% of its words) · not found: chloroplast
longest run copied word for word from the source: 2 words
next: ask about each MISSING point with a question (rung 1), not the answer; then the learner explains it again.
rule: a key point counts as covered when at least 60% of its content words (stemmed, common words dropped) appear in the explanation, alone or inside a longer word (a choice); 12 or more words in a row from the source reads as copied (a choice).
source: Chi, de Leeuw, Chiu & LaVancher 1994, Cognitive Science 18:439-477 (self-explanation); Dunlosky, Rawson, Marsh, Nathan & Willingham 2013, Psychological Science in the Public Interest 14(1):4-58 (self-explanation: moderate utility)
limit: a word match is not understanding: a learner can name every point and still not get it, or say it rightly in other words. Read the explanation; this only says where to look.

evidence · WHAT THE RESEARCH SAYS WORKS, AND WHERE IT STOPS

$ tutor.py evidence
WHAT THE RESEARCH SAYS WORKS · utility as rated by Dunlosky et al. 2013 where they rated it
- practice testing (retrieval) [HIGH]: Quizzing yourself beat restudying on tests two days and a week later, though restudying looked better after five minutes and felt surer (Roediger & Karpicke 2006).
- distributed practice (spacing) [HIGH]: Spreading the same study over days beats cramming; the best gap grows with how long you need to remember it (Cepeda et al. 2006, 2008).
- interleaved practice (mixing) [MODERATE]: Mixed maths problems scored 63% a week later against 20% for blocked practice, while scores during practice were the reverse, 60% against 89% (Rohrer & Taylor 2007, experiment 2).
- self-explanation [MODERATE]: Explaining each step to yourself helped understanding (Chi et al. 1994).
- elaborative interrogation [MODERATE]: Asking "why is this true?" helped learning of facts in the studies reviewed.
- trying before being told [-]: A wrong guess before the answer improved later recall over just studying the answer (Kornell, Hays & Bjork 2009).
- rereading, highlighting, summarising [LOW]: Feel productive, did little on their own in the studies reviewed.

the common thread: desirable difficulties. What feels harder while practising (recalling, spacing, mixing) often sticks better; what feels easy (rereading) often fades (Bjork 1994).
in practice: close the book and recall, then check; review the same thing again after a day, then after a week; mix problem types; explain it back; keep a record of what you get right.
rule: techniques ranked by the evidence reviewed; each study named.
source: Dunlosky, Rawson, Marsh, Nathan & Willingham 2013, Psychological Science in the Public Interest 14(1):4-58; Roediger & Karpicke 2006, Psychological Science 17(3):249-255 (the testing effect); Cepeda, Pashler, Vul, Wixted & Rohrer 2006, Psychological Bulletin 132(3):354-380 (spacing); Cepeda, Vul, Rohrer, Wixted & Pashler 2008, Psychological Science 19:1095-1102 (best gap against test delay); Rohrer & Taylor 2007, Instructional Science 35:481-498 (shuffled maths practice); Chi, de Leeuw, Chiu & LaVancher 1994, Cognitive Science 18:439-477 (self-explanation); Kornell, Hays & Bjork 2009, Journal of Experimental Psychology: Learning, Memory, and Cognition 35(4):989-998 (a failed try before the answer); Bjork 1994, Memory and metamemory considerations in the training of human beings, in Metcalfe & Shimamura (eds), Metacognition, MIT Press, 185-205 (desirable difficulties)
limit: effects vary by subject, age, material and test; most studies used college students and short lab tasks, and the mixing evidence comes mostly from maths and learning categories. None of this guarantees a grade.

THE RUNS ARE PASTED FROM THE CARD’S OWN WITNESS.MD AS PRINTED. FIGURES MARKED FAKE ARE THE STUDIO’S OWN EXAMPLES, INVENTED FOR THE RECORD; PUBLISHED FIGURES ARE NAMED WITH THEIR SOURCES. NO PRIVATE PERSON IS NAMED.

QUESTIONS AT THE TABLE

Asked before the wire.

What does it need from me?

Name. Ask what the work is and what it must agree with.

Does it send anything anywhere?

No. The machine is a standard-library Python program that runs on your own computer and sends nothing anywhere; the words run wherever an AI reads Markdown. No server, no account, no key.

What is on the record?

The self-test (33/33) and 7 of its 7 commands run on real inputs, pasted into WITNESS.md as they printed; the punch on the face is the SHA-256 of SKILL.md, so a downloaded copy can be checked against this room.

Can it be wrong?

Yes — it is an AI reading a text file. Plans, review gaps and the mastery threshold are planning choices, labelled as such, and SM-2 is a 1987 rule of thumb. The research findings are real but vary by subject, age and learner, and none guarantees a grade. It never writes graded work to hand in or takes a test for anyone, collects no child's personal data, does not diagnose learning difficulties, and is not therapy.

What will it never do?

← EVERYDAY LIFE · THE FLOOR →