Dominic and Major

Two ways to turn a two-digit number into something you can see. They are usually presented as rivals. They are better read as answering different questions, and this corpus wants both answers.

The mechanism, side by side

Major maps each digit to a consonant sound. Vowels, and w, h, y, carry nothing, so you are free to pack them in wherever a word needs them.

42 is r-n, so RAIN, RHINO, RUIN, WREN. Any of them. The system hands you a choice of objects.

Dominic maps each digit to a letter, and the pair is somebody's initials.

42 is D·B, so David Bowie. The system hands you a person — and, because a person can do things, an action and an object as well.

The same number, both ways

Take 4215, a year you might want inside an era.

Major. 42 is RAIN and 15 is TAIL. Two objects, and you must invent a link: rain falling on a tail. It works, but the bond between two nouns is arbitrary, and arbitrary links are what decay first.

Dominic. 42 is David Bowie and 15 is Albert Einstein. You take the person from the first pair and the action from the second: David Bowie sticking his tongue out. Four digits, one image, and the image has a grammar — somebody doing something — which is the shape memory is built for.

That is the whole case for Dominic, and it is a strong one. Person-Action-Object takes it further: with an object slot too, one image carries six digits.

Where they actually differ

Major gives you options; Dominic gives you constraints. Every number has many valid Major words — this repo ships three verified alternatives for every cell. Dominic gives you two fixed initials and you take whoever exists. You cannot decide that 42 should be someone memorable; you can only search D·B.

Dominic cannot be checked by machine. A Major word either encodes to its number or it does not, and re-deriving it from its spelling is arithmetic — akasha lint does it for all 400 words in the corpus. Whether Daniel Barenboim is vivid to you is not a question a computer can answer.

Major's failure mode is ambiguity; Dominic's is collision. Is the g in "gig" hard or soft? Does the t in "match" count? Building the candidate list here, the verifier rejected 29 words on their first pass — nine were my arithmetic and the rest were silent letters and affricates I had waved through. Dominic never has that problem: letters are letters. Instead it has B·C — Bill Clinton or Bill Cosby? — and if you have not settled that in advance you lose the cell under pressure.

Dominic assumes a shared culture. Its cells are only as good as your population of famous people, and some pairs simply have none. Filling all 100 here, seven cells had no widely known figure at all: 01 O·A, 03 O·C, 05 O·E, 07 O·G, 55 E·E, 80 H·O and 84 H·D. Those have to come from your own life — a colleague, a relative, a character from something you love. That is not a flaw so much as the price of the method, but it is a price its promoters tend not to quote.

Major does other work. Because it encodes sounds rather than initials, it also spells: you can peg a word, a name or a phone number with the same table. Dominic only does numbers.

What competitive memorisers use

Dominic O'Brien won eight World Memory Championships with the system that carries his name, and the top of the field is broadly PAO — because for cards and long digit strings, compression per image is what wins. That is a strong signal for speed at scale, and a weak signal for anything else. This corpus is not trying to memorise a deck in thirty seconds. It is trying to make a few hundred numbers permanently available, which is a different problem and rewards vividness over throughput.

What to do here

Keep Major as the table, and build Dominic on top of it rather than instead of it.

The Major list already exists, is verified, and is the object layer. Dominic's people are the actor layer. Together they are a PAO system, and neither has to be thrown away:

• Objects — the settled Major word for the cell. Already drilled, both directions, and machine-checked.
• People — the Dominic candidate for the cell. New, and the thing to add if you want four-digit images.

Mixing them within one role is the thing to avoid. If 42 is RAIN on Tuesday and David Bowie on Thursday, decoding breaks — and decoding is half of what a peg table is for.

The idea worth stealing

Dominic's real insight is not the digit-to-letter table. It is that a pair of initials retrieves a person, and a person can act. That shape fits something else in this corpus exactly.

Country codes are already pairs of letters. HR, DE, PT, CZ. Give each country a person with those initials — Helen Reddy, David Essex, Portia Trent, whoever lands — and every country gains an actor. Then a multi-country event stops being a list and becomes a scene: two people, doing something, somewhere on a tour you can already trace.

That has been an open thread in this project for a long time, and Dominic is the technique that closes it. It needs 197 letter-pairs rather than Dominic's restricted ten-letter set, which is harder — but you only ever need the countries you are actually studying.

The list as it stands

The settled table has been through a rhyme pass. Major decodes by sound, so two pegs differing only in their first consonant are two pegs you can confuse — and that consonant is the digit you are trying to recover. The list shipped with SAIL, TAIL, NAIL, RAIL, JAIL: five cells whose only difference was the thing being decoded.

akasha rhymes reports what is left. Thirteen cells moved, and the worst cluster went from five to three:

The x5 column now reads SEAL, TOOL, NAIL, MULE, ROLL, LILY, JELLY, COAL, FILE, BELL — ten different vowels. Two swaps created new pairs and were taken further: RAKE rhymed with CAKE at 77, so 47 is RAG, which suits a plague and a partition better anyway; PET rhymed with NET at 21, so 91 is BEAD, and fifteen beads coming off a broken necklace is a better 1991 than a bat was.

Three individually weak pegs went at the same time: GNOME needed you to remember the g was silent, CHOO-CHOO was hyphenated, and VIP was an initialism rather than a word. Seven pegs still rely on a pronunciation exception — COMB, TOMB, LAMB, KNOB, KNIFE, MATCH, NOTCH — and a test pins that list so it cannot quietly grow.

Choosing

pegs/candidates.md holds the options: three verified Major words and two or three Dominic people for every cell, with an action attached to each person because a person without one only encodes two digits.

akasha peg major.42                 # show the shortlist for 42
akasha peg major.42 --pick 2        # take one
akasha peg major.42 "your own"      # better than either

The last of those is the one that matters. A borrowed peg is measurably weaker than a generated one, and this corpus shipped with a hundred borrowed ones. Options exist to close that gap cheaply: picking from a shortlist is nearly as good as inventing, and unlike inventing, it actually gets done.