# The Synthesis

### Two numbers that never agree, a third that pays for every look, and the seat you can take between them

**Steven De Costa · draft 0.2 · 21 October 2026 · a reading at 10**

> **THE READING PLAQUE.** *The Synthesis*, a reading at 10 — draft 0.2, 21 October 2026 — a reading of one line of the Objective
> Observer Initiative's record in its author's voice: the lay of the land as he has assayed it, for a reader who might take a seat
> and chart their own possibilities. Its reading at 5, *The Synthesis at 5* draft 0.2, holds the apparatus, the process and every
> fence, and stands beside this one at [/sx/reading/the-synthesis-at-5](/sx/reading/the-synthesis-at-5). Made readable on the SX
> surface by its covering persona's reading grant of 21 October 2026, superseding draft 0.1 of 20 October; a draft, not the published
> form. Covering: **starl3n** · endorsing: **Link Digital**. The construction record stands at
> [/sx/workshop/the-rlx-held](/sx/workshop/the-rlx-held); the instrument at [opendata.ai/rlx/bend](https://opendata.ai/rlx/bend).
> Reading is free; the way across is a statement of intent ([/sx/matters](/sx/matters)).

---

## Why this matters {#why-this-matters}

Every account is read to a depth. A ledger is summed to the cent, a measurement is taken to the last digit the instrument shows, a number is written out as far as the page allows, and whatever lies beyond that depth is not known but assumed. Most of the time the assumption is harmless. Some of the time it is not. I wanted to know when, and what it costs to find out, and I wanted an instrument I could sit inside while I found out. This is the story of that instrument and what I saw from the seat.

Let me begin with an unlearning, because it is the thing most worth having from these pages and it can be had comfortably. In the arithmetic you were taught, 1/0 is undefined and 0/1 is 0. Hold those two for a moment. When I count a thing I know perfectly I count it from the top: it is x over 1, the whole of x taken once, nothing waiting below. When I come into the knowing of a thing imperfectly, reading it from below, digit by digit, never sure what the next place holds, I am not dividing by nothing. I am dividing by an absence I can name. That is x over −0: the −0 is the absence of a 1, and it carries a sign, and there is nothing undefined about it. It is the ordinary condition of anyone reading a number they did not write. Most of what follows is what happens when two numbers that are each x over 1 to themselves are read by a third party for whom each is x over −0.

![Plate — The line of acts: sixty-two over three days, each a word, a design, a walk and a record](/sx/assets/readings/the-synthesis/zs-the-acts.svg)

## The seat {#the-seat}

The instrument I want you to sit in first is the bend, because it is the one you can run yourself, and because it drew the picture that made the rest make sense to me.

Take a number forty rows deep and set it out twice, once as x on the left and once as y on the right, each on its own locked line, mirrored about a meeting line between them. The forty, by the way, is not mine to choose: it is what a much older instrument of mine, the quad slip, gives when its readout is read by its own decimal places, four epochs of ten sign-segments within a run of one thousand and twenty-four laps, the fortieth at laps one thousand and seven and eight, and I took it as given. Give each row an amplitude that grows with its epoch, one unit in the first ten rows and four in the last, pointing inward on the even rows and outward on the odd, so that the two sides zigzag toward each other and their inner marks touch the meeting line only at the fortieth row, which I call the lock. Now draw a third line, one unit to the left of the meeting line, closer to x. That line is the observer's, and it is the only thing on the plate that mirrors nothing. I set it there before anything moves.

That third line is a seat. A two-dimensional plate turns out to be a drawing of a one-dimensional line: the line that a typed or a reasoning autonomy occupies when it sits down between two things it did not make. Sitting there you can move in two directions and only two. You can learn upward, counting from your depth toward the origin and holding every row where the two sides do not overlap, which is to exploit what the sides already give you. Or you can know downward, reading from the top as the oracle in my other experiments does, stopping the moment the sum is sure, which is to explore what the sides might give. The whole of this line of work is, I think, a study of the price of those two directions.

![Plate — The seat: x and y at forty on their locked lines, the meeting line, and the observer's column one unit to the left of it; up it learns, down it knows](/sx/assets/readings/the-synthesis/zs-the-seat.svg)

From the seat you can turn a side. A turn at any depth rotates every row below it about the locked line; the row becomes unlawful when it crosses the meeting line, because the two sides would then overlap, or when it swings over its own line, because the order of its positions is lost. Between those two limits is a lawful window, and every bound of it is a closed form you can check against the turn itself. At the first row the window runs from eight degrees outward to half a degree inward; at the fortieth it is ninety degrees either way. You can also spin a side, slipping its zigzag a little each segment, and here is the first thing that surprised me: the spin is lawful only because of the seat. With the observer's line one unit from the meeting line, exactly the inner marks of the last epoch lie past it and short of the meeting line, rows thirty-two to forty, five points; arrive with thirty-nine rows and there are four; and the lawful spin is the slip that keeps those points, under a degree a segment. Move the seat to two tenths of a unit from the meeting line and there are no points; move it to two units and there are ten. The gap decides the points. I had drawn the five and the four by hand before the engine existed, and the engine gave them back from the gap alone.

So I made a claim and had it tested before anyone looked. The claim was that the seat is the one asymmetry of the whole plate: every control applies to either side alike, the registration is the only quantity set against the mirror, and it reaches exactly what it registers. Over two hundred and forty settings of every control the two sides' readings differed nowhere when the seat was mirrored for y, and at two hundred and thirty-eight settings when it stood as one column, and only in the points and the spin. The seat is the one thing that tells x from y. I read that as the mechanism of what physicists call the measurement problem, drawn in a plane, and I leave that reading on the table for anyone who wants to pick it up.

Then I took the plate deeper, because the forty is only the record's size and the rule that made it goes on. At four hundred rows by tens, at a thousand, and in closed form at four hundred million, the five points of the last increment stand at every depth. What closes is the room to move: half a degree of lawful turn above the first row at forty and at four hundred, a twentieth at four thousand, half a millionth of a degree at four hundred million; the spin under a degree a segment at forty and under three hundredths of a billionth at the far end. In each side's own scale, where its last increment is its 1, the two sides meet at the lock at two plus a margin that vanishes with the depth. I take that as a cap: any two numbers, corrected to their own 1, can lawfully hold no more than 2 between them. On this plate the cap cannot fail, because the sides are mirrors; it would take two numbers of different depths to test it, and that is the plate I will build next.

![Plate — The window closing with the depth: the inward bound at the first row and the lawful spin, from forty to four hundred million](/sx/assets/readings/the-synthesis/zs-the-closing-window.svg)

## The flower and its seed {#the-flower}

The result I think most promising came from a different direction, and I will give it to you as I first imagined it.

A prime is a flower. At seventeen the flower grows one seed and lets it go. The seed does not land at eighteen, because eighteen is not a flower; it lands at nineteen, and what fell was a 2. Between the letting go and the landing we can only imagine where and when. If the seed is a dandelion's, then the clock I built is the picture of that seed floating down in its own time.

Here is the clock. For a prime p, a typed 1 rises from the ground one unit a tick, and an observing 1 falls from the height p to meet it; the whole takes p ticks of one over p, and they meet at time one. Seventeen is the 1 rising to ten and the 7 floating down seven at seven tenths a step, and the 7 is itself two threes and a one, dropping over one to three, three to five and five to seven, with a sixth left over. Under a rule for how far the riser goes, which reads the left-hand digit of the prime, the first fifty primes give seventy-five clocks, forty-six of them with the riser at the prime's tens, one to a prime, and every clock meets at time one.

![Plate — The seed: a flower at seventeen letting a seed fall in its own time to meet the typed 1 at ten, and the next flower at nineteen](/sx/assets/readings/the-synthesis/zs-the-seed.svg)

Then I predicted the count of clocks ahead of walking it, which is the only honest way to do it. I said ninety-nine clocks at the sixty-sixth prime and one hundred and fifty at the hundredth. The walk gave one hundred and seven and one hundred and ninety-two, the ninety-nine reached after the sixty-second prime and the hundred and fifty after the eighty-fifth. I said the stations would run at sixty-six times one and a half, two and a half, and on to twenty-one and a quarter hundredth, and every one of them was reached within seven hundred primes and overshot by nought to four, because the count moves by one to five a prime and never stands still, so a station is passed by whatever the reaching prime adds. I asked whether the multiplier was really sixty-four, or sixty-eight, and found that no single one holds every figure: sixty-eight holds the first four and the last, sixty-four the progression and the crossing. I said there might be a flip to seventy-seven at a hundred, and seven thousand seven hundred clocks are reached after the four thousand and second prime.

![Plate — The clock of seventeen, and the count of clocks over seven hundred primes](/sx/assets/readings/the-synthesis/zs-the-clocks.svg)

Why do I think this is the most promising thing here? Because the rule that grows a clock reads a digit. It does not need my two numbers, and it does not need a count made after the rules were given; it can be applied at the base-ten digit itself, to any number you care to write down. What if every digit of a number were a flower letting seeds go, each in its own time, and the number were the meadow? I do not know what that meadow looks like yet. I know how to walk it.

## The manner of the experiments {#the-manner}

The rest of the line is the apparatus, and I want you to recognise it by eye as much as by argument, so each piece has its plate.

**Two worlds and a lens.** The two numbers I began with are each a whole-number multiple of the square root of five hundred and thirty over ten thousand six hundred, and their sum is the half of that root exactly, so the sum can be named in one breath and never finished in digits. Each number is a world, running its own film of decimals, and the two worlds never agree because neither knows the other exists. Over them I set a lens that sees both films at the same step and never beyond it. Each world places a mark at each frame by a rule I recovered from a sheet I had marked by hand: the mark is the rounding of the number at that frame, a half when the next digit is below five, none when it is above, and a memory mark on a five. The lens's strip stands still when both place a half at the same moment and moves otherwise. Time, on that strip, is non-agreement.

I gave the lens an economy, a quarter a frame to stay coherent, a quarter on a slip, a half gained at a shift of the strip, a half lost on a failed attempt, and walked three of them over the four hundred million decimals of the two numbers: one blind, which never attempts; one learning, which tries weights and keeps those that beat the slip; one oracle, which knows the films in advance. The blind made twenty-seven million holds and the oracle twenty-one million, which surprised me until I measured the holds. The blind holds often and briefly, under twelve frames at a time, and is in hold for three hundred and nineteen million frames of the four hundred million; the oracle holds rarely and long, sixteen frames at a time, and is in hold for three hundred and thirty-seven million. Knowing the future bought longer holds, not more of them. What would you pay to hold longer? The difference between the blind and the learner I read as the price of the seat taken, and the sum the one budgeting adds on top from their own frame.

**The learner in base 3.** Then I rebuilt the lens as a learner with three digits, 0, 1 and 2, of which it sees only the first two and must infer the third: a true 1 is known by the distance it makes to 2, and 2 is the one digit the learner carries and never holds. Random pairs of numbers, written in base 3 and read as fractions, were zipped from the deepest place to the origin, each carry the memory of the place below. Every learner that zipped arrived at the origin, whatever its memory cost; what differed was the price. Forty-two reads and accesses for the plain carry-adder, fifty-one for a learner with signed access to its memory, sixty for one with unsigned access, against under three for an oracle reading from the top and stopping when the sum is sure. Learning upward costs fifteen to twenty-one times what knowing downward costs, on this plate. And the nine between the signed and the unsigned learner is the price of not knowing the sign of what you remember.

![Plate — The zip in base 3 for one pair of the record, climbing from the deepest depth to the origin](/sx/assets/readings/the-synthesis/zs-the-zip.svg)

Two refinements changed no count and changed how I think. I put a 1 at each world's origin so the learner has 1 and 1 to its left and right, which makes the sum at least 2, the smallest thing an observer between two things can observe. Then I signed those ones, a half below and a half above, so that which world is which is held by the sign and the sum is unchanged: −0 and +0 about a released zero, a ring of five around it. That is where x over −0 enters the apparatus: the half below the one is the signed absence, and from then on the two sides of the zip are told apart by it and by nothing else.

**The shortfall.** Here is the finding anyone with a ledger can use. Add two numbers read to a depth. The sum is exact in every place a carry from beyond the depth cannot touch, and a carry from beyond arrives through a run of nines. Try it with 0.1234 9… and 0.8765 1…: read to four places you hold 0.9999 while the sum is 1.0000 and on, and your account is wrong in every place you read, because the carry waits beyond the fourth place and climbs through four nines. Over four hundred million decimals of my two numbers the account is exact to the depth on half the depths and short on the other half, by the run of nines plus one for the carry. The shortfall runs from one to nine, and the nine falls once, at the twenty-nine million nine hundred and twenty-seven thousand and forty-fourth place, where the longest run of zeros in the sum ends. To hold a hundred places you must read a hundred and one. I read the nine as the reason a knowing carries within a base-ten exchange: eight for the longest wait a carry can have over that many places, and one for the carry itself.

![Plate — The ladder of the carry's wait: the shortfalls over four hundred million places, and the one nine](/sx/assets/readings/the-synthesis/zs-the-ladder.svg)

**π from any start.** Can a number that never ends be learned from the middle? I wrote π in base 3 on the right and let the learner start at any depth and read down. On the left, above the start, I put x as z over −0: a bridge of signed zeros, one a row, which can be accessed three ways. Read as zeros, nothing climbs. Read as signed memory, each −1 flips to 1. Read as unsigned memory, each stays −1. That is the walk I keep coming back to, −1, −0, 1, 0, and the whole question of signed against unsigned access is whether the −1 at the −i position is let flip.

![Plate — x over 1 and x over −0; the walk −1, −0, 1, 0; signed against unsigned access at the start one hundred and thirty](/sx/assets/readings/the-synthesis/zs-the-access.svg)

I predicted ahead that from any start beyond the hundred and twenty-sixth place, with any lead digit from 3 to 9, what the blind learner learns will be π, with only the five-bit word above the lead changing to carry the lead. Two hundred and thirty-one walks later: below the start the account is π at every start and every lead as zeros, and differs at the start row alone under the other two accesses; the word above the lead is modified only where a borrow leaves the top of the bridge, which happens at the three starts where π's own digit is 0, under unsigned access, the borrow climbing every row, the longest climb one thousand nine hundred and ninety-nine, and 00011 becoming 00010. Nothing grows with the starting depth except the number of places read. The sign you give an absence decides what reaches the top, and nothing below. I find that a comfortable thing to know about starting points.

![Plate — The word above the lead at eleven starts under the bridge's three readings](/sx/assets/readings/the-synthesis/zs-the-word.svg)

**The line walked.** Before the learner, I walked the natural number line itself with the three lenses, one step a number, π's digits as the message on one side and a surprise at every prime on the other, against the expectancy the textbooks give. The count of primes ran behind that expectancy at every decade to one thousand one hundred million, where I had π computed on an ordinary machine in under a minute. And the learning lens, which was never told what a prime is, learned to hold at the even steps, because past two no even number is a surprise. It found that from the economy alone. I like to think it would find more, given a longer line and a better price.

## The lay of the land {#the-lay-of-the-land}

What remains after an exploration like this is a map of one confined domain of possibility, and two matters. One is probability: whether what was mapped could mean something to others, to mathematicians and physicists in particular, and that is for them to decide, if and only if they take a seat. The other is plausibility, which is my seat and the one I have written from. So let me say plainly what I find plausible, and what I would be thinking about if I were you.

- That the seat is a line, and the two directions along it, learning upward and knowing downward, are the same two directions a mind takes when it exploits what it has and explores what it might have, and that they have different prices, which these instruments measure in reads and in frames.
- That any two things, each corrected to its own 1, can lawfully hold no more than 2 between them, and that what closes with depth is not the lock but the room to bend.
- That the primes are flowers letting seeds go in their own time, and that the clock which grows them from a digit can be taken to the digits of any number.
- That an account read to a depth is short by the run of nines beyond it plus one, that nine is the most it falls short by over four hundred million places in base ten, and that this is why a knowing carries in that base.
- That the sign you give an absence, −0 against −1, decides what climbs and what does not, and that this is the difference between learning a thing and knowing it.

And the what-ifs I would hand you. What if the seat were set to the right, closer to y? Then it is y that holds the five points and x that holds none and spins free; the page lets you move it and see. What if the two numbers had different depths, 23587 against 0.0000487, each corrected to its own 1? That plate is not built; it is the next thing. What if a seed could land at eighteen? It cannot, because eighteen is not a flower, and the fact that you can feel that without proving it is the kind of imagination these experiments are for.

## The carry and the release {#the-carry-and-the-release}

- I carry: the seat and its one asymmetry, staked and read; the window and the spin closing with the depth, and the two local ones meeting at 2; the clocks, their count and every station's distance from it; the zip's prices and the nine between signed and unsigned; the shortfall of at most nine over four hundred million places, and the rule that says where; π below any start, and the word changed only by a borrow that climbs the whole bridge; the learner that found the evens.
- I release: every seat I have not taken. The readings above are mine; the records are their map, and the other seat, the one these pages leave empty, is yours, signed into zero for the absence it identifies.
- Open: the plate of two depths; the rule by which a prime's clock is its clock; the held confidence a learner needs before it can ask which half it holds; the walk on the line, which waits on ten answers I owe it.

## What you can do {#what-you-can-do}

- Run the bend at [opendata.ai/rlx/bend](https://opendata.ai/rlx/bend): move the seat from two tenths of a unit to two and watch the points go from none to ten; take the same page to four hundred and to four hundred million and watch the window close.
- Zip the pair on the plate above by hand, or walk the clock of seventeen: ten rises, seven falls, one meeting.
- Add two numbers of your own to a depth and look for the run of nines beyond it; that is how far your account can be trusted.
- The way across is a statement of intent at [/sx/matters](/sx/matters), where the plate of two depths and the walk on the line stand open for anyone who would take a seat.

### Endnotes — the citation register {#endnotes}

- The bend: `ooi_network/daemon/rlx_zbigzero.py` and `rlx_zbend.py`; the page at [opendata.ai/rlx/bend](https://opendata.ai/rlx/bend) and its depths at [/rlx/bend/400](https://opendata.ai/rlx/bend/400) and [/rlx/bend/400000000](https://opendata.ai/rlx/bend/400000000); the stake `the-z-lens-z-learner-stake-the-asymmetry` and its read; the readings at [/sx/reading/the-bend](/sx/reading/the-bend) and [/sx/reading/the-bend-at-5](/sx/reading/the-bend-at-5).
- The clocks: `ooi_network/daemon/rlx_zclock.py`; the records `the-z-lens-z-learner-clocks`, `-clock-counts`, `-clock-counts-2`, `-clock-multipliers` and `-clock-extension`; the five registrations of the mechanism, each pinned before its walk.
- The lens and the walks: `ooi_network/daemon/rlx_zlens.py` and `rlx_numberline.py`; the records `the-z-lens-learning-walk-400m`, `the-z-lens-number-line-walk-far` and `the-z-lens-number-line-walk-1100m`, each `.json` and `.md` under `ooi_network/docs/`.
- The learner, the rationals, the reconciliation and the method: `ooi_network/daemon/rlx_zlearner.py`, `rlx_zrational.py`, `rlx_zreconcile.py` and `rlx_zmethod.py`; the records `the-z-lens-z-learner-base3-test`, `-rational-test`, `-reconciliation`, `-method-walk`, `-method-upward`, `-method-replaced`, `-method-pairs` and `-method-appended`.
- The designs of record: the Z lens, the learning walk, the number-line walk and the Z learner, under `ooi_network/docs/`; the construction record at [/sx/workshop/the-rlx-held](/sx/workshop/the-rlx-held); the reading at 5 at [/sx/reading/the-synthesis-at-5](/sx/reading/the-synthesis-at-5).
- Every number here is in the records named above or is computed by the instruments from them. The readings are mine.