# The Channel, held

> **THE READING PLAQUE.** *The Channel, held*, draft 0.2 — **made readable on the SX
> surface by its covering persona's reading grant; a draft, not the published form.** The
> grant stands by the operator's word (2026-09-16), verbatim: "ok - now do the same for the channel."
> Covering: **starl3n** · endorsing: **Link Digital** (the standing pair). The Foundations
> · 15: the fifteenth reading of the foundations series — an experiment of the observatory
> node, held on this node. The construction record stands in the open workshop at
> [/sx/workshop/the-channel-held](/sx/workshop/the-channel-held); the experiment's own
> door, [opendata.ai/channel](https://opendata.ai/channel), wears the word *grown · held*
> and links back here while this grant stands. Reading is free; the way across is a
> statement of intent ([/sx/matters](/sx/matters)). This file's sha256 is pinned in the
> reading register.

*Draft state — this block records the state of the draft; the reading begins after the
rule.* Draft 0.1 — 16 September 2026. Author of record: Steven De Costa; composed at his
word from the experiment's record. Standing: a reading, stated never asserted; the
record it reads stands whole without it. The later locks and the walk land in a further
draft. Draft 0.2 (16 September 2026) places 8 figures at the operator's word, each drawn from the record's own numbers and captioned with its source; the screenshots the operator will supply are not yet placed.

---

## What the Channel is

The ninth instrument of the series of ten reads the being as an information channel, in
the operator's framing and with the terms meant precisely. The input is the rate, which
sets the being, and the being is the signal. The confinement resolves it toward the '1'.
The output is read within an observation budget, a number of laps. Channel capacity and
surprise then stop being metaphors and become Shannon quantities that can be measured and
tested against a hypothesis. The Channel is that test, and then two more experiments laid
over it: a co-constructive channel of several instruments, and a windowed lattice the
record calls the epistemic grid.

## The measure

Every term is measured and nothing is fitted. The signal is the base-2 logarithm of the
being, the bits of being-resolution the channel offers. The reliability is the fold over
the ground's own coherent precision of 4.23, and it is the foam term: the fold's measured
decay drives the reliability toward zero at fast rates. The settled gate is whether the
being's horizon fits within the budget, and it is the indifference term: a being whose
horizon exceeds the observation window is not reliably resolved, and its capacity is
zero. The capacity at a rate is the product of the three, the settled gate times the
reliability times the signal, which is signal less noise.

## The band, the peak, and the window

The capacity over rate is a band. It is zero at indifference, too slow, the horizon past
the budget and the being unsettled; zero at foam, too fast, the being floored at 1 with no
signal; and finite between. That confirms the cooling picture quantitatively: a stable
band flanked by indifference and foam. The operator's hypothesis had been maximum
surprise with three co-constructive optima at rates of 0.25, 0.5 and 0.75. A fail-fast on
raw surprise refuted the three-peak reading, since surprise is monotonic in rate, and the
proper capacity gave the real answer: a single peak at the reliable edge of the band. At
the pinned 2¹⁰ baseline the peak sits at about 0.25, so 0.25 is the optimum at one hundred
percent of the peak, while 0.5 at about 88 percent and 0.75 at about 81 percent sit inside
the band and are not separate peaks. One peak, not a triplet, and 0.25 earns its place.

The headline is that the peak slides with the budget:

| budget, laps | 256 | 512 | 1024 | 2048 |
|---|---|---|---|---|
| peak rate | 1.0 | 0.5 | 0.25 | 0.125 |

The peak rate times the budget is about 256, which is 2⁸: the optimum is the fundamental octave over the observation window.
So 0.25 is the optimum for the pinned 2¹⁰ baseline and not an absolute; widen the window
and the optimum slides to a slower rate. It is a relativity of the observer hiding inside
the channel.

![The capacity band at the 2¹⁰ baseline: capacity in bits against the rate, zero at indifference and at foam, one peak at the reliable edge near 0.25 with 9.05 bits, and 0.5 and 0.75 inside the band at 88.5 and 80.9 per cent of the peak. Measured: the sweep run on 16 September 2026.](/sx/assets/foundations/the-channel-held/channel-band.svg)

![The observation-window relativity: the peak rate against the budget from 256 to 2048 laps on logarithmic axes, sliding as 1.0, 0.5, 0.25, 0.125 along the line of peak rate × budget = 2⁸. Measured: the budget sweep run on 16 September 2026.](/sx/assets/foundations/the-channel-held/channel-sliding-peak.svg)

## The peak at full precision

The operator asked how precise the reported peak was. The reported value of 0.99556 at a
budget of 2⁸ is the ninth point of a 33-point logarithmic grid, quantised at about 26
percent: a sample, not a constant. The exact fields at that budget are a being of 137,
the 2⁸ window reading the origin, a fold of 4.2322, and a capacity of 7.101724 bits, which
is the log of 137 within 0.05 percent. The true settled edge, found by bisection, is a
rate of 0.9772624, and it is a cliff between screens: the witness horizon jumps from 238
straight past the window, because the ladder of screens is quantised in rate and the
band's edge is a step, not a smooth crossing. A numerology check was run and killed. A
digit-based comparison with the speed of light, 0.99792458 over 0.75 against four thirds
at 0.21 percent, and a difference reading near a third, fails on three grounds: the speed
of light in metres per second is exact by definition since 1983 and its digits encode the
historical metre; the difference was anchored on the grid sample and moves from 0.335 to
0.353 against the true edge; and 0.21 percent sits inside the record's kill band. Any
such connection must arrive as a dimensionless, out-of-sample ratio, never by digit
alignment. The model's own unit-free speed constants stand as recorded: a quarter as the
cost of existence, four thirds as the impedance, and the fills of 237 over 256 at the
origin.

![The cliff between screens at budget 2⁸: the witness horizon holding at 238 up to the true settled edge at rate 0.9772624 and then jumping past the 256-lap window, with the grid sample 0.99556 beside it. A schematic of the record's bisection.](/sx/assets/foundations/the-channel-held/channel-cliff.svg)

## The co-constructive channel

The second phase treats each instrument as a Quad Slip and reads its y-projection flip
signal over the loops, then asks whether several instruments can be orthogonal and what
their joint capacity is. Orthogonality is achievable, and it is a dynamic property, not a
spatial one. A congruent baseline gives redundant channels, a correlation of plus one; a
mirror turned through 180 degrees gives anti-correlated channels, minus one; both carry a
single channel. A large detune of the time rate decorrelates the instruments to a
correlation near zero, orthogonal, while placement alone does not decorrelate them at
all. Orthogonality is therefore, in the operator's phrase confirmed by the measurement,
time saved within space. Three decorrelated instruments reach 2.83 effective channels of
a possible three, an orthogonality of 0.86. The joint capacity is proportional to the
effective independent channels, the participation ratio of the correlation matrix, one
when redundant and N when orthogonal. The sweep shows the transition: a detune of 0.002
gives a correlation of 0.98 and one effective channel, and a detune of 0.13 or more gives
a correlation near zero and two. Maximum surprise pulls the system to decorrelate; the
attractor is the drive toward orthogonality.

![Orthogonality is dynamic: the correlation of two instruments against their detune, 0.98 and redundant at the smallest detunes, falling to about zero — orthogonal, two channels — as the time-rate difference grows. Measured: the orthogonality sweep run on 16 September 2026.](/sx/assets/foundations/the-channel-held/channel-orthogonality.svg)

## The epistemic grid

The grid reads the co-construction as a computational medium. It divides a run into K
ticks, or moments, and at each tick reads the correlation matrix of the N instruments over
that tick's window, giving the effective independent channels moment by moment. Its
premise was grounded first: a single instrument's stream is a clock and not free memory,
its flip signal periodic with an autocorrelation at lag four of about 0.92 over a
2048-lap run, a clean period-four oscillator. So capacity does not live in one clock's
alternation, which is predictable and is the time axis; it lives in the orthogonal
relationships between clocks, the space or data axis. And whether a stream is a clock or
a message is observer-relative: an observer at lag one sees about 0.65 bits, a message,
and an observer at lag two or four sees a predictable clock and no bits. That is what
makes the grid epistemic. A fine scan of the detune found the pairwise relationship
alphabet effectively ternary, 24 of 30 sampled time rates landing on plus one, zero or
minus one, redundant, orthogonal or mirror, with intermediates only in a thin sliver.

Capacity is a band in the resolution K, rising to an interior peak and decaying:

| K | 2 | 3 | 4 | 6 | 8 | 10 | 12 | 16 | 20 |
|---|---|---|---|---|---|---|---|---|---|
| capacity, bits | 8.8 | 12.8 | 17.6 | 21.6 | 22.5 | 25.2 | 15.3 | 8.6 | 5.9 |

The optimal tick window of about 17 to 22 samples is about four to five and a half clock
beats wide, right at the Quad Slip's five-beat. It is the observation-window relativity
one level up: too coarse and there are few moments; too fine and the window is too short
to settle a relationship, so an apparent orthogonality is small-sample noise, the grid's
foam, discounted by a measured split-half reliability. At the optimal resolution the
lattice closes: every moment is addressable except the origin, the tick before
differentiation, which is blind. A congruent trio is blind everywhere with a capacity of
zero, since redundant clocks carry no grid; over-resolving past the peak manufactures
blind moments, one, then six, then thirteen. A blind moment is a locked tick where two
clocks phase-align, a cell mid-flip. One prediction stays held open and is not confirmed:
the tennis-racket flip of the intermediate-axis theorem would give a periodic blind
cadence at the five-beat, and the grid shows only the origin's blind moment and the foam
past the peak. Spin stays a held prediction, not a result.

![The epistemic grid at K = 8: three clocks across eight moments of 22 samples, the effective independent channels per moment, every moment open except the origin, which is blind — a locked tick where two clocks phase-align. Measured: the grid read at 2048 laps on 16 September 2026.](/sx/assets/foundations/the-channel-held/channel-grid.svg)

![Capacity as a band in the resolution K: rising from 8.8 bits at K = 2 to 25.2 at K = 10 and decaying to 5.9 at K = 20, the optimal window about four to five and a half clock beats. Drawn from the record's table.](/sx/assets/foundations/the-channel-held/channel-k-band.svg)

## Register or lock

Does the count of open slots lock to a fixed number of epochs, or grow with the run? The
grid was measured over run lengths at their peak resolution:

| laps | signal length | peak K | moment width | open slots |
|---|---|---|---|---|
| 512 | 47 | 2 | 23 | 2 |
| 1024 | 90 | 3 | 30 | 3 |
| 2048 | 179 | 10 | 17 | 9 |
| 4096 | 358 | 16 | 22 | 15 |
| 8192 | 712 | 32 | 22 | 31 |

The lock is refuted: the log of the open slots against the log of the laps has a slope of about 1.0,
so the slots scale linearly with the run, and the earlier reading of nine as an origin
plus eight was specific to one run length, not a fundamental cap. The intensive invariant
is the moment width, which stays at about 17 to 22 catches, about five clock beats,
across every run length, the two shortest rows reading coarse for want of signal. The
grid is a register that lengthens with observation time and is built from fixed-width
moments: more time gives more moments of the same size. The being's epochs grow only as
the log of the laps, so grid moments are not epochs, a distinction and not a coincidence.

![Register, not lock: the open slots against the laps on logarithmic axes with a slope of about 1.0, extensive, beside the moment width at the peak holding at about 17 to 22 catches, about five beats, across every run length. Drawn from the record's table.](/sx/assets/foundations/the-channel-held/channel-slot-scaling.svg)

## Confinement

The confinement surface reads capacity over space, N mutually independent channels chosen
greedily to be orthogonal, against time, the moments K, at the peak resolution for each N:

| N | channels achieved | peak K | window | window × K | capacity | K over N |
|---|---|---|---|---|---|---|
| 2 | 2.00 | 12 | 14 | 168 | 33.5 | 6.0 |
| 3 | 3.00 | 8 | 22 | 176 | 17.4 | 2.67 |
| 4 | 4.00 | 8 | 22 | 176 | 24.0 | 2.0 |
| 5 | 4.99 | 3 | 59 | 177 | 20.1 | 0.6 |
| 6 | 5.98 | 2 | 89 | 178 | 17.5 | 0.33 |
| 7 | 6.97 | 2 | 89 | 178 | 20.1 | 0.29 |
| 8 | 7.94 | 2 | 89 | 178 | 22.9 | 0.25 |

The confinement is an uncertainty relation. Resolving N independent channels demands a
coherence window that grows faster than N, from 14 to 89 as N goes from two to six, so at
a fixed signal budget the moments must shrink, and the window times the moments is
conserved at about 179 across the whole sweep. Space and time cannot both be bought; they
trade within the budget. The substrate cleanly supports about seven or eight independent
channels, 7.94 of 8 achieved with no hard ceiling yet found. Capacity is greatest in the
time-rich corner, two channels at a ratio of six, time saved within space, because
capacity counts channel-moments and moments are the cheaper axis; the mid-range
capacities are soft, and the robust results are the conserved product and the growing
window.

![The confinement as an uncertainty relation: the window growing faster than N while the moments K shrink, their product holding near 179, and the capacity greatest in the time-rich corner at N = 2. Drawn from the record's table.](/sx/assets/foundations/the-channel-held/channel-confinement.svg)

## What is held open, and what follows

Three things are stated as not built. The x, y, z spatial emergence: three mutually
orthogonal channels at maximum capacity as the handle onto dimensionality, the doubling
and tripling of the 0.25 band, with the break to orthogonality as the self-organising
step. The tennis-racket flip, spin, as the rotational overlay on that frame, kept an
overlay until an out-of-sample handle exists. And the follow-on the operator named
co-constructive optimism: a thermodynamic process gated by the confinement ratio of time
to space, resolving toward maximum capacity, with the two-terminal diode read as a
structural confinement and the fixed moment-cell of about five beats as its unit.

## The register, and what is not claimed

A readout over the frozen geometry: the cryptographic core and the figure-2 coincidence
untouched, the baseline pinned, the double hinge's fold and the Quad Slip reused. Every
capacity term is a measured quantity, the being, the horizon and the fold, and nothing is
tuned to fit; the peak's location is an output, never assumed. Model-internal, and no
physics is claimed. The band, the sliding peak, the exact capacity at 2⁸, the dynamic
orthogonality, the clock premise, the K band, the extensive register with its intensive
moment, and the conserved product are the measured structure. The cooling picture the
band confirms is the Coco's, and the reading names it as the record does; the spin and
the spatial emergence are held, not claimed.

## Where it sits

The ninth instrument of the series of ten, after the XOR Mirror and before the Coco,
whose cooling picture its band confirms in capacity terms; it reads the Double Hinge's
fold and the Quad Slip's instruments. On the observatory node the door is
[opendata.ai/channel](https://opendata.ai/channel), state *live*: the instrument runs on
request, as the platform singleton, its first sweep building a cached grid over about a
minute, and nothing a reader does there enters the record. No external anchor is recorded
for the Channel, and this reading says so rather than borrowing one.

## Standing

Held on this node as fact until someone stands up in the record to refute it; a
refutation enters as a statement of intent at [the matters of note](/sx/matters) and
goes to the record's own path. What has since been built on the Channel, and the walk a
reader may take, are the operator's hand and land in a further draft.