Lunar surface engineering, decided exactly — the routing, the whole loaded field, heat that leaves and will not return, and the ground beneath the site.
lunar surface missionsexact, not sampledyour data stays yours
A lunar surface mission has to build on ground nobody can walk out to and inspect.
You cannot ship the supercomputer. You cannot send the technician. The surface swings
roughly 290 K between a fourteen-day day and a fourteen-day night, and the regolith
beneath it is one of the best insulators ever measured. The four questions that decide the
build — does the route exist, what does the whole field do, does the heat
leave, what is under the site — are precisely the ones that do not survive
being sampled, estimated, or run to a residual that merely looks flat.
Selenoform uses the Helix Processor to answer them as exact objects. The route census is a truth test, not a search. The
field query returns the complete path-dependent field, not a scalar probe. The thermal
schedule is an integer channel count, not a tuned parameter. The subsurface reading comes
back with its own integrity band and refuses outside it. One published operator, compiled
once, evaluated where re-running the conventional method is physically impossible.
Designed for the Moon, validated on Earth — where you can still go and dig, and where a reader can check the record without taking our word for it. Measured against the fields' own references — a full-field independent calculation, banded and
sparse and dense solvers in agreement, public SIT4ME mine-noise records, and exact quorum
certificates — and the failures are printed beside the wins. The mathematics is published
in full, today, as a single 44-page paper with a DOI. The compiler that runs it is not.
The processor beneath all four instruments
The Helix Processor does not merely run the same calculation faster. Its exact closed form changes
how the work grows with the lattice floor. The benchmark below asks both paths for the same exact integer
at seven increasing sizes: the direct referee streams the whole floor, while the receipt-bearing
Helix path reads the answer from the floor's atoms.
Same target, same exact integer, measured end to end at every point. At
m = 881,188,485, direct exact streaming took 3,482.09 s
(58.03 min); the receipt-bearing Helix Processor took 457.34 µs —
7,613,805× faster on the measured endpoint. Across the seven completed runs, the
measured wall-time exponents are 1.016 for direct streaming and 0.036 for the sealed path. No
projected point appears. The sealed time includes receipt construction; the
public evidence receipt carries every value and
the pinned audit.
Four data sheets — engagements are direct
Sw
Switchyard
ROUTE · the surface's switch book
Which sources, ports, and clocks can actually be wired together on a built
surface — decided by an exact membership test against your policy, over a command room too
large to enumerate.
A command room of 12,432,960 words answered without materializing
them. Restricting the exhaustive divisor census to the support policy cuts the probe count
from 3,108,240 pairs to 12,015 — an exact 258.7× reduction, not a
heuristic prune. All 1,036,080 pairs round-trip and all 324 glue/clock
representatives agree with an independent path calculation.
The complete path-dependent field of a distributed, switched surface — every
trajectory, not a sampled few — compiled once and then evaluated at horizons where
streaming the answer is no longer possible.
A raw history room of 1618 words collapses to ten congruence
skeletons whose exact multiplicities sum to the full count. The compiled evaluator is
logarithmic where streaming is linear: the measured operation ratio climbs
52.2× → 11,842× from L = 600 to 600,000. On one identical
target the generic frontier took 70.87 s; the ten-skeleton evaluator took a median
23.1 µs.
A scheduled, passive thermal valve for a surface that swings hundreds of degrees
between day and night — sized by an integer channel count that jitter cannot shift, not by a
parameter you tune on site.
Over one lunation on a 390 K day / 100 K night surface, the
scheduled valve turns a net absorber into a 13× net radiator —
Qout/Qin = 13.0 against 0.87 for the same surface bare. The
ring solve is O(N) in time and storage, run to one million sites in
4.8 s, where the dense complex Green matrix alone would be 14.9 TB before
factorization.
RESOLVE · resource contacts below the channel limit
A whole-assembly seismic instrument for thin buried structures: pull two acoustic
contacts out of one conventional peak, qualify the result on held-out field data, and refuse
beyond the measured frontier.
In three public SIT4ME mine-noise looks, a registered pair 19.23 m
apart — 0.60× the common channel limit — folds into one matched-field peak
but returns as two held-out contacts in 3/3 fixed-address reads. All 16/16
one-class deletion stresses pass at the measured 9 dB operating point; noise alone is refused
3/3. The product behind the fixture is not 48 independent sensors: it is a separately
audited 1,024-address quorum volume with 40,338,372 whole-assembly operating modes.
Four instruments, one substrate — and one thing no part can compute
Switchyard, Superhighway, Valve, and Ground are four questions put to
one published operator: route the surface, carry the field, keep the heat, listen below.
Underneath them sits a conservation law with an unusual property — the complete assembly
crystallizes a mode that no coalition below the full assembly can compute. Every part's current
is nonzero; they sum to zero. It is conserved by the whole and by no part, and it has no closed
form, because the object is not the kind of thing a formula can name. Where a delivered capability
rides that mode, it is structurally absent — not hidden, absent — from any
incomplete install.
The same substrate carries further confirmed results on this surface, published in the same paper
and available on engagement rather than as data sheets:
Sparse sensing. Four scatterers, two of them sub-Rayleigh pairs, recovered with no
ghosts by testing source irreducibility — a 105× source/sink separation, and all
20 tested looks reject the coherent false positive.
Survey and lifecycle. Array aliasing is a divisor event, so a design's failure modes are
arithmetic, not statistical; deletion and mistuning occupy disjoint channels, and 24
injected faults classified with zero cross-talk.
Volume quorum. On a three-axis frame, an equal-share quorum exists exactly when the
sector count is divisible by the characteristic — the first democratic quorum is
three-way.
Selenoform is offered to vetted partners by direct engagement — an instrument pointed at
your mission, or the operator pointed at a question of your own. Either way the rule is the same as
the rest of Lattice OS: your problem in, the answer out, no copy of your data kept.
