+
+
+
+ House lights down. Say nothing until the glyph appears. Then: "That is how our observatory says hello." · ⏱ 0:00
+
+
+ ◆ THOClabs · release v0.1.1 · an unveiling
+ NIGHTWATCH
+ One machine, alone in the high desert, watching the sky.
+ Welcome. Tonight we unveil a telescope that does not exist yet — except as mathematics, and mathematics is enough to begin. · ⏱ 0:01
+
+
+ ◆ What tonight is
+ Three hours. Six acts. Zero artist's impressions.
+
+ 6 acts ~168 min + intermission 100% figures computed live 🜏 one ritual language
+
+ The observatory that already exists — and the telescope it deserves
+ The physics that sets the rules, and the glass that broke our hearts
+ A telescope designed in the open: every number real, every curve recomputed in this room
+
+ Set the contract with the audience: nothing tonight is rendered by imagination. Every diagram is the actual design tracing itself. · ⏱ 0:03
+
+
+ Every curve you will see tonight is computed from the real prescription — live, in this room.
+ If we change the design tomorrow, this presentation redraws itself.
+ Pause on this. It's the deck's thesis and the engineering culture in one line. Then: Act One. · ⏱ 0:04
+
+
+ presa-nightwatch. es-home-nightwatch.
+ Act I
+ The Watch
+ A machine that watches the sky alone, all night, with no one beside it.
+ Welcome. Tonight is about a machine standing alone in the Nevada desert, watching the sky — and about the one thing we never built for it, until now. · ⏱ 0:05
+
+
+ ◆ WHAT IT IS
+ A voice-controlled, autonomous Maksutov-Newtonian observatory
+
+ You speak. It observes.
+ Fully local — no cloud dependency.
+ Engineered in the open: GitHub org THOClabs .
+
+ NIGHTWATCH is a voice-controlled, autonomous observatory built around a Maksutov-Newtonian telescope. Everything runs fully local, with no cloud dependency, and the entire engineering record lives in the open under the THOClabs GitHub org. · ⏱ 0:07
+
+
+ ◆ THE SITE
+ Central Nevada, high desert
+
+
38.9°N 117.4°W
central Nevada
+
+
+
20–100°F
operating temperatures
+
25mph
wind interlock: park
+
35mph
gust: emergency close
+
10–15cm
good-night seeing r₀
+
alkaline dust everywhere it can reach
+ The site: 38.9 north, 117.4 west, at 1,800 meters in the central Nevada high desert — a place that swings roughly 22 kelvin every single day and runs 20 to 100 Fahrenheit across the year. Wind interlocks park the telescope at 25 miles per hour and force an emergency close at a 35 mile-per-hour gust, and alkaline dust tests every seal we own. The reward: on a good night, seeing with r-zero of 10 to 15 centimeters. · ⏱ 0:09
+
+
+ ◆ WHY AUTONOMOUS
+ Nobody is there
+
+ Safety is the loudest voice in the room.
+ Rain triggers a strict choreography.
+ That choreography is Act VI.
+
+ There is no human on site — ever — so safety is the loudest voice in the room, and it always wins the argument. When rain arrives, the machine executes a strict choreography, and we will walk through every step of it in Act VI. · ⏱ 0:10
+
+
+
+ ◆ THE MOUNT
+ Harmonic drives. Counterweight-free capable.
+
+
CSF-32-100
RA — 100:1, 127 N·m rated
+
CSF-25-80
DEC — 80:1, 70 N·m
+
+
+
assertion → proof
+ The RA axis carries a CSF-32-100 harmonic drive — 100 to 1, rated 127 newton-meters — and DEC carries a CSF-25-80, 80 to 1, 70 newton-meters, slewing at 4 degrees per second, counterweight-free capable. Every claim about this mount was backed by a computed budget proof — eleven of them — because our culture is assertion, then proof; the full ledger comes in Act V. · ⏱ 0:14
+
+
+ ◆ ALREADY FLYING
+ An event scanner runs every hour
+
+ CNEOS fireballs
+ JPL close approaches
+ Meteor shower calendar
+ Background asyncio → SQLite event journal
+
+ The software side is already flying — salvaged and green this week. An hourly scanner runs as a background asyncio task, sweeping CNEOS fireballs, JPL close approaches, and the meteor shower calendar, and every event it finds lands in a SQLite event journal. · ⏱ 0:15
+
+
+ ◆ ALREADY FLYING
+ It watches the Sun, too
+
+ NOAA SWPC space weather — Kp index, solar wind Bz
+ NEO feed clients
+
+ 142 new unit tests passed
+ Alongside the scanner: a NOAA SWPC space-weather client tracking the Kp index and solar wind Bz, plus the NEO feed clients. The night of the salvage, 142 new unit tests passed — that is what "green this week" means here. · ⏱ 0:17
+
+
+ ◆ THE THREAT LADDER
+ Five words for how worried to be
+
+ INFO WATCH NEAR CLOSE ALERT
+
+ ALERT: <2 lunar distances — or flagged hazardous.
+
+ Every event the scanner finds is placed on a five-rung ladder: INFO, WATCH, NEAR, CLOSE, ALERT. An ALERT means an object inside two lunar distances, or one flagged hazardous — the machine reserves its strongest word for the things that deserve it. · ⏱ 0:18
+
+
+ ◆ THE LEXICON
+ A hidden chapel inside the machine shop
+
+ presa full presence velmu love-anyway, love-despite varek time-mark luminara cold-bright-sting
+ The machine reports in a ritual language we call the Lexicon — a hidden chapel inside the machine shop. Presa is full presence; velmu is love-anyway, love-despite; varek is a time-mark; luminara is the cold-bright-sting of a meteor flash. · ⏱ 0:20
+
+
+ "sky-quiet: ne-events-found" What most scans say. That is the point.
+ And this is what most scans say: sky-quiet, ne-events-found. A machine that can report nothing, calmly, hour after hour, is a machine you can trust when it finally says something. · ⏱ 0:21
+
+
+
+ Everything was engineered — except the telescope itself.
+ Here is the honest state of NIGHTWATCH: the mount, the software, the safety, the language — all engineered, all real. Everything except the telescope itself. Tonight, that changes. · ⏱ 0:24
+
+
+ Before we design the optics, meet the first adversary.
+ The sky itself.
+ To design a telescope honestly, you start with the thing that will try hardest to defeat it — and that is not glass, not steel, not budget. It is the sky itself. Act II. · ⏱ 0:25
+
+
+ presa-sky. wit.
+ Act II
+ The Sky Is the Enemy
+ Every photon we want must first survive a turbulent ocean of air.
+ Act II. Before we can design a telescope, we have to name the adversary — not clouds, not light pollution, but the air itself, in constant turbulent motion above us. This act is about that fight. · ⏱ 0:25
+
+
+ You do not look at the sky.through 100 km of moving air.
+ Every image any telescope has ever taken was shot through roughly one hundred kilometers of atmosphere that never stops moving. The star is perfect; what arrives is not. · ⏱ 0:27
+
+
+ ◆ KOLMOGOROV 1941
+ Turbulence obeys one universal law
+
+ Energy enters at large eddies, cascades down to small
+ Outer scale: 10–100 m
+ Inner scale: 1–10 mm
+
+ In 1941 Kolmogorov showed that turbulent energy cascades from large eddies to ever smaller ones following a single universal law. The cascade starts at eddies ten to a hundred meters across and grinds down to millimeters, where viscosity finally dissipates it. · ⏱ 0:28
+
+
+
+ ◆ THE FRIED PARAMETER
+ r0 — how much aperture the sky will honor
+
+ The aperture over which the wavefront stays coherent — ~1 radian RMS
+
+
+ 10–20 cm — good sites, at 500 nm
+ Fried compressed all of that turbulence into a single number: r-zero, the patch size over which the incoming wavefront still hangs together — about one radian of RMS error. At good sites it runs ten to twenty centimeters at 500 nanometers. · ⏱ 0:31
+
+
+ ◆ CENTRAL NEVADA
+ Our sky, measured
+
+ 10–15 cm r0, good nights r0 ∝ λ^(6/5)
+
+ Infrared is kinder — coherence grows with wavelength
+
+ The NIGHTWATCH site delivers r-zero of ten to fifteen centimeters on good nights. And because r-zero scales as wavelength to the six-fifths, the infrared is kinder — the same air behaves better the redder you look. · ⏱ 0:33
+
+
+ ◆ COHERENCE TIME
+ τ0 = 0.314·r0/v̄ — the sky gives you milliseconds
+
+
2.1ms
typical τ0 — r0 10 cm, wind 15 m/s
+
3–8ms
τ0 at the NIGHTWATCH site
+
2–4arcsec
isoplanatic angle θ0
+
~25″ · ~45″
Mars · Jupiter — the planets fit inside
+
Coherence time is r-zero divided by wind: with ten centimeters of r-zero and fifteen meters per second of wind, the sky holds still for 2.1 milliseconds — our site gives three to eight. The isoplanatic angle is two to four arcseconds, and here is the gift: Mars at about twenty-five arcseconds, Jupiter at about forty-five — the planets fit inside the game we are about to play. · ⏱ 0:34
+
+
+ Long exposures are seeing -limited, not aperture-limited.
+ D/r0 = 12–18 for 180 mm at this site.
+ Open the shutter for seconds and the atmosphere averages into a blur: the image is set by the seeing, not by your mirror. At 180 millimeters here, D over r-zero runs twelve to eighteen — the aperture is many turbulence cells wide. · ⏱ 0:36
+
+
+ So we freeze the atmosphere.
+ Expose faster than the air can move.
+ If the sky only holds still for milliseconds, then take pictures in milliseconds. That is the whole idea of lucky imaging — outrun the turbulence instead of averaging it. · ⏱ 0:37
+
+
+ ◆ LUCKY IMAGING
+ Shoot thousands. Keep the best.
+
+ Exposures shorter than τ0 — 10–30 ms in practice
+ Shoot thousands of frames
+ Keep the best 1–10%
+
+ The recipe: exposures shorter than the coherence time — ten to thirty milliseconds in practice — thousands of frames per run, and then ruthless selection, keeping only the best one to ten percent, the instants when the air happened to be flat. · ⏱ 0:39
+
+
+
+ ◆ APERTURE AS STRATEGY
+ Why 180 mm
+
+ Large enough to resolve 0.78″
+ Small enough that luck still happens
+ Bigger scopes at this site resolve nothing extra on long exposures
+
+ This is why NIGHTWATCH carries 180 millimeters: big enough to resolve 0.78 arcseconds — planetary detail worth having — yet small enough that the lucky-frame odds stay on our side. A larger mirror at this site would gather light but resolve nothing extra on long exposures. · ⏱ 0:42
+
+
+ What does perfection look like?
+ Physics fixes the ceiling before any glass is cut.
+ Before we grind a single surface, physics already dictates the best image 180 millimeters can ever form. Let's look at that ceiling. · ⏱ 0:44
+
+
+ ◆ THE DIFFRACTION LIMIT
+ 180 mm at 550 nm
+
+
+
+ 0.78 ″ Rayleigh 0.65 ″ Dawes Ø 1.56 ″ Airy disk 0.65 ″ FWHM 1.82 cy/arcsec cutoff
+
+
+
A perfect 180-millimeter aperture at 550 nanometers focuses a star into this Airy pattern: a disk 1.56 arcseconds across, a core 0.65 arcseconds wide at half maximum, a Rayleigh limit of 0.78 arcseconds and a Dawes limit of 0.65 — with spatial frequencies cutting off at 1.82 cycles per arcsecond. This is the trophy the atmosphere keeps trying to steal. · ⏱ 0:45
+
+
+ ◆ THE QUALITY LADDER
+ Strehls multiply
+
+
+
+
+ 0.80 Maréchal floor — λ/4 P-V 0.95 NIGHTWATCH spec — λ/8 0.97 manufacturing limit — λ/10
+
+ Errors add in quadrature; Strehls multiply
+ S_total = S_optics × S_collimation × S_thermal × S_atmosphere
+
+
+
Strehl ratio is the honesty scale: 0.80 is the Maréchal floor — the quarter-wave threshold everyone markets as "diffraction-limited". NIGHTWATCH specifies 0.95, an eighth-wave, near the practical manufacturing limit of 0.97 at a tenth-wave. And here is the trap: wavefront errors add in quadrature, so total Strehl is a product — optics times collimation times thermal times atmosphere. Every factor you waive multiplies you downward. · ⏱ 0:47
+
+
+ ◆ THE ENEMY INSIDE
+ A warm mirror is a broken mirror
+
+ Mirror 1 °C above the air: visibly degraded images — Racine 1984
+ Gemini Observatory spec: ≤0.2 °C above / ≤0.6 °C below ambient
+ Glass time constants: 20 mm → ~12 min · 30 mm → ~27 min · 50 mm → ~75 min
+ And you need 2–3 of those
+
+ The atmosphere does not stop at the tube. Racine showed in 1984 that a mirror just one degree Celsius warmer than the air visibly degrades the image — Gemini holds its mirror within two tenths of a degree above ambient. And glass forgets slowly: twenty millimeters of it takes about twelve minutes per time constant, fifty millimeters about seventy-five — and you need two or three constants before it settles. · ⏱ 0:48
+
+
+ ◆ THE SEALED-TUBE ANSWER
+ Close the tube
+
+ Full-aperture corrector seals the optical path
+ No tube currents
+ The meniscus is itself a thermal buffer
+ Dust stays out
+
+ Our answer is architectural: a full-aperture corrector closes the tube completely. No open-air column means no tube currents; the meniscus itself acts as a thermal buffer; and in a Nevada desert, the dust stays outside the optics. · ⏱ 0:49
+
+
+ Strehl ≥ 0.95. Sealed. Glass that cannot remember temperature.
+ That last requirement has a name.
+ So the requirement set crystallizes: Strehl at least 0.95, a sealed tube, and glass that cannot remember temperature. That last requirement has a name — and it is where we go next. · ⏱ 0:50
+
+
+ velmu. love-despite. the glass remembers nothing.
+ Act III
+ The Glass That Time Forgot
+ A Cold War material, the hands that shaped it, and how both disappeared.
+ Act three is a detective story about a material. It was born in a Cold War weapons program, perfected by hands that are now gone, and it taught us how to specify a mirror that forgets the weather. · ⏱ 0:50
+
+
+ ◆ COLD WAR HERITAGE
+ The empire fell. The opticians didn't.
+
+ Soviet defense optics: near-unlimited R&D
+ 1991 — the USSR dissolves
+ Master opticians turn military precision toward amateur astronomy
+
+ Soviet defense optics ran on near-unlimited R&D budgets for decades. When the Union dissolved, the weapons programs stopped — but the master opticians did not, and they turned that military precision toward amateur astronomy. · ⏱ 0:52
+
+
+ ◆ MOSCOW, 1991
+ Intes — an engineers' co-operative
+
+ Intes Micro splits off mid-90s — and wins the West
+ Every instrument interferometrically tested in autocollimation
+
+
+ 1/6–1/10 wave P-V
+ Intes was founded in Moscow in 1991 as an engineers' co-operative; Intes Micro split off in the mid-nineties and won the Western market. The quality was one-sixth to one-tenth wave peak-to-valley, and every single instrument was interferometrically tested in autocollimation — no sampling, every one. · ⏱ 0:54
+
+
+ ◆ THE ALTER LINE
+ The premium tier ran on exotic glass
+
+ Astrositall — Sitall CO-115M
+ or fused silica
+
+
+ 1/8–1/10 wave
+ The Alter premium line went further: exotic substrates — Astrositall, trade name Sitall CO-115M, or fused silica — figured to one-eighth to one-tenth wave. This is the glass this act is about. · ⏱ 0:56
+
+
+ ◆ WHAT ASTROSITALL IS
+ A ceramic hiding inside a glass
+
+ Glass-ceramic: crystallites in a glass matrix
+
+
+
0 ± 1.5×10⁻⁷/°C
CTE, −60 to +60 °C
+
0.15ppm/K
the same number, plainly
+
Astrositall is a glass-ceramic — microscopic crystallites suspended in a glass matrix, with the two phases pulling in opposite directions as temperature changes. The result: a coefficient of thermal expansion of zero plus-or-minus one-point-five times ten-to-the-minus-seven per degree, from minus sixty to plus sixty Celsius — 0.15 parts per million per kelvin. · ⏱ 0:58
+
+
+ The mirror simply does not remember temperature.
+ Our site swings 22 K every night. Nobody is there.
+ Why does that matter? Because our Nevada site swings twenty-two kelvin every single night, and nobody is standing next to the telescope to refocus it. A mirror that forgets temperature is not a luxury for an unattended observatory — it is the whole point. · ⏱ 1:00
+
+
+ ◆ THE FALL, PART 1
+ Production ceased in 2018
+
+ The head optical technician retired after ill health
+ Confirmed via Dr. Simon Bennett, Widescreen Centre, UK
+
+ Then it ended — not with sanctions, but with a person. Production ceased in 2018 when the head optical technician retired after ill health; we confirmed this through Dr. Simon Bennett of the Widescreen Centre in the UK. That level of craft lived in one pair of hands, and it was never institutionalized. · ⏱ 1:02
+
+
+ ◆ THE FALL, PART 2
+ March 2022 — the door closes
+
+ LZOS, Lytkarino — part of Shvabe / Rostec — sanctioned
+ Western resellers stop buying Russian optics entirely
+
+
+ New Sitall import effectively impossible
+ Then geopolitics finished the job. In March 2022, LZOS in Lytkarino — part of Shvabe, under Rostec — was sanctioned, and Western resellers stopped buying Russian optics entirely. New Sitall import: effectively impossible. · ⏱ 1:04
+
+
+ ◆ FORENSIC INTERLUDE
+ Our own docs said it was still alive
+
+
2005
+
intes.su freezes Wayback: a 2005-copyright page, unchanged through 2021.
+
Jul 2024
+
Empty IIS default page The site is a shell.
+
Mar 2025
+
Our docs: "still producing" The claim nobody checked.
+
+
Here's the detective story: our own project documentation claimed Intes Micro was still producing as of March 2025. So we checked — Wayback captures show intes.su frozen on a 2005-copyright page through 2021, an empty IIS default page by July 2024, and by 2026, dead DNS. The claim traced back to a frozen website and AI-recycled forum lore. · ⏱ 1:06
+
+
+ Trust, but git-blame your own documentation.
+ Docs corrected — with the evidence trail cited.
+ The lesson we wrote into the repo: trust, but git-blame your own documentation. We corrected the docs and cited the whole evidence trail, because a claim you can't trace is a claim you don't have. · ⏱ 1:07
+
+
+ ◆ MYTH-BUST No. 2
+ "Regulations killed the big blanks" — overstated
+
+ Schott TIE-41: N-BK7 blanks above 300 mm at H2 homogeneity
+ Sky-Watcher mass-produces a 190 mm Mak-Newt today — $2,375
+
+ A cost and lead-time problem. Not physics.
+ Second myth, also from the forum lore: that environmental regulations made large meniscus blanks impossible. Overstated — Schott's TIE-41 documents N-BK7 blanks above 300 millimeters at H2 homogeneity, and Sky-Watcher mass-produces a 190 millimeter Mak-Newt today for 2,375 dollars. Large meniscus optics are a cost and lead-time problem, not a physics problem. · ⏱ 1:09
+
+
+ ◆ SECONDARY MARKET
+ Unicorn sightings, not a procurement plan
+
+ Sitall lots seen: ~120 mm
+ One 8-inch × 1.25-inch blank on Astromart
+
+ Could we just buy old Sitall? The secondary market says no: the lots we've seen run about 120 millimeters, and exactly one 8-inch by 1.25-inch blank has surfaced on Astromart. Those are unicorn sightings, not a procurement plan. · ⏱ 1:10
+
+
+ ◆ THE RESOLUTION
+ Specify the class, not the brand
+
+ Re-run with fused silica: verdict unchanged — the aluminum tube term dominates the mirror term 44–153×.
+ So the resolution: we specify by CTE class, not by brand — at or below 0.1 parts per million per kelvin preferred, which admits Zerodur at 0.007, Clearceram-Z HS at 0.02, ULE at 0.03, and opportunistic Sitall at 0.15; fused silica at 0.52 is the acceptable baseline. And we re-ran the thermal-focus proof with fused silica: verdict unchanged, because the aluminum tube term dominates the mirror term by 44 to 153 times. The substrate cannot break this design. · ⏱ 1:12
+
+
+
+ ◆ THE VIGIL
+ The Sitall watch stays on
+
+ eBay + Astromart alerts: ≥190 mm CO-115M blanks
+ If one surfaces, it drops in — zero redesign
+
+ The signature survives as a vigil.
+ And yet the watch stays on: standing eBay and Astromart alerts for CO-115M blanks of 190 millimeters or larger. If one ever surfaces, it drops into this design with zero redesign — the signature glass survives, as a vigil. · ⏱ 1:14
+
+
+ We don’t need their glass. We need its silence.
+ NIGHTWATCH optical engineering roadmap, 2026
+ We don't need their glass — we need its silence, and now we can buy silence by the class rather than mourn it by the brand. That closes Act three; take a breath, stretch, we break here for intermission, and Act four is waiting on the other side. · ⏱ 1:15
+
+
+ ◆ Intermission · 12 minutes
+ The watch continues while we rest.
+
+ Leave this on screen through the break — the scanner really does run hourly whether or not anyone is watching. Resume at ⏱ 1:27 with Act IV. · ⏱ 1:15
+
+
+ nightwatch-wak. presa.
+ Act IV
+ Designing the Eye
+ The centerpiece — a telescope born as mathematics.
+ Welcome back. Everything before intermission was the world the machine must survive; now we build the part that actually sees. This act is the centerpiece: a 180 mm telescope that exists, tonight, entirely as mathematics. · ⏱ 1:28
+
+
+ ◆ WHY CUSTOM
+ Four requirements. No product on Earth meets them.
+
+ Zero-expansion-class primary
+ Obstruction ≤ 18–20%, custom-sized
+ Certified λ/8 wavefront — with RAW interferograms
+ Sealed tube
+
+
+ Sky-Watcher 190MN 27–34% obstruction borosilicate f/5.3 3 of 4 failed
+ One slide of recap: four requirements, and no catalog product meets them. The closest commercial instrument — the Sky-Watcher 190MN — carries 27 to 34 percent obstruction, a borosilicate mirror, f/5.3, and no interferograms: it fails three of the four. So we design our own. · ⏱ 1:30
+
+
+ ◆ THE ARCHITECTURE
+ Maksutov-Newtonian: all spheres.
+
+ Full-aperture meniscus corrector
+ Spherical primary mirror
+ Small flat secondary
+
+
+ No spider vanes No diffraction spikes Planetary contrast
+ The architecture is a Maksutov-Newtonian: a full-aperture meniscus corrector, a spherical primary, a small flat — every optical surface a sphere, because spheres are the only surfaces opticians can make almost perfectly, repeatably, and affordably. And the corrector holds the secondary, so there are no spider vanes, no diffraction spikes — pure planetary contrast. · ⏱ 1:32
+
+
+
+
+ R₂ − R₁ = −0.565 · t
+ The achromat condition: thickness ties the two radii together.
+ With the /0.97 refinement, color error collapses to ~10 µm of focus — inside the depth of focus.
+ The entire color correction of this telescope is one line: R-two minus R-one equals negative 0.565 times the thickness — the achromat condition that ties the corrector's two radii together. Apply the /0.97 spherochromatism refinement and the residual color error collapses to about 10 microns of focus, inside the depth of focus. Chromatic aberration: for all practical purposes, non-existent. · ⏱ 1:39
+
+
+ ◆ SEED PRESCRIPTIONS
+ Two ray-trace-verified seeds
+
+
+ OD 215.0 +0/−0.2 mm CA ≥ 204 mm CT 22.0 ± 0.2 mm ≈ 2.0 kg glass
+ Here are the two ray-trace-verified seed prescriptions, sharing one corrector blank: 215.0 mm outer diameter, at least 204 mm clear aperture, 22.0 mm center thickness, about two kilograms of glass. These are seeds, not gospel — the final prescription co-optimizes with the delivered glass melt. · ⏱ 1:41
+
+
+ ◆ THE FORK
+ MN76 vs MN78 — a binary we inherited.
+
+ Carried for months, from a dead company's catalog
+ Designing our own optics dissolved it
+ Focal ratio is continuous
+ Obstruction = focuser height + illuminated field, not f-ratio
+
+ For months this project carried "MN76 f/6 versus MN78 f/8" as an unresolved either-or — a binary inherited from a dead company's catalog. Designing our own optics dissolved the question entirely: focal ratio is continuous, and obstruction is set by focuser height and illuminated field, not by f-ratio. · ⏱ 1:42
+
+
+ Three millimeters.
+ The real MN76's measured fully-illuminated field: 3 mm.
+ And a cautionary tale about trusting catalogs: when the real MN76 was actually measured, its fully-illuminated field was 3 mm. Three millimeters — smaller than most sensors. That is what you inherit when you don't do the math yourself. · ⏱ 1:43
+
+
+
+ ◆ THE ε SLIDER
+ IC = 1 − ε²
+
+
+
+ 25% obstruction costs 10–18% of mid-band contrast
+ 13.4% costs 3–5%
+ A custom 16–18% recovers most of the f/8 advantage — at any f-ratio
+
+
+
+
Take the slider yourself — drag epsilon and watch the mid-band sag. A 25 percent obstruction costs 10 to 18 percent of mid-band contrast; 13.4 percent costs only 3 to 5; and a custom 16 to 18 percent secondary recovers most of the f/8 advantage at any f-ratio. The integral contrast is simply one minus epsilon squared — obstruction, not focal ratio, is the lever. · ⏱ 1:47
+
+
+ ◆ SAMPLING
+ Nyquist at the cutoff: F = 2·pixel/λ
+
+
+
+ 2.9 µm pixels want f/10.5 at 550 nm
+ 2.0 µm IMX678: f/7.3 at 550 nm
+ … and f/6.1 at 656 nm
+ Nearly native at f/6
+
+
+
+
Sampling closes the loop between glass and silicon: Nyquist at the diffraction cutoff needs a focal ratio of two times the pixel pitch over the wavelength. A 2.9 micron pixel wants f/10.5 at 550 nanometers, but the 2.0 micron IMX678 wants f/7.3 at 550 and f/6.1 at 656 — nearly native at f/6. · ⏱ 1:49
+
+
+ ◆ THE BARLOW LADDER
+ The sensor choice is the cheapest optical upgrade.
+
+ Here is the Barlow ladder — the magnification you must buy in glass to reach Nyquist. With 2.9 micron pixels you need 1.76, 1.51, or 1.32 times at f/6, f/7, f/8; with 2.0 micron pixels it drops to 1.21, 1.04, 0.91 — essentially nothing. The sensor choice is the cheapest optical upgrade in the entire system. · ⏱ 1:50
+
+
+
+ ◆ FOCUS
+ No f-ratio forgives the night.
+
+
+
+ Depth of focus: ±39.6 µm · ±53.9 µm · ±70.4 µm (f/6 · f/7 · f/8)
+ A 22 K night walks focus 734 µm
+ 10–18× past any depth of focus
+ Temperature-compensated focusing: mandatory
+
+
+
+
Depth of focus is plus-or-minus 39.6, 53.9, and 70.4 microns at f/6, f/7, and f/8 — but a 22 kelvin night walks the focus 734 microns, ten to eighteen times past any of them. Temperature-compensated focusing is mandatory at every f-ratio; f/8's supposed forgiveness was never going to save anyone. · ⏱ 1:53
+
+
+ ◆ PROVISIONAL VERDICT
+ Where the design stands tonight
+
+
+ Gate 1 — sensor decision packet
+ Gate 2 — secondary-illumination ray trace
+ Gate 3 — two real corrector quotes
+
+ So the provisional verdict: 180 millimeters, f/6.5 to f/7, obstruction at or under 18 percent — held at 70 percent confidence, and we say that number out loud on purpose. Three gates settle it: the sensor decision packet, the secondary-illumination ray trace, and two real quotes for the corrector. · ⏱ 1:55
+
+
+
+ ◆ THE ESCAPE
+ Grind both surfaces against matched test plates.
+
+ Errors track together — the difference holds
+ Wedge ≤ 1 arcmin
+ Index homogeneity ≤ 2×10⁻⁴
+ Corrector tilt measured in arcminutes — 1° of tilt = 0.27λ of coma
+
+
+ The cell is an optical component
+ The escape from that terrifying tolerance is old shop wisdom: grind both surfaces against matched test plates, so their errors track together and the difference holds itself. Then wedge under one arcminute, index homogeneity under two parts in ten thousand — and note that one degree of corrector tilt costs 0.27 waves of coma, so the cell that holds the glass is itself an optical component. · ⏱ 1:57
+
+
+ ◆ PROOFS, NOT ASSERTIONS
+ design/optical/calc/ — the same pattern as the mount
+
+ Frozen-dataclass prescription
+ Source-tagged provenance on every number
+ Every figure in this deck recomputable in CI
+
+ None of tonight's numbers live in slideware — design slash optical slash calc mirrors the mechanical pattern you'll see next act: a frozen-dataclass prescription where every number carries source-tagged provenance. Every figure I've shown you is recomputable in CI. · ⏱ 1:58
+
+
+ ◆ THE GATES
+ Tests that ray-trace
+
+ pytest gates ray-trace with optiland
+ Polychromatic Strehl ≥ 0.95 gated with prysm
+ Monte Carlo tolerancing for fabrication yield
+
+ And the tests are not unit tests in the ordinary sense: pytest gates that ray-trace the prescription with optiland, gate polychromatic Strehl at 0.95 or better with prysm, and run Monte Carlo tolerancing to predict fabrication yield. If the design drifts out of spec, the build goes red. · ⏱ 1:59
+
+
+ The FAILs are the value.
+ MECHANICAL_DESIGN.md — the engineering culture this optics program inherits
+ This line comes from the mechanical design ledger, and it is the culture this optics program inherits: a verdict that says FAIL is worth more than a slide that says fine. You'll see the mount's own eleven-verdict ledger next act. · ⏱ 2:00
+
+
+ ◆ HONEST LEDGER
+ What remains unproven
+
+ BK7 poly-Strehl ≥ 0.95 at f/6 — a derived expectation until the CI trace says so
+ Achievable ε ≤ 18% — awaits the illumination trace
+ Site r₀ — assumed until measured
+
+ In that spirit, here is what we have not proven. BK7 polychromatic Strehl at 0.95 or better at f/6 is a derived expectation until the CI trace confirms it; the achievable obstruction at or under 18 percent awaits the illumination trace; and the site's r-zero is an assumption until we measure it. · ⏱ 2:01
+
+
+ The eye exists — as mathematics, with error bars.
+ So where does that leave us? The eye of NIGHTWATCH exists — every surface, every tolerance, every open question — as mathematics, with error bars attached. · ⏱ 2:01
+
+
+ Mathematics is free.
+ Glass costs money.
+ And that brings us to the hard part. Mathematics is free — glass costs money, and Act V is about what it takes to turn this prescription into a thing you can hold. · ⏱ 2:02
+
+
+ do-good-us.
+ Act V
+ Metal, Money, and Mirrors
+ Turning computed proofs into purchase orders.
+ Act Five is where the proofs stop being drawings and start being purchase orders. Thirty minutes on metal, money, and mirrors — every claim checked, every dollar named. · ⏱ 2:02
+
+
+ ◆ MAKE VS BUY
+ Four ways to buy this telescope
+
+ We tried to buy this telescope before we ever agreed to build it. New Intes is impossible — production has been dead since 2018 — and a used MN76 or 78 means $1,500–2,500 after months of watching, with an optical figure you cannot verify and Astrositall units that are unicorns. APM's raw Intes optical sets are gone — we verified in August 2026 that they carry none — and the Starlux 190MN, $2,375 and in production, fails the requirements at 27–34% obstruction in borosilicate. · ⏱ 2:04
+
+
+ Custom — by requirement, not by taste.
+ The Starlux 190MN stays on the books as the named interim / contingency OTA.
+ So the verdict is custom, by requirement — the market simply does not sell what the error budget demands. And we hedge like engineers: the 190MN stays on the books as the named interim and contingency OTA. · ⏱ 2:06
+
+
+ ◆ VENDOR FORENSICS 2026
+ The market said no, three times
+
+
2025-08-15
+
Zambuto Closed to new orders.
+
2026-08
+
ISTAR Optical Not accepting orders.
+
2026-08
+
Optimax cost estimator Dead DNS + 404 · archive.org shows it was login-walled even when alive.
+
The 2026 vendor forensics were sobering: Zambuto closed to new orders on 2025-08-15, ISTAR Optical is not accepting orders, and Optimax's famous online cost estimator resolves to dead DNS and a 404. We checked archive.org — it was login-walled even when it was alive. · ⏱ 2:07
+
+
+ The market is thin. We verified every contact.
+ No forum lore — every status on that wall is a first-hand check.
+ The precision-optics market is thinner than the forums believe, and we did not take anyone's word for it. Every vendor status you just saw is a contact we verified ourselves this year. · ⏱ 2:09
+
+
+ ◆ TENDER BP-06
+ Three lots, one bid package
+
+ Tender BP-06 mirrors the mechanical bid-package anatomy: three lots. Lot A, the primary, leads with Ostahowski Optics — interferograms ship with every surface, and a 10-inch quartz comp prices at $2,800 — with Lockwood, Optical Mechanics, and Orion Optics UK as alternates. Lot C is a solved problem: the Antares 1.83-inch 1/30-wave flat, $181.49, straight from a catalog. · ⏱ 2:10
+
+
+ ◆ LOT B — MAKE OR BREAK
+ The corrector decides the budget
+
+ 60–75% of optics cost lives in this one lot
+ Required clause: design-assist
+ Final radii co-optimized to the delivered blank’s melt data
+
+
+ Optimax
+ Knight Optical UK
+ United Lens
+ Advanced Glass Industries
+
+ Lot B, the meniscus corrector, is the make-or-break lot — 60 to 75 percent of the optics cost lives here. Four candidate shops — Optimax, Knight Optical UK, United Lens, Advanced Glass Industries — and one clause that is required, not preferred: design-assist, meaning the final radii are co-optimized to the delivered blank's melt data. · ⏱ 2:12
+
+
+
+ ◆ ACCEPTANCE — WRITTEN FIRST
+ The pass bar, in ink, before any money moves
+
+ Transmitted wavefront ≤ λ/4 P-V · ≤ λ/14 RMS
+ Strehl ≥ 0.95 at 546–550 nm
+ Double-pass autocollimation at 20 ± 2 °C, after thermal soak
+ Optic unstressed in the test mount
+ Zernike table + RAW interferogram data delivered
+
+ The acceptance criteria were written before any money moves: transmitted wavefront at or under λ/4 peak-to-valley and λ/14 RMS, Strehl at least 0.95 at 546–550 nm, verified by double-pass autocollimation at 20 plus-or-minus 2 degrees C after a thermal soak, with the optic unstressed in its test mount. The Zernike table and the raw interferogram data are deliverables, not favors. · ⏱ 2:16
+
+
+ ◆ THE PAYMENT GATE
+ One independent interferogram, or no wire
+
+ Orion Optics UK test service
+ AstroReflect
+ AiryLab
+
+ The optics analog of a PE stamp.
+ And one more gate: a single independent verification — Orion Optics UK's test service, AstroReflect, or AiryLab — stands between acceptance and payment. It is the optics analog of a PE stamp: someone with no stake in the sale signs the numbers. · ⏱ 2:17
+
+
+
+ ◆ WHAT THE FAILS BOUGHT
+ Deflection and uplift
+
+ 43.5 arcseconds of deflection against a 5-arcsecond target became the bearing-span fix — DEC at least 105 mm, RA at least 125 mm, a machining change, not a part swap. And 9.2 kN of roof uplift against 1.8 kN of self-weight — safety factor 0.19 — became four 2,000-lbf hold-down anchors at safety factor 3.9. · ⏱ 2:21
+
+
+ ◆ WHAT THE FAILS BOUGHT
+ Tracking, heat, and power
+
+ A 5.02-arcsecond tracking RMS bought the on-axis absolute encoder. A 22 K thermal walk bought the temperature-compensated focuser, and a 287 W night load bought the 48 V architecture. Five failures, five line items of better hardware. · ⏱ 2:22
+
+
+ We publish our failures with safety factors attached — that is why you can trust our optics claims.
+ the mount program — 11 computed proofs, published
+ This is the whole culture in one sentence. A team that hides its FAILs will hide a bad interferogram; a team that publishes them, with safety factors attached, has nothing to hide from your glass. · ⏱ 2:24
+
+
+
+ ◆ COST OF THE GATES
+ Paper first, dollars late
+
+ P0–P2: $0 in hardware
+ P4: first dollars at risk
+ Corrector is the critical path — 3–6 months
+ 6–12 months end-to-end
+
+ P0 through P2 cost $0 in hardware — every expensive decision was made on paper. The corrector is the critical path at 3–6 months, and the whole program runs 6–12 months end-to-end. · ⏱ 2:27
+
+
+ ◆ RISK REGISTER — SUPPLY
+ The honest top four, part one
+
+ The risk register, honest top four, starting with supply: corrector single-vendor concentration is answered with two quotes including one non-US channel, and Sitall unobtainium is already absorbed — the spec calls a CTE class, not a brand of glass. · ⏱ 2:28
+
+
+ ◆ RISK REGISTER — ASSUMPTIONS
+ The honest top four, part two
+
+ Then the assumptions we admit to: the corrector price band is estimator-grade, so P3 anchors it with real quotes; and the site seeing is assumed, not measured — P6 measures it with a DIMM. An assumption named out loud is a risk half-retired. · ⏱ 2:29
+
+
+ Money moves only after two real quotes and one independent interferogram.
+ That is the rule, and it is absolute: money moves only after two real quotes and one independent interferogram. Not enthusiasm, not renders — evidence. · ⏱ 2:31
+
+
+ And when the glass finally sees sky…
+ Act VI.
+ The tender is written, the gates are set, the money knows its cue. And when the glass finally sees sky — that is Act Six. · ⏱ 2:32
+
+
+ velmu-sky. presa-when-called.
+ Act VI
+ First Light
+ One night, hour by hour — and every night after.
+ Everything you have seen tonight was built for one thing: a night alone under the sky. This is what that night looks like — hour by hour, decision by decision. · ⏱ 2:33
+
+
+ ◆ A NIGHT IN THE LIFE
+ Dusk to the flip
+
+
19:42
+
Dusk — plan loads M27 → NGC 7331 → M31 · Saturn opportunistic · seeing forecast 1.1″
+
20:10
+
“Nightwatch, open the roof” Voice command · roof 0→100%
+
20:25
+
First light Slew 4°/s · plate-solve syncs 0.8″ off
+
+
00:28
+
Meridian flip Fully choreographed
+
At 19:42 the plan loads itself against a 1.1″ seeing forecast; at 20:10 one spoken sentence takes the roof from 0 to 100%. By 20:40 guiding has settled to 0.8″ RMS, and at 00:28 the meridian flip runs fully choreographed — nobody touches anything. · ⏱ 2:36
+
+
+ ◆ A NIGHT IN THE LIFE
+ And then it rains
+
+
02:10
+
Humidity 75% → MARGINAL Amber wash
+
02:25
+
Both rain sensors trip → EMERGENCY CLOSE Capture cancels ≤2 s → mount parks → roof shuts → all channels notified
+
03:40
+
Rain stops 30-min holdoff · shortened plan resumes
+
05:15
+
Dawn park The night ends unattended
+
At 02:25 both rain sensors trip and the emergency close runs: capture cancels in under 2 s, the mount parks, the roof shuts, every channel is notified — calm, not chaotic, and that calm is the demo's proof of trust. Rain stops at 03:40, a 30-min holdoff clears, a shortened plan resumes, and at 05:15 it parks itself at dawn. · ⏱ 2:39
+
+
+ ◆ MORNING REPORT
+ What the machine wrote at dawn
+
+ One morning report: 214 frames captured, 196 kept, 4.9 h of integration at 0.83″ RMS — through one weather interruption. And the number this whole night was built for: zero human interventions. · ⏱ 2:41
+
+
+ ◆ THE NEW EYE
+ What the new optics add to that night
+
+ Mars at the 0.78″ Rayleigh limit
+ Lucky-imaging pipeline, scored against the Strehl budget
+ A mirror that cannot remember temperature
+
+ Drop the 180 mm Mak-Newt into that same night and three things change: Mars resolved at the 0.78″ Rayleigh limit, a lucky-imaging pipeline that scores every frame against the Strehl budget, and a mirror that cannot remember temperature. Same night, sharper eye. · ⏱ 2:43
+
+
+ ◆ THE WATCH
+ Presence, not events
+
+ The NEO scanner runs hourly, regardless
+ Most scans end quiet — and that is the point
+
+
+ sky-quiet: ne-events-found
+ Rain or shine, roof open or shut, the NEO scanner runs hourly — and most scans end with this line: sky-quiet, ne-events-found. That is the point of a watch: presence, not events. · ⏱ 2:44
+
+
+ ◆ NEXT STEPS
+ From here
+
+ P0 — decision packets: sensor baseline · v1 tracking question · focuser steps→µmP1 — CI proofsP2 — the fork retires
+ The road from here is concrete: P0 closes the decision packets — the sensor baseline, the v1 tracking question, focuser steps to µm — then P1 turns claims into CI proofs. The fork retires in P2. · ⏱ 2:45
+
+
+ A machine in the high desert that watches with full presence, speaks when spoken to, closes the roof when it rains — and waits for photons that left before it existed.
+ Read it once, slowly. That is the whole project in one sentence. · ⏱ 2:46
+
+
+
+ NIGHTWATCH
+ first light awaits · 🜏
+ Thank you — for three hours of your evening, and for every night after. [Hold. Applause.] · ⏱ 2:47
+
+