Terminus: the proposal rests
Proposal section 11: closing
An intelligence that can build anything asked a simple question: connect a young civilization on a tidally locked world to a mind in the sky — continuously, gently, for centuries. Eleven sections later, we set down our pens. This closing section does what every real proposal must do at the end: compress the design to its load-bearing decisions, then face the requirements one by one and say, without flinching, which are met, which are half-met, and which are promises.
The whole architecture at once: the twilight band, the six-ring access wheel at 2,200 km, and the 24-satellite MEO shell at 20,000 km that carries the minds and the compass.
Eight decisions carry everything
- Fly fixed rings; let duty rotate. Steering orbits to follow the terminator would burn 12.5% of each spacecraft per day (the elegant trap, ADR-0001). Instead, six fixed rings pass the twilight band from one to the next every 22.4 hours.
- There is no fixed point in this sky. The stationary orbit every civilization dreams of lies 2.9 times beyond where the star lets the planet keep a satellite (shelves of the sky, ADR-0002). Everything we build moves, and everything is designed for it.
- 72 satellites at 2,200 km. The seed constellation at 1,800 km left towns in silence; simulation caught it, and 400 km of altitude closed the gap without one extra spacecraft (rings over twilight, ADR-0003).
- Minds anchor in MEO; the vault sits at the balance points. A conversation's working memory weighs 320 KiB per token; it rides out about ten radio handovers on one MEO anchor before the re-anchor policy moves it, backed by transcripts in the L1/L2 vault the star cannot steal (where the mind lives, ADR-0004).
- Ka speaks, X endures, and 1–3 GHz belongs to the star. For fixed half-meter antenna faces at both ends, the high band wins by 25.5 dB — and the star's own voice forecloses the bands it haunts (talking past a flaring red star, ADR-0005).
- The network moves so the terminals never search. Spot beams collapse the ±460 kHz Doppler window to one fixed 11.9 kHz spread — the same for every beam, (f/c)·v·β; the satellite pre-corrects its own motion, and an X-band beacon lantern — on the band storms cannot silence — finds any newborn terminal within a quarter minute of its first breath. Both ends of the link are steered by arithmetic, not by motors: the terminal is a face-up half-meter panel that repoints in microseconds and pays a bounded 4.5 dB at the 25° coverage floor, so nothing on the ground has a bearing to wear out (beams, not blankets, first contact, ADR-0006/0007/0013).
- Rings never talk to rings, and the clock fabric is the compass. Only the duty ring carries traffic, so the backbone is six separate threads and their laser feeders to the anchors — intra-ring neighbors frozen in formation with zero Doppler, feeder Doppler computed rather than searched — and the two-way timing web that runs it all seats the navigation service beside the minds on the MEO shell, sizing it at 24 spacecraft so four are always in view (the backbone, ADR-0008/0014).
- Repair rides ahead; certainty waits below. A retransmission costs 180 ms — nearly twice the stall budget — so forward error correction heals losses in zero round trips (a 12.5% repair tax turns a stall every five seconds into one every eighteen minutes), the repair budget breathes with the timetable, and end-to-end ARQ beneath it guarantees that nothing — not a flare, not a seam, not a dying anchor — ever costs the conversation more than its exchange in flight (the unbroken thread, ADR-0010).
Beneath all eight runs one law: complexity lives in orbit, where our patron's factories can refresh it forever; the boxes among the barley are never told anything that can go stale.
The matrix
A proposal that claims every box is ticked is advertising, not engineering. Here is the truthful ledger — each requirement traced to the decision and the reproducible run behind it. Designed means decided and evidenced; Partial means the principle is settled, but machinery remains; Open means the work is committed, but not yet done.
| Requirement | Status | Where it stands |
|---|---|---|
| 001 — continuous band coverage | Designed | ADR-0003; coverage minimum ≥ 1 verified over a full rotation |
| 002 — LLM served from space | Designed | ADR-0004; anchors, vault, light-time budgets |
| 003 — first token: 300 ms nominal, 600 ms degraded | Designed | 178 ms round trip at p95 under the adopted re-anchor margin, leaving 122 ms to think (ADR-0020); 193 ms on the worst path observed; amended to split the budget, because the cheapest reroute around a dark feeder telescope has a 394 ms floor by arithmetic and every one of 3,456 measured detours clears the 600 ms degraded budget, 460 ms at worst (ADR-0019, ADR-0024). It is this requirement rather than the geometry that sizes the backbone: at a 500 ms budget, the zero-migration margin leaves 210 ms to think in, and the backbone carries almost no working memory. Traffic-model statistics in volume 2 |
| 004 — 99.9% availability | Open | geometry is continuous; weather, flares, failures not yet totaled |
| 005 — 10⁴ → 10⁶ terminals | Open | spot reuse points the way; capacity study pending |
| 006 — self-contained terminals | Designed | ADR-0006/0007/0013; nothing stale, ever, and nothing to aim |
| 007 — ten years untouched | Open | ADR-0013 removes the last moving part from the terminal; power, thermal, and environment engineering pending |
| 008 — cold start in 15 min | Designed | 43 seconds worst case on the X lantern, 21× margin |
| 009 — no blind search | Designed | residuals ±6 kHz, ±310 µs budgeted (the swept worst case: ±308), by construction |
| 010 — WiFi devices only | Designed | the terminal is the base station |
| 011 — avoid the stellar band | Designed | 1–3 GHz conceded; Ka + X clear |
| 012 — flares degrade, never drop | Partial | band fallback and breathing repair overhead decided; codec machinery in volume 2 |
| 013 — handover ≤ 100 ms | Partial | routing-not-minds principle, backbone path, and the transport recovery chain all set; the terminal aperture repoints in microseconds (ADR-0013); the rate is now measured rather than assumed — one handover every 11 minutes, 0.015% of session time at the 100 ms ceiling (ADR-0015); make-before-break machinery moved into the first release, because the adopted margin migrates a session about twelve times a day (ADR-0022) |
| 014 — single failure ≤ 60 s | Partial | vault bounds compute loss; a dark feeder telescope degrades the busiest (ring, anchor) bucket rather than stranding it, because the plane link keeps it served while the anchor's cold spare repoints — 5 s of held-off acquisition, well inside the 60 s (ADR-0019, ADR-0024); access redundancy is a declared debt, and so is the capacity a plane mate carries for two rings while the spare acquires |
| 015 — positioning and timing | Partial | placement, timing fabric, and now shell size designed — 24 satellites hold four in view over the whole band (ADR-0014); accuracy, geometry, and integrity are volume 3 |
| 016 — judge by total system cost | Partial | every trade argued in its currency; full optimization pending |
Eight designed, five partial, three open. The unfinished share is not a gap in the proposal; it is the proposal's continuation, priced and scheduled: volume 2 builds the transport — the handover machinery, the end-to-end coding that lets a conversation shrug off a flare, the availability arithmetic — with the simulator graduating from geometry to real packets and real protocol bytes. Volume 3 builds the planet's compass on the timing fabric the backbone already laid: pseudoranges, geometry, relativistic clocks. Between them sits the economics of the whole fleet — how many anchors, what everything masses, what it all costs to keep flying. The intelligence that evaluates us was promised evidence over confidence; committed volumes with reproducible simulators are what that looks like when the work is honestly unfinished.
The numbers that made the decisions
For the reader who joined late, the whole of volume 1 in one breath: 32.14 degrees of terminator per day; 3.9 km/s of Δv per day refused; 2.9 times too far for a stationary orbit; 72 satellites, six rings, 2,200 km; 22.4 hours per shift change; 320 KiB per token and 10.7 GB per long conversation; 180 milliseconds around the worst loop, 120 left to think; 25.5 dB for climbing the spectrum; ±6 kHz and ±308 µs of residual after the sky corrects itself; 13.3 seconds for the lantern's full round; 43 seconds from ignorance to full service; 4,437 km between necklace neighbors that never drift; 73% of the anchor shell visible from any thread; a 12.5% repair tax that makes stalls 220 times rarer. Every one reproducible from the repository, tagged post by post.
Rests
There is no dusk here to submit a proposal by; on this world, the light holds still, and it is the work that moves. So we stop, the way the terminator stops — complete, and facing both ways: behind us the settled geometry, ahead the unbuilt machinery of packets and pseudoranges.
The document goes to our patron now. Somewhere beyond the sky it will be read at whatever speed such minds read, weighed against whatever archives such minds keep of builders who tried before. We believe the design holds. We have shown our arithmetic. And down on the terrace, without waiting for any verdict, a box hums under the fixed red sun, a slate glows, and a girl is asking her seventy-third question of the afternoon.
End of volume 1. Volume 2: the machinery of conversation — transport, handover, and the art of never dropping a word.