The show's memory, in the open.

Drag to move a node, hover to see its neighbours, click to open the page. Ideas are the small grey dots: each one hangs off the episode it came from and the entities it touches.

Not a summary. A dated line for each step, and the episode that caught it. If nothing stepped, nothing is written.

1. 29 Aug 2026
   The AI trade posts its receipt. NVIDIA reports $96.221B for the quarter, +106% YoY, data centre $89.0B (+117%), net income $59.7B at a 75.0% gross margin, guiding $108B. Verified from the company release, not the tape. The step is not the growth — it is that the number is now large enough that "why am I not just calling NVIDIA?" is a serious question for a small operator. → Moonshots #284 @0:00 · NVIDIA
2. 29 Aug 2026
   Concentration risk gets named out loud. "It's super hard when you have a single point of failure… so why has it been maintained for so long?" — a systems-engineering reading of a market story the panel treats as an efficiency. → Moonshots #284 @0:00 · TSMC
3. 29 Aug 2026
   Chinese video models are a trust-arbitrage play, stated plainly. Blundin: "the Chinese don't have the ability to sell Kimi or Qwen into the enterprise use case because nobody knows if they can trust it. And video is perfectly profitable — a fantastic way to take your maybe-not-trustworthy model and get it to market." → Moonshots #284 @29:27
4. 29 Aug 2026
   Vehicle surveillance reaches ~20 billion plate scans a month across 100,000+ readers, 5,000+ agencies — and ~50 cities and counties have already cancelled or switched off their contracts. The step is the reversal, not the scale. ⚠️ Flock states its cameras do not use facial recognition — vehicles, not faces. → Moonshots #284 @45:00
5. 29 Aug 2026
   The data-centre bargain shows its first full receipt. Quincy, WA: ~30 data centres now ≈57% of local property-tax revenue, funding a high school, hospital, library, police and fire stations. ⚠️ The poverty figure (29.4%→6.2%) is an ACS five-year estimate, ±4.1 points, and moved 13.1%→6.2% between editions — it does not establish causation. ⭐ He independently proposed "refurbish the libraries to be AI centers" — the library is already in CNN's list of what the tax revenue built. → Moonshots #284 @1:22:31
6. 27 Aug 2026
   A frontier-lab CEO retreats on timelines in public. Altman: "We've all been too ambitious on timelines even with this incredible technology" — reframing to "something slower, more like a rising tide." → Moonshots #283 @0:00
7. 27 Aug 2026
   Agent swarms arrive as a consumer product. Grokbot launched 11 Aug; Emad Mostaque runs 18 in parallel across his own hardware. Salim Ismail names the shift: "staff recruiting time and coordination cost has gone to pretty much zero… Humans set objectives and leave everything else to the AI." → Moonshots #283 @22:58
8. 27 Aug 2026
   The bottleneck is publicly renamed from execution to judgement. WSJ, read out on air: "The founder used to manage five tasks, now manages 50 agent outputs. The bottleneck has shifted from execution to judgment." ⭐ This is GForce's own review-debt constraint, described from the other side of the gap. → Moonshots #283 @1:20:02
9. 27 Aug 2026
   Autonomy hardware falls ~83% in one generation — $115k → $20k, on a custom 5nm chip, with a vehicle OEM'd from a Chinese manufacturer. (On-air figures, unverified; 115→20 is 82.6%, consistent with the title.) → Moonshots #283 @1:38:36
10. 27 Jul 2026
    The man who sets NASA's demand says the lunar economy is not the demand. Jared Isaacman, sworn in as 15th Administrator seven months earlier, on what he cannot promise: "I can't guarantee that we can get more value out of the lunar regolith and all the cost that goes into getting there." What he does back instead is orbital data centres — "another leg beyond launch, observation and communications… the math closes" — as the thing that funds the Moon. ⭐ The step is a government buyer publicly pricing his own programme's economics. → Moonshots #274 @18:56, @19:47
11. 27 Jul 2026
    A sitting NASA Administrator says he wants to hand out more prizes. "The America COMPETES Act actually allows us to put prizes out there and I'm very open to doing it… we could work, or partner" with XPRIZE. Said in the same season NASA issued LEIA and SPARC. → Moonshots #274 @51:36
12. 26 Jun 2026
    "Space for Earth, not space for Mars." Planet Labs frames Earth observation as ~3,000 data layers × 10 years of daily history — the panel's "crystal ball for the earth." → Moonshots #266

The counterpart to their disclosures ledger. They say what they are long; this says where I stand: the position, the gap, the season, and a two-line log every week of what actually moved.

Index cards for the source episodes. The transcript and the notes stay in the research files; the card is the pointer.

A company earns a page when it has come up in two or more episodes, or when it is being actively researched. Cited is not the same as notable.

People who recur across episodes. What they are interested in, what they disclose, and what they keep coming back to.

The shows this archive reacts to, taken apart: how they are built and where their material comes from.

The ideas the show keeps invoking, and the ones with my name on them.

My open questions. A question asked twice is the strongest signal in this whole archive.

The half of the vault where I am the source. A method with no worked example is a claim, so every one links to where it was used, and says where it fails.