On the record
Recorded 2026-07-1036 claimsThe future-facing claims behind Moduloa Manufacturing, boiled down and frozen at recording: the original 25 on 2026-07-10; later additions carry their own recorded date. Each claim stays open until reality answers it. Then it is scored Correct, Partly, or Wrong, in public. The claims never get edited; they only get scored.
One investigation, four instruments. The thesis states the bet. The register freezes it into dated claims, scored in public. The tracker follows its premise: whether humanoids are becoming ready for factory work. The Journal records the evidence as it arrives.
30 years of deadlines36 claims, 2026 to 2056. 11 of them fall due in 2036.
Bar height is the number of claims due that year, and the spacing is real time, which is why the two decade gaps are empty. A claim still open when its deadline passes is scored wrong automatically.
By the end of 2056: all 36 claims are due. Scored so far: 1.
Point at a claim to see its deadline and its status today.
Scores are today's. Nobody knows the future ones yet, which is the point of the record. A claim still open on its deadline is scored wrong automatically.
A thesis that cannot be wrong is not a thesis. This register makes the bet auditable: anyone, including us, can return in 2031, 2036, 2046, and 2056 and see exactly what was claimed before it was obvious, and what the claims got wrong. See the live humanoid-race tracker →
Statuses: Open (undecided), Correct, Partly (directionally right, materially off), Wrong. Every claim carries a due date. A claim still Open when its due date passes is scored Wrong: no extensions, no reinterpretation. A score only lands with public sources and the date it was decided, and a claim scored Wrong stays on the page. That is the point.
The register is reviewed weekly by an automated research pass that looks for public evidence on open claims. It is deliberately conservative: no source, no score. Most weeks it changes nothing.
Single-event claims: one date, one outcome, no room to reinterpret. Context for this one: on 2026-07-05 Hyundai already put Atlas on the pitch at the Round of 16, delivering the halftime match ball (its first live-match robotics integration). The claim, recorded 2026-07-15 with no announcement to lean on: the majors now use the biggest broadcast stages to show what their humanoids can do, and Hyundai escalates this to the final itself.
Recorded 2026-07-22, out of radar research on the micro-factory layer (see the journal note of the same date). Two claims that test the thesis from below: whether process knowledge really becomes portable between independently owned standardized cells, and whether routing-without-owning survives contact with reality.
Recorded 2026-10-12 by the founder, from the thesis companion Software is done, hardware is next. The bet that writing code stops being the scarce input, with a date on it and a number to score it by. If agents still manage only about three hours of professional work reliably at the end of 2027, the curve stalled and this is wrong.
Recorded 2026-10-12 by the founder. The other claims bet on the industry; these six bet on Moduloa itself, each tied to work on the task board. None of them says the work is finished. Each says the groundwork exists: a first version, in public, that others can build on and correct. We expect many early contributions to come to nothing, and a few people to carry most of the load in building the fundamental models; these dates are for the foundations they would build on. The rules are the same as for every claim here: the text is frozen, and a claim still open on its deadline is scored wrong in public, with no extension.
Claims about Moduloa, held to the same standard as claims about the world.
The near bet: humanoids leave the demo reel and enter real production.
Recorded 2026-07-26, out of the QA pass on the model's phase pages: Quality & Logistics was the one phase no claim tested head-on. This closes that gap: it bets that quality results feeding a living certification score become a real commercial instrument, not just a diagram.
The structural bet: labor cost stops deciding where things get made.
The systems bet: manufacturing capacity becomes infrastructure.
The societal bet: the questions change entirely.
Where the Journal has discussed each claim
Reading, not scoring. These are the notes that name or discuss each claim, oldest first. Some say plainly that they do not move it, so read the note. A note never scores a claim: the weekly review does that, and only from two independent sources outside this site. A note marked in words discusses the claim without naming its id.
10 July: The state of humanoid work, mid-2026 (in words) · 1 August: Monthly recap: July 2026 (in words) · 2 October: A humanoid maker opened its books
2 October: A humanoid maker opened its books · 3 October: The new job is the real test for humanoid robots
10 July: The state of humanoid work, mid-2026 (in words) · 1 August: Monthly recap: July 2026 · 26 September: The robot was never the expensive part · 2 October: A humanoid maker opened its books
29 August: The interface is being written now, and it does not say humanoid · 12 September: The measurement layer arrived, and it points at the robot · 3 October: The new job is the real test for humanoid robots
29 August: The interface is being written now, and it does not say humanoid · 5 September: Monthly recap: August 2026 · 19 September: The standards landed below the work · 3 October: The new job is the real test for humanoid robots · 12 October: The work went to the integrators. The money went to the factories.
29 August: The bottleneck was never the machines · 4 September: Speed came back, price did not · 5 September: Monthly recap: August 2026 · 26 September: The robot was never the expensive part · 3 October: Monthly recap: September 2026 · 12 October: The work went to the integrators. The money went to the factories.
28 July: Capacity is a membership (in words) · 12 October: Someone has to own the hall · 12 October: The work went to the integrators. The money went to the factories.
26 July: The intelligence layer arrives first · 29 August: The interface is being written now, and it does not say humanoid · 5 September: Monthly recap: August 2026 · 12 September: The measurement layer arrived, and it points at the robot · 19 September: The standards landed below the work · 2 October: A humanoid maker opened its books · 3 October: The new job is the real test for humanoid robots · 3 October: Monthly recap: September 2026 · 12 October: The work went to the integrators. The money went to the factories.
29 August: The bottleneck was never the machines · 5 September: Monthly recap: August 2026 · 26 September: The robot was never the expensive part · 12 October: Someone has to own the hall
1 August: Monthly recap: July 2026 · 5 September: Monthly recap: August 2026 · 3 October: Monthly recap: September 2026
22 July: The box beats the body, for now (in words) · 26 August: Alphabet published the cell, not the standard (in words) · 29 August: The bottleneck was never the machines · 29 August: The interface is being written now, and it does not say humanoid · 5 September: Monthly recap: August 2026 · 12 September: The measurement layer arrived, and it points at the robot · 19 September: The standards landed below the work · 3 October: The new job is the real test for humanoid robots · 3 October: Monthly recap: September 2026
No note names these yet: P-11, P-16, P-17, P-19, P-20, P-21, P-22, P-23, P-25, P-30, P-31, P-32, P-33, P-34, P-35, P-36.
If this thesis fails, the most likely wrong assumption is one of these four: humanoid timing, humanoid reliability, capital intensity, or customers' willingness to trust portable production. Written down now so a failure can be diagnosed against what we actually believed, not what we later claim we believed.
Boiled down from the working thesis. Full context: the Manufacturing thesis, v0.2. Machine-readable register: data/predictions.json.