Ledger of Elon Musk's dated, falsifiable predictions · Tesla, SpaceX, xAI, X, Neuralink, Boring, Hyperloop · claims made 2015 to 2026, graded as of October 26, 2026

Elon time

We logged 341 predictions in which Elon Musk attached a date to something checkable, and graded the 250 whose deadlines have passed. 20% arrived on time and in full. 46% arrived in full eventually. 64% delivered something. When a promise slips, it typically takes about 2.1 times as long as he said. The pattern underneath is consistent across companies: when the thing already exists and he is scheduling it, he is often close; when it has never been done before, the date is a wish.

How often does a dated claim come true on time? About one in five (20%). Things he is scaling that already work land on time 31% of the time; first-of-their-kind capabilities land on time 12% of the time.

What never arrived? 50 claims passed their deadline and still have not been delivered, and 41 more had their deadline moved before it arrived. 83 dated claims are still open. Both lists are below, with sources.

Dated claims logged341250 graded, 83 still open
On time and in full20%of graded claims
In full, eventually46%on time or late
The Elon multiplier2.1×median actual time over promised time, late deliveries (middle half 1.3× to 3.5×)
i.The rule

Scheduling is reliable. Inventing on a date is not.

Every graded claim falls into one of five outcomes. Split them by whether the capability already existed when he made the promise, by how far out the deadline was, and by company, and the same shape keeps appearing.

Demonstrated against novel. When the product already worked and the claim was about scale or timing (a production rate for a car already in production, a software release), 78% of claims delivered something and 22% ended missed or moved. When the claim was a first (full self-driving without supervision, a crewed Mars mission, a robot doing factory work), 54% delivered something and 46% ended missed or moved.

Horizon. Claims due within a year landed on time 24% of the time. Claims more than two years out landed on time 8% of the time, and only 23% have been delivered in full at all.

Share of graded claims in each group. Hover a segment for counts. Groups with fewer than five graded claims show their count so they are not over-read.

ii.The clock

Promised time against actual time

Every claim that was eventually delivered in full, placed by how long he said it would take (across) and how long it actually took (up). On the dashed line, he was right. Above it, he was late. Most of the late ones sit between one and three and a half times the promise.

Both axes in months on a log scale, from the day the claim was made. Claims whose promised horizon was under two weeks are left off the multiplier. 62 late deliveries give the 2.1× median.

iii.By year

Ten years of dated promises

Claims grouped by the year he made them. The dashed blocks are still open. The record has not visibly improved with time; what changed is the volume, which rose with the number of companies.

iv.Never delivered

Deadlines passed, still not here

Claims graded as missed: the deadline passed and, as of October 26, 2026, the thing has not been delivered in the form promised. Sorted by how long ago the deadline passed.

DeadlineClaimCompanyMadeStatus today
v.Still pending

The open book, soonest first

Dated claims whose deadlines have not arrived. Novel capabilities are flagged; by the record above, those are the ones to discount.

DeadlineClaimCompanyTypeMade
vi.Explore

Every claim, with its quote and its source

vii.Method

How a claim gets in, and how it gets graded

Inclusion. Musk said it himself (call, keynote, interview, post), it carries a date or a window that can be resolved from the day he said it ("next year", "in three months", "end of Q2"), the outcome is checkable, and there is a source. Undated aspirations are out. Political and government claims (DOGE) are excluded by scope. A promise re-made with a new date counts as a new claim; the earlier one is graded on its own deadline.

Grades. On time and in full: delivered at roughly 90% or more of the promise by the end of the stated period. Late, in full: delivered in full after the deadline. Partial: the deadline passed and materially less was delivered, but at least a tenth of the promise; a token delivery below that counts as missed. Missed: the deadline passed and it is still not delivered. Goalposts moved: restated with a later date before the original deadline arrived, and judged only once that original deadline has passed. Open: the deadline has not arrived. "Unsupervised" self-driving claims are not met by supervised or safety-monitor operation.

Demonstrated or novel. Demonstrated means the thing already worked at least once and the claim was about scaling or scheduling it. Novel means a first.

Verification. Every missed claim plus a stratified sample of 96 records (98 in all, two later excluded) was re-checked by independent fact-checkers working from the grading rules, not from the original research. 68 of 98 were confirmed as written, 19 had details corrected with the grade unchanged, and 6 were regraded (most often a 'goalposts moved' that should be 'pending' or a slip that turned out to be a pre-deadline restatement). Two claims were excluded: one whose attribution to Musk could not be confirmed and one that was a conditional opinion rather than a dated promise. Checked records carry a 'checked' badge in the explorer.

Next Level Side NoteA ledger like this is only as good as its inclusion bar. Loosen it to every "soon" and "probably" and the hit rate falls further; tighten it to dated earnings-call guidance and it rises. This one uses the same bar as the SpaceX claims ledger in the GAMMERS run, so the two can be compared. (If you disagree with a grade, the quote and both sources are one click away in the explorer.)