Backtest

The result

At 2010 the score ranked 13 of 21 pairs the right way (0.62); counting papers alone ranked 14.5 (0.69), and 2,694 of 10,000 shuffles did as well. By the rule fixed in commit 094380f on 2026-09-22, the score did not rank arrivals.

Without faster-than-light messaging and full-dive, the score ranked 6 of 13 pairs (0.46) and papers alone 8 (0.62).

Revised after the result

At 2010 the score ranked 11 of 18 pairs the right way (0.61); counting papers alone ranked 12.5 (0.69), and 3,173 of 10,000 shuffles did as well. By the rule fixed in commit 3c855e1 on 2026-09-30, the score did not rank arrivals.

Without faster-than-light messaging and full-dive, the score ranked 5 of 11 pairs (0.45) and papers alone 7 (0.64).

Each cutoff, revised:

• 2010: 11 of 18 pairs (0.61); papers alone 0.69; 3,173 of 10,000 shuffles as high; anchor cloud computing
• 2013: 5 of 7 pairs (0.71); papers alone 0.79; 2,961 of 10,000 shuffles as high; anchor cloud computing
• 2016: 1 of 1 pair (1.00); papers alone 1.00; 4,952 of 10,000 shuffles as high; anchor cloud computing

Commit

3c855e1 on 2026-09-30

Revision sha256

9dc963e8c05b632b60971e4ea42c18e3ceef050198b92ab60582895a305dff2d

First pre-registration sha256

645456e20d913b9c74d4be3880ca9b9617801707303fbc2aeb85d8bbf6ef9f06

First result sha256, known when revised

c12150b67873f653955814d7fe22646fccdd5f12f0647941b098195ed7657a14

Ruled

2026-09-30

Run

Python 3.14.3; the shuffle counts belong to this interpreter

This pre-registration was ruled on 2026-09-30, after the first result was known, and committed before any score was computed from it. Its arrival rule, cutoffs, measure, seed and anchor candidates are the first's; only the outcomes in the table changed. The pair counts, 18, 7 and 1, follow from the arrival years alone and were known before the commit.

What was fixed first

Commit

094380f3e745833bed5d403684219a8c750375e3 on 2026-09-22

Pre-registration sha256

645456e20d913b9c74d4be3880ca9b9617801707303fbc2aeb85d8bbf6ef9f06

Run

Python 3.14.3; the shuffle counts belong to this interpreter

Cutoffs

2010, 2013 and 2016; the headline is 2010

Arrival

the first year a product in the category reached a buyer: delivered to someone other than its developers and funders, by sale, lease or service performed; a contract signed for something not yet delivered does not count, and neither does a demonstration. Wearable computing's years are the ones that chain declared on 2026-09-20, before any backtest existed, each the first consumer or mass-market product of its category; they are kept as declared, and each entry's note says where the strict rule finds an earlier delivery

Measure

Within one chain, every two tested products of which one arrived before the other, or the other not by 2025; right when the earlier scored higher at the cutoff, half for a tie, scores compared at 2 decimals

Baseline

The same pairs, read by each product's research works counted from 2000 to the cutoff

Shuffle

10,000 shuffles of the scores within each chain at each cutoff, seed 20260921

Verdict rule

At 2010, the score ranked arrivals only if its share is above 0.5, above the baseline's, and fewer than 500 of 10,000 shuffles reach it

Read again without

Faster-than-light messaging and Full-dive, at every cutoff; this line never changes the verdict

Anchor candidates

large language model, graphene, smartphone, cloud computing, social media, deep learning, CRISPR

Mature control

smoke detector

Before the pre-registration was committed, a feasibility read computed some pieces as of earlier years: for wearable computing, citation reach at 2010, 2012 and 2014, which of its barrier documents survive each cut, and growth against the live boom anchor; for every chain, whether its root crosses a tenth of its peak in the window's first year. No past score was computed before the commit. The pair counts, 21, 10 and 4, follow from the arrival years alone and were known before it.

Found after the run: the label for debris shielding, first flown micrometeoroid shield in 1973, is contradicted by a shield flown in 1968 and 1969. The first pre-registration is hashed and stays as it was; the correction is made in the revision.

What arrived

Each cutoff

The second reading

Limitations

• Barriers were not read at any cutoff: the documents on file were chosen with later knowledge, so every score here is read with barriers unsampled and carries no barrier penalty.
• The products, their search terms and which stages count as measured were chosen in 2026, knowing which products had arrived.
• Counts come from today's index, which has added works to past years since; a cutoff reads the record as it stands now, cut at the year, not as it stood then.
• The 19 arrival years were typed by hand from public sources, which backtest.json carries beside each year.
• The arrival years for orbital debris, humanoid robots and quantum entanglement were first ruled overnight by the model that ran this backtest, under the author's instruction to document rather than wait; the author ruled on them on 2026-09-30, after the first result was known, and the revision changed 5 outcomes. Wearable computing's are the years that chain declared before the backtest existed.
• The revision was made with the first result known, so it is not a pre-registration of the same strength as the first; the first result, fixed before any outcome was read, leads this page.
• The anchor candidates were named in 2026; the rule chose cloud computing at 2010, 2013 and 2016.
• The measure rests on 21 pairs at 2010, 10 pairs at 2013 and 4 pairs at 2016.
• Hearables had no work carrying references published by 2010 and 2013, so its citation share scored as no edge measured.
• Precedence was undetermined for 13 of 13 tested products at 2010, 10 of 10 at 2013 and 7 of 7 at 2016, so those scores average the other parts.