Corrections

Until 24 September the League divided each account’s fee by the middle of its trade-size band instead of by the real trade size. Below 0.1 SOL the “trade” was often a 0.002 SOL account-rent payment, and some rows rested on a single trade, so the table showed fees of 56 to 77 percent of trade size, which no bot charges. Each rate is now the median fee over the median real trade size, from trades of 0.1 SOL and up with at least five trades per rate. The table’s 41 rates now run from 0.10 to 1.02 percent, in line with the two fee levels our census found, about 1 percent and about 0.47 percent.

Root cause: dividing by an assumed trade size instead of the measured one.

After we corrected Photon’s fee on 9 September, the League said Photon and GMGN tie at every size, yet its chart, its table and one sentence still showed the old crossover at 0.35 SOL. Its size ladder named GMGN cheapest at every size although Photon costs exactly the same, and GMGN is one of the bots whose referral links pay us. On the Calls page, our 21 September fix broke ties alphabetically, as it said, which in practice always picked GMGN. The League also said GMGN fills 75 basis points better without saying that the figure comes from a study that ended on 19 August. Now the League shows one line, “tie at every size”; ties read as ties, and where a page must pick one bot, tied bots take turns. The execution figure carries its end date and our referral disclosure.

Root cause: a correction applied to one sentence, not to the chart, the table and the data behind them.

The Graduations page counted every outcome row in our scanner file, 59,875 at the time, including repeats and tokens outside our checked index, while the homepage counted the checked index, 57,639. Both pages now count the same checked index, one row per token.

Root cause: two pages counting the same thing from different files.

The trade-size box on the Calls page still priced Photon at 0.0035 SOL a trade, the flat figure we retracted on 9 September. Above 0.35 SOL it named Photon the cheapest bot, and in automatic mode its Trade buttons opened Photon through our referral link. At its real rate of 1.00 percent per leg, Photon ties GMGN at every size. The box now uses the published rate, says when bots tie, and breaks ties alphabetically.

Root cause: a corrected number copied into a second page was fixed in only one.

We published that our scanners had run “109 days without missing one” since 1 June. They had not. Both the homepage and the data page computed the number as the span between our first record and our last — arithmetic that cannot tell continuous recording apart from a gap. There were four outages, of 22, 11, 11 and 10 days. The true record is 44 consecutive days since 8 August, plus 14 scattered earlier days: 58 days of data across a 111-day span. The homepage now carries a chart of every day, including the 54 we missed.

Root cause: a span mistaken for continuity, in two independent generators.

Sixteen accounts we had labelled PHOTON are a different service. Different programs, and zero payer overlap with Photon’s real collector in a sampled comparison. They are now labelled UNNAMED_PFN — a name prefix, and nothing more.

Root cause: attribution by name rather than by evidence.

Our +21 bps Photon execution row rested on the same pfn prefix we had just disproved. The 278 trades behind it were probably not Photon trades, so the number measured something we cannot name. We pulled the row and the total-cost column rather than leave a figure we no longer believed.

Root cause: attribution by name rather than by evidence.

We published a flat 0.0035 SOL per leg. We then ran four controlled trades through Photon from a phone and read them on chain. The real charge is 1.00 percent of trade size per leg, confirmed across 196 priced trades from 54 wallets and six size bands. The old figure overstated Photon on small trades by about three and a half times and understated it on large ones by about three.

Root cause: attribution by name rather than by evidence.