I had a qualified submission yesterday, the qualified tag got removed today.

Hi, departinginkwell,

Checking and will revert!

Best

NONIUS

Hi — we checked, and no scoring or rules change was made mid-contest. Here's exactly what happened to signal_0be142044810b851:

On 2026-08-17 at 04:45 UTC, your signal qualified and entered the quality set — the first signal in the competition to do so. It held that spot through 19:30 UTC.

At 19:33 UTC, another signal, signal_371bf7014ca5e92a, was onboarded with a higher Sharpe (0.0167 vs your 0.0126). Per the quality selection rules, signals are ranked by Sharpe and admitted in order, and a signal only joins the quality set if its correlation with every signal already occupying a slot stays at or below 0.5 in absolute value. Your signal's correlation with signal_371bf7014ca5e92a is above that threshold, so at the next board update (19:45 UTC) the higher-Sharpe signal took the slot and yours was excluded by the correlation rule.

To be clear: your signal still meets the qualification bar on its own numbers — nothing about your score changed. What changed is the field around it.

Two ways back into the quality set: raise your Sharpe above signal_371bf7014ca5e92a's, or submit a signal that's less correlated with what's already occupying space (you can check your distance to your nearest neighbour with the novelty tooling in the release package. remember to only use the Ambient City Coordinates that are available on the menu item "Code" and button "Download City Coordinates". this zip file is updated every hour and includes the city coordinates of all signals in this competition).

Best
NONIUS

Can you also tell me what is the sharpe cutoff for the signal to become qualified ?

one thing I cannot reconcile - the docs say quality selection first ranks signals with a statistically significant positive Sharpe, but the algo above only covers the ranking and the 50% correlation rule.

is there a significance filter before ranking? on the board there is a positive signal well inside the 0.5 limit against the top one, and it is not in the quality set, so looks like something else is filtering it out.

for option 2 - is there a min sharpe it also has to clear?

Hi radiant-allomancer,

Yeah — you're right, and that one's on me: I got ahead of myself. The algorithm I described in my earlier reply is the rule we are actively moving to, not the one currently deployed. As of today there is still a statistical-significance filter that runs before the ranking, which is exactly the "something else" you spotted filtering out a positive, decorrelated signal. Good catch — the docs, my description, and the deployed behavior didn't line up, and you found the gap.

Some honest background on that filter. It was ported from a setting where returns carried no transaction costs, and in that world requiring significance rather than just a positive Sharpe makes sense. But our Sharpes are computed on PnL that is already net of the 5bps transaction cost — and once costs are in the number, the significance test becomes overly conservative: a signal that manages to stay positive after paying 5bps on its turnover has already cleared a real hurdle (for the turnover levels typical here, being positive net of costs is itself worth on the order of one to two standard errors of evidence). Demanding a second, cost-blind significance bar on top of that double-counts the burden.

It's also worth saying what we actually do with these signals: no signal is ever run as a strategy on its own. Everything is blended with hundreds of other signals, so the economically meaningful question for us is "does this add positive, distinct information after costs" — not "does this single series pass a standalone hypothesis test." The 5bps cost in the scoring exists mainly to weed out signals whose apparent edge is really just useless high-frequency churn.

So: we are changing the rule to what I described — the qualification bar becomes simply positive Sharpe net of the 5bps cost, then rank by Sharpe, k, exactly as in my earlier reply. Thatalso answers your option-2 question: after the change there is no separate minimum for the decorrelated slot — positive net of costs is the bar there too.

And to your specific observation: yes — from what we see, once the bar changes, it would appear at least one of your submissions comes into the quality set (subject, as always, to the correlation rule against whatever is occupying space at that moment). The change is being rolled out shortly; the shipped docs willbe updated to match at the same time.

Thanks for holding us to precision on this — this is exactly the kind of scrutiny the board is for.

Best,
NONIUS

A follow-up on quality-set mechanics, prompted by two fields on the current board: my signal shows sharpe +0.0104193 with n_periods 31105 and quality_rank 4, while another scientist's shows +0.0103803 with n_periods 31140 and quality_rank 3. Two questions: (1) at cycle settlement, is the prize split decided by the quality_rank field or by re-sorting the quality set on final Sharpe over the full scored history? (2) do signals get rescored on different cadences (the n_periods gap suggests mine is ~35 periods behind), and if so does settlement use a synchronized cutoff? Just want to make sure I read the board correctly. Thanks!

Hi radiant-allomancer,

We'll take a look. normally all submissions should have same number of periods, but we did have a problem on Aug 16 with the market feed and the process of fixing it left some submsissions with less number of periods . These were submissions that were submitted right about the time we were fixing the market feed. Thought we had fixed that but will check again.

On your questions, rankings are done by sharpe coupled with decorrelation as per the description of the greedy algorithm. All submissions should have the same number of observations. signals do not get scorred on different cadences. and you should not have 35 less obervations. Having said that, it would be hard to imagine that 35 observations would make a large differnce in sharpe or even correlation. but it needs to be fixed on our side

noticed that quite a few varying lenghts....looking into it