Sure! Thrilled to answer what I can:

I am using actual point-in-time constituents from Norgate's Current & Past watchlists with the index constituency series. I'm using daily bars, signal from the close of day t-1, fill at the open of day t. When I run monthly rotations, I rank on the last trading day's close and trade at the next month's first open. Your alternative day selection is something new I hadn't considered, I will try it out and compare.

Costs are meant to simulate a IBKR Pro account, simulated to $0.005/share ($1 min) plus 1 bp slippage, equal weight, compounded. This is deliberate, since the RealTest OrderClerk tool exclusively talks to IBKR (IIRC).

For Machine-readable table, yes I think this would be great. Most fields exist already; the rule sequence, rebalance frequency and holding count have to be pulled out of the strategy files. Is there a standard format that would be helpful to most individuals? Or one that you personally have in mind?

All strategies I've posted under paid tiers have a 6 month post-publication forward record waiting period. This incubation period is meant as a personal sanity check - I'd hate to release anything that looks good in backtesting and walk-forward only to find I'd overfit or made a mistake and have 20 angry people emailing me about it.
I started publishing algorithms in July, so the earliest published ones should start opening up for sale around Dec or Jan.

On downloadable monthly return histories, agreed, and the monthly grid is already on every strategy page, so the data exists - I just have to convert it to a CSV download. The algo-catalog.com/compare-combine page does pairwise combination math from that same data in the meantime.

Q: Do you test neighboring parameter values, or only the published spec?
A: Across time, yes. Across values, no. Every declared parameter set is scored on each calendar year 2008-2019. I keep the mean-minus-one-SD winner. The value grids are small (usually two values/parameter) and I don't yet check for spikes against neighbors or publish a grid; I'll add this to the todo list.

The rejects list is mostly algorithms that produced negative CAGR, either on the 2020+ out-of-sample window or over the full history. There are about 1,900 of them as of this posting. I couldn't decide if I wanted to publish them. I keep all of them with full stats, except stats that can be derived from the core stats. Do you have an interest in seeing the rejects as well?

The catalog page does contain filters on universe, mechanism, tier, OOS and lifetime CAGR and max drawdown, with Sharpe sorting, but I believe they can be enabled simultaneously. I will look into it.
Rebalance frequency, holding count and a Sharpe filter are missing though; I'll look into adding them.

Thank you very much for spending the time on these questions. I've been meaning for a few months to poke around the internet and ask for feedback, and this is great.