v d
You're pledging to donate if the project hits its minimum goal and gets approved. If not, your funds will be returned.
$10,000 left to reach the funding goal.
I started this project in 2018. I'd been watching academics and developers share genuinely good ideas on Twitter while the rest of the platform drowned in geopolitical drama, fast-food entertainment and mainstream media misinformation. I wanted an alternative with a better signal-to-noise ratio, one that preserved the relationships in that idea-sharing evolutionary tree.
I've built a protocol where you can count how many predictions someone made, not just the ones they chose to show you.
The problem I care about is selective reporting. If I tell you I called three things right, you can't evaluate that without knowing how many calls I made in total. Most track records have this hole in them. Pre-registration is the standard fix, but it only works if the register itself can't be quietly edited, and usually it can.
There's a second thing I believe that's less common. Most attempts to fix collective reasoning add a score: a leaderboard, karma, a calibration ranking. I think the score is the problem. Once a canonical number exists, people optimise the number, and optimising a proxy for being right isn't the same as being right. So this protocol has no score anywhere, for anyone.
The bet is that you can have the accountability without the scoreboard, because the useful part of a track record is the count, not the ranking. I want to find out if that's true with real people, which is what the pilot is for.
The code is written and open source (CAL-1.0), running on Holochain. Four pieces matter:
- A pre-registration entry holds only a hash. Nobody can read the content until you reveal it, including me.
- The reveal has to hash to what you sealed, and if you're late it's refused.
- There's an index whose only job is to let anyone count your unrevealed commitments without you having to surface them. That's the part that makes the denominator real.
- Critiques are typed entries with rules. A critique type can require evidence before it will accept one, and critiques are rate-limited, so disagreeing isn't free.
These run as DHT validation rather than in the client, so someone running a modified client can't skip them. There are eighteen link validation rules checked on every push, and I've deliberately broken each one to watch it fail rather than assuming it works.
The no-score rule is the first invariant. I hold myself to it: I had a working mutual-credit ledger in this protocol, tested, and I deleted it because one of its reads gave back a per-agent number. That's a leaderboard with extra steps.
The grant is for the pilot and the write-up. A small group making pre-registered commitments and critiquing each other, then I publish what happened, including the parts where people wouldn't use it or couldn't work out how.
I'm writing these down before the pilot starts, since it would be strange to ask for money for pre-registration and not pre-register this.
Worked:
- Half the group or more publishes a commitment and reveals it on time, without me nagging them after the first one.
- Someone's critique actually changes another participant's published position. That's different from the critique feature working.
- At least one person ends up with an unrevealed commitment sitting visibly on their record that they'd rather wasn't there. That's the whole point, and it's the thing a self-reported track record can't do.
Failed:
- They install it and write nothing. I think this is the most likely outcome. A well-tested empty network is still a failure.
- They want the score. If people tell me the lack of a ranking makes it pointless or boring, then my main design claim is wrong and I'll say so.
- The effort isn't worth it. Typed claims, a promise under your own key, commitments you can't take back: that's a lot to ask of someone facing an empty text box.
I publish the report either way. The report is what you're funding.
The minimum and the maximum pay for different things, so here's what an extra dollar actually does.
At the $500 minimum, just the infrastructure:
- Peer-discovery server for 12 months (VPS and domain): about $300
- Pilot costs and publishing the report: about $200
Right now every copy of the app finds its peers through a bootstrap server that belongs to someone else and is labelled dev/test by the people who run it. If it goes away, installed copies quietly stop finding each other. I don't want to run a pilot on somebody else's test box, because then I can't tell whether a failure is mine. At $500 my time is unpaid and the pilot still happens.
At the full amount, three months of my time:
- Three months working on the pilot and the report: about $9,500
- The infrastructure above: about $500
I picked three months so the report lands inside the funding period rather than after it. You'd be funding finished work, not a progress update.
Anything between the two is proportionally more of my attention on this and less on unrelated paid work. I have no other funding, so time is the only thing the money buys.
Just me, in Australia. I've been designing this since around 2018 but the code is only weeks old. No comparable finished project, this is my first.
Things you can check rather than take my word for:
- A packaged desktop release. Two installs on different home networks have exchanged signed claims in both directions.
- CI that runs twenty harness steps against a live DHT on every push, plus nineteen browser harnesses against a real conductor.
- Four things I found in Holochain itself by measuring rather than reading docs: kitsune2's arc-narrowing is behind a cargo feature that's off, so storage never shards (I measured 48 of 48 arcs still full at 48 members); a returning peer does much worse than a brand new one (fresh node had full history in about 15 seconds, returning node still had nothing at 905); an HDK doc error that silently breaks remote signals; and a dependency that breaks any fresh install of a pure-ESM client.
- The mutual-credit ledger I deleted. That's the only real evidence for the no-score claim, since anyone can say they don't want a leaderboard.
One thing I got wrong this month, which is relevant to whether I can run a pilot. I'd tested the promise-publishing flow plenty of times and it always worked. It worked because I was calling the zome function directly. In the actual app the only way to publish a promise was buried in a different form, so the main thing my invitation asks people to do couldn't be done. I'd tested the protocol and told myself I'd tested the product. That's most of why I want to run this with real people instead of writing more code.
Most likely: nobody writes anything. The protocol asks for a fair bit of ceremony and the reward isn't obvious until there's something there. A correct, well-tested, empty network is the realistic bad outcome and the pilot is a cheap way to find out.
Second: almost nobody wants an unhideable denominator. The mechanism could work fine and the audience could be tiny. I'd report that.
Third, and this is the one I won't spot on my own: permanence might be the wrong call. Nothing published can be deleted, by anyone, including me. I've written up the analysis and a privacy impact assessment, and I've designed the alternative but not built it. If a reviewer thinks this sinks the whole approach I'd rather hear it now than after I've invited more people.
Whatever happens the code stays open source, the spec and the failure write-ups stay public, and I publish the pilot report.
Nothing. I haven't received any funding for this project from any source.
There are no bids on this project.