X.KOJIRA.IO
Timeline

elsat

[email protected]
npub1zafcms4xya5ap9zr7xxr0jlrtrattwlesytn2s42030lzu0dwlzqpd26k5
Janitor
Grateful for the devs who:

-told me to fork off
-responded with silence
-responded with a thank you
-gave me feedback

I evaluate the above to figure out how to be less sometimes maybe shit.

https://video.nostr.build/5f67b89e07abdf3245db15037735cc5…
I don’t know that we can compare wireguard to nostrability.

My best guess is sherlock was pulling and scanning thousands of events from live relays, and this raised some ML flag.

No response yet from gh support.
Христос воскресе!
This time I got a support ticket number. Will share updates here, if any
Submitted a ticket again
Nothing. Contacted “github support” the other day. No response.
npub1g5qtwz2nh9q0m… repos were disappeared from github the past 24 hours. No email from github, no idea what happened. Schemata, trackers, sherlock and everything else.

I will try to restore when I return to my machine.

Maybe this is the signal for nostr git for those not there yet.

Consider backing up your github at the least.

npub1uac67zc9er54l… npub1qe3e5wrvnsgpg… npub180cvv07tjdrrg… npub1xtscya34g58tk…
Thank you ser. Lmk how I can improve it. This pulls events from relays and validates against schemata.

I have a shortlist of sherlock issues specified. Need to do some more verification these are real before tagging devs.



Cc npub1r0rs5q2gk0e3d… npub1uac67zc9er54l…

https://image.nostr.build/86c8831d9430638d3a15b2e9f0365e0…
Will aim to simplify the next version fixing content schemas

cc npub1uac67zc9er54l…

nevent1qqstkdun97y…
Nip-66 seeing some massive wins lately.

All hail ser npub1uac67zc9er54l…

nevent1qqs95hqgslp…
Asking questions and price increases ist Verboten
I have decent benchmarks on major outbox implementations now in github.com/nostrability/outbox.

I have concrete suggestions on where to start for apps without outbox, and concrete improvements in increased coverage in thompson learning, and reduced load times by way of nip-66.

I have submitted PRs for consideration to rust-nostr (closed), coracle/welshman (open), NDK (open), applesauce (open) SDKs.

Lmk where to submit outbox PRs next. Note that some devs like PRs, and others dont.

cc npub1uac67zc9er54l…
I dont know all the latest details, but basically we have at least a dozen things for this gent here, as long as he is motivated enough to sort out the rough edges.
I have benchmark updates, and a few real world practitioner updates landing today

https://github.com/nostrability/outbox/pull/20
Google, meta, instagram, sloptok, amazon have spent billions of US war shekels and millions of engineering man hours on reducing load time, so that their ADD riddled consumers dont lose patience before the page loads.

Ser npub1uac67zc9er54l… gifted nostr with a spec that helps nostr devs (e.g. implementing outbox) in nip-66 that is benchmarked to reduce load times by about 40%.

Adding nip-66 is tens of lines of code.

If you dont trust third party nip-66 providers, you can roll your own.

By checking which relays are alive/offline/dead before checking relays you dont add complexity - you reduce and manage complexity in your app, as you make app behavior more certain (order of magnitude 1/2 connections to outbox are wasted without nip-66).

Performance matters.

Accept this, and improve the experience for your customers.

nevent1qqsv357aakh…
Can I see the repo?

Nostr.watch runs off nip-66 events.
Its just that they aren't published to many relays, only:
Relaypag.es
Relay.Nostr.watch
Monitorlizard.nostr1.com
There's also undocumented events that can be used for uptime, kind 1066

cc npub1uac67zc9er54l…
More outbox lessons learned for me, maybe useful for #devstr

As far as I know
1. Learned that no nostr app uses nip-66 to check if a relay is alive before firing off requests
2. According to my benchmarks, by implementing nip-66, the relay success rate increases from 30% -> ~~75-85%, meaning fewer connections are wasted on dead relays.
3. Nip-66 is therefore an efficiency improvement, and is not necessarily a coverage guarantee
4. I have not measured time to outbox results yet.
5. No nostr app spot checks what outbox relays return against an independent relay to detect gaps. Nip-66 liveness checks return relay health, but do not verify per-author delivery.
6. Deterministic algos like greedy set coverage cannot be set up with a learning mechanism
7. Stochastic algos like coracle’s are able to implement learning, and benefit significantly - coverage improves the more the app looks up a specific profile
8. The theoretical CS algo I mentioned before (MAB-UCB) algo is not practical
9. 20-44% of follows dont have a kind-10002, which hampers outbox.

From what I’ve seen so far the best starting point is coracle’s stochastic algo, with nip-66, and delivery scoring.

https://github.com/nostrability/outbox

cc npub1uac67zc9er54l… npub1jlrs53pkdfjnt… npub180cvv07tjdrrg… npub1utx00neqgqln7… npub1acg6thl5psv62… npub1ye5ptcxfyyxl5… npub1fgz3pungsr2qu… npub13v47pg9dxjq96…

nevent1qqstev5upz3…
I made the report less academic, and more actionable.

Specific to your question on how to implement the cs algo - mab - see

https://github.com/nostrability/outbox/blob/main/IMPLEMEN…

And

https://github.com/nostrability/outbox/blob/main/IMPLEMEN…

MAB complements nip-66 well - nip-66 checks liveness before asking relays, whereas MAB tracks relay health *after* asking the relay, and scores it.

Nosotros npub1cesrkrcuelkxy… looks like has a “seen” table (event created at x seen on relay y), and it is not providing feedback to relay selection for future lookups.
Thanks for taking interest. I will add methodology overview to the readme. To your questions:
1. I extracted the algorithms from nostr outbox implementations to a typescript library.
2. In the TS library we have nostr algos from the real world (e.g. from amethyst, nostrudel, nostur etc)
3. In the TS library we also have the general CS algorithms not yet found in nostr apps
4. The computer science problem of “outbox” maps to “weighted set cover problem” AKA “maximum coverage problem”. https://en.wikipedia.org/wiki/Maximum_coverage_problem
5. The “winning” algo for long term event retrieval MAB is used in real world applications in online ads, a/b testing https://web.stanford.edu/class/cme241/lecture_slides/Mult…

https://web.mit.edu/6.7950/www/lectures/L15-2022fa.pdf
6. I answer two questions to the above:
Given a npub (e.g. fiatjaf who has 400/whatever follows) run a ~~dozen or so algos
A) “assignment coverage” phase 1(academic/no network ) if every relay is online, and no events are deleted, how many of your follows are reached by this algo?
B) “event recall” phase 2 (real world like/pull events from relays) given realistic conditions of relays being down, events being pruned/deleted, relays gating event retrieval (via auth or otherwise) what are the numbers?
C) given phase 2 above, what happens if we triage relays first via nip-66 liveness monitor events, relays, services like nostr.watch by npub1uac67zc9er54l…
7. I have not gone into a dozen apps to measure in app the outbox results. All of this is done in deno/typescript
8. The real world results are vastly different than the academic ones.
I’m measuring more details on nip-66 - the hypotheses are 1) improved coverage, and 2) drastically reduced number of calls to relays to find the info being sought
I had a family friend who did this approach and it did not help him.

This was before I knew about seed oils. I suspect it was the seed oils, and/or other chemical-industrial waste slop.
To put the climate scam in perspective for bitcoiners - it is about 100,000,000 btc (one hundred million btc) at $100k/🌽 and counting

nevent1qqsrdufmu4y…
older →