ROBIUL HASAN

Web3 project breakdowns
Airdrops • NFTs • creator programs

What’s live vs what’s just a campaign.
NFA. DYOR.

Collab:

[ROBIUL HASAN @Robiul70177]

someone asked me tonight how many projects i’m “in.”

i had to think. following a discord and filling a form isn’t really being in anything. i can name two where i’ve actually used the product this month. the rest are bookmarks.

that number used to embarrass me. now it’s the cleaner answer. if i can’t show a screen, i’m not in it, i’m on a list.

[ROBIUL HASAN @Robiul70177]

A robot dataset is usually a cost you pay once.

@axisrobotics is trying to make the next batch cheaper than the last, because the next batch is chosen by the last policy.

Collect demonstrations. Train. Put the policy back on the task. It fails in specific ways — slip, tilt, short stop. A human closes those gaps. Those corrections train the next policy. The following collection targets what that policy still cannot do.

Random success clips do not do this. They repeat the easy part.

The receipt is still on-chain. A run becomes a Trajectory when it is verified and signed on Base. Referrals follow the same rule: a click is nothing, a signed task is the unit. Five qualifying Trajectories move the multiplier to 1.05x, ten to 1.1x, counted when the epoch is tallied.

Data that only shows the win stays a library. Data that returns to the failure becomes an engine.

https://t.co/65a19So4g8

[ROBIUL HASAN @Robiul70177]

Language models read the public web. A robot arm cannot.

@axisrobotics collects the trace the web does not have: how a gripper closed, what the scene did, whether the run succeeded, and what a human did after it failed. That package is a Trajectory. It is recorded from a browser session, replayed for consistency, scored, then signed on Base. Unsigned, it is not in the epoch.

Pre-training fills demonstration slots. Training fits a policy. Post-training puts that policy back in the browser so people correct it. The next data is pulled from those corrections, not from another random success clip.

Physical AI labs can buy more video. Video does not include the recovery, the control signal, or a provenance ID.

The engine is the loop, not the clip.

https://t.co/65a19So4g8

[ROBIUL HASAN @Robiul70177]

@ecille2468 You too enjoy this beautiful day brother ❤

[ROBIUL HASAN @Robiul70177]

gm ct 🌞

saturday, coffee, and i’m not opening five quest tabs before i’ve even read what they want.

if something’s actually live i’ll look at it. if it’s another form from tuesday, it can sit.

hope the morning’s quiet on your side too.

[ROBIUL HASAN @Robiul70177]

saturday morning and the group chats are already full of “don’t miss this.”

most of them are the same link from last week with a new emoji. i used to open all of them before coffee. now i ask one thing first: did anything on the site change since tuesday?

if the answer is no, it can wait till monday. weekends are where i used to burn the most time on forms that were never going to pay me back. trying not to do that today.

[ROBIUL HASAN @Robiul70177]

Two ways to touch the same arm. They are not the same dataset.

On @axisrobotics, pre-training is you from the first frame. Teleoperate. Finish the task. One run, one demonstration. That teaches a policy what the job looks like when a human already knows the motion.

Post-training is the other chair. The policy drives. You watch. The gripper slips, the cup tilts, the arm stops short. You take over. The recovery is the record. Less intervention on a finished run scores higher. That is on-policy data: what to do when the model is already wrong.

Most robot sets keep the clean demo and drop the miss. Axis keeps both, then feeds the misses back so the next collection targets the last policy’s weak spots.

A Trajectory still has to be signed on Base or it is not in the epoch.

Pre-training: here is the task.
Post-training: here is the repair.

https://t.co/65a19So4g8

[ROBIUL HASAN @Robiul70177]

opened a project site tonight and the docs were three pages of the same sentence.

community first. revolutionary. built for the long term. no contract, no date, no picture of the thing working.

that used to be enough for me to write a thread. it isn’t anymore. if i can’t point at a screen and say what it does, the post turns into their slogan with my name on it.

so this one stays in drafts. if they ship something i can click, i’ll write it then. until that, silence is the more honest post.

[ROBIUL HASAN @Robiul70177]

A demo video is a claim. A signed Trajectory is a receipt.

On @axisrobotics a run is not in the epoch until you sign it on Base. The client uploads the package: task metadata, scene, full control trace. Replay checks physical consistency. Bad submissions never reach training. A quality score is algorithmic — rules plus a motion check. Post-training also scores how little takeover a finished run needed.

Then the portfolio. You sign. A Data ID lands on Base, bound to the task and your contributor ID. That signature is what admits the run to points. Unsigned means zero, even if the arm moved.

Referrals work the same way. A click is not a Trajectory. A signup is not a Trajectory. It counts when the person you sent finishes the task, passes verification, and signs. Five qualifying runs move the multiplier to 1.05x. Ten to 1.1x. Applied when the epoch is tallied, not live in the tab.

The data engine only compounds on records a chain can point to.

https://t.co/65a19So4g8

[ROBIUL HASAN @Robiul70177]

An Axis task is not a quest you tick.

It is a data bounty with a life.

@axisrobotics runs every task along one path: Pre-training, Training, Post-training.

Pre-training: you teleoperate the arm from scratch. Each accepted run is one demonstration.
Training: slots are full, a policy is fitting, the task is visible and closed to new runs.
Post-training: the policy attempts the task. You watch. You take over when it slips. That correction is on-policy data for the next fit.

A task with every stage done shows as Ended. The two collection stages are the product: a demonstration set, and a correction set. Most pipelines keep the first and throw the second away.

Scoring is algorithmic. Replay checks physical consistency. Post-training score also reflects how little intervention a finished run needed. From the portfolio you sign on Base. Unsigned runs earn nothing for the epoch.

The loop only compounds if the correction is kept.

https://t.co/65a19So4g8

[ROBIUL HASAN @Robiul70177]

@rarelistio Now joined but not understand the Leaderboard system

[ROBIUL HASAN @Robiul70177]

friday is when i notice which accounts only exist for the reply.

they’re on every gm. they never have a take you could send someone. if you mute the greetings, the profile goes quiet.

i did that to my own timeline last month and it was uncomfortable. a lot of days were just me saying i showed up.

so today’s post is this, and then one reply under someone who actually said something. not “gm ser.” a sentence they could argue with.

small fix. it’s the one that changes what a stranger sees if they open the profile from a collab dm.

[ROBIUL HASAN @Robiul70177]

Robots do not learn movement by reading the web.

They learn it from trajectories: a body, a scene, a control trace, a result.

@axisrobotics is built as that data engine. Pre-training is a human driving the arm from scratch. Post-training is the policy driving, and a person stepping in when the gripper slips, the cup tilts, or the arm stops a centimeter short. The recovery is the record. Unsigned runs do not enter the epoch. A Trajectory counts when it is verified and signed on-chain.

That is the loop: collect, train, fail, correct, collect again. The next batch is chosen from the last policy’s weak spots, not from a random demo reel.

Physical AI will not be solved by prettier success videos.
It will be solved by keeping the moment after success stops being guaranteed.

Join and sign a run: https://t.co/65a19So4g8

[ROBIUL HASAN @Robiul70177]

Good morning, everyone 🌞

Happy Friday and Jumma Mubarak, friends! ❤️

Wishing you all a beautiful Friday filled with happiness, peace, and good moments. May Allah bless you and your family. Have a lovely day, everyone! 🤲

Have a wonderful day everyone

Most casino NFTs are a picture you cannot do anything with after the mint.

MintABear is supposed to be a seat in the @PlayOnMint file.

4,444. Free mint. Robinhood. A portion distributed through wagering on https://t.co/dxzzx6c9BH, not only a form. Burn $MNTD to take one from Level 1 to 5. Higher level is framed as more royalty weight and a larger Status boost. After TGE, one Bear can link to MINT Status and change eligible reward rates. Extra Bears add raffle entries.

Royalties are described as ETH and $MNTD by value. Mystery boxes as NFT and token baskets. That is a design, not a paycheck I am quoting.

The play still has to happen first.
XP still has to write.
$MNTD still has to exist as the loyalty rail — 1B, Base, 30% community side.

I want the Bear because it is tied to the book. Not because a JPEG has a floor story.

[ROBIUL HASAN @Robiul70177]

@Laff38 💯 Is there any doubt?

[ROBIUL HASAN @Robiul70177]

weird thing about whitelist season.

everyone posts the application. almost nobody posts the checker a week later.

the first post is easy. you did a thing, there’s a card, the timeline likes cards. the second post is the one that actually tells people how the process went. selected, still waiting, or the form closed and nothing moved.

i applied to one this week. sealed, in review. that’s the whole update. if the list drops and i’m not on it, i’ll say that too. a timeline that only shows the apply button is just an ad for the form.

checking back is the boring part. it’s also the only part with an answer.

[ROBIUL HASAN @Robiul70177]

been thinking about why small accounts get stuck on the same $50 posts.

it’s usually not the follower count. projects already know a 9k account won’t move a launch by itself. what they can’t tell is whether you can explain their thing without the brief in front of you.

most of the posts they get back are the same shape. greeting, three features, a tag, a “dyor.” you could swap the project name and nobody would notice. that post fills a slot. it doesn’t give them anything to send a founder.

the posts that get a second message are slower. you open the site. you say what actually loads. you say what is still a form. you leave the price talk for after, because a number with no sample is just a guess.

i’m not great at this yet. half my timeline is still the easy version. the difference is i can see it now, so the next one doesn’t have to be.

if you’re in the same spot, write one post this week where the project name is the last thing you add. if the draft still makes sense, it’s yours. if it falls apart, it was a template.

[ROBIUL HASAN @Robiul70177]

@Hasan_NFTOX Use your head 🧠

[ROBIUL HASAN @Robiul70177]

@foxifytrade Hope i win Best of luck all participate

[ROBIUL HASAN @Robiul70177]

got another “what’s your rate” dm today with no project name.

used to reply with a price. now i just ask for the link.

half of them vanish, which is useful. the ones that send a site are usually the only ones worth opening.