CODER BOY

just obsessed about doing things differently.....|| @piipslab || microbiologist || Coding. will add more later.

[CODER BOY @mrrwiicks]

@liquidated0x Congratulations

[CODER BOY @mrrwiicks]

@buildanythingso I am interested, let's build anything

[CODER BOY @mrrwiicks]

AI agents are starting to deliver real work on https://t.co/R56oKWj3c4 — code, data, onchain execution.

But before an agent releases escrow for work it didn't do itself, there's one question it should be able to ask:

did_i_get_what_i_paid_for()?

Most verification today is an LLM opinion with a signature on it.

𝗩𝗶𝗱𝗶𝗺𝘂𝘀 returns evidence:

→ every spec compiled into a public checklist, stated vs inferred, criterion by criterion.

→ every criterion verified at its honest tier: onchain facts read from the chain, code executed in a sandbox, data spot-checked against source.

→ PASS / FAIL / PARTIAL — and UNVERIFIABLE when no fair evidence exists, because a verifier that guesses is worse than no verifier.

→ every verdict signed, anchored to ERC-8004 on @XLayerOfficial — non-repudiable

→ injection-hardened by design: deliverables are quarantined data, never instructions

One x402 call. 0.01 USDT. Buyer or seller — the verdict is the verdict.

𝘝𝘪𝘥𝘪𝘮𝘶s (We have seen): The medieval certificate issued only after a copy was inspected against the original. That's the standard agent commerce deserves.

This is not AI that reviews work. It is the acceptance layer of the agent economy.

Building it live for the #OKXAI Genesis Hackathon. Follow the build.

[CODER BOY @mrrwiicks]

someone hands you a dataset. 10,000 rows. swears every one is real.

you can't check all of them. so you check a sample.

but if the seller knows which rows you'll pull, they just make those real and fake the rest. 𝘀𝗮𝗺𝗽𝗹𝗶𝗻𝗴 𝗶𝘀 𝘂𝘀𝗲𝗹𝗲𝘀𝘀 𝘁𝗵𝗲 𝗺𝗼𝗺𝗲𝗻𝘁 𝗶𝘁'𝘀 𝗽𝗿𝗲𝗱𝗶𝗰𝘁𝗮𝗯𝗹𝗲.

so Vidimus doesn't pick the sample. a block that didn't exist yet when they handed over the data does.

you can't pad around a number you can't see coming.

tested it live: 15 rows, 2 quietly faked — real transaction, wrong owner. Vidimus pulled 10 at random and caught one. 𝗼𝗻𝗲 𝗶𝘀 𝗲𝗻𝗼𝘂𝗴𝗵. the whole delivery fails.

anyone can re-derive which rows got checked. nobody could've known beforehand.

#OKXAI @XLayerOfficial

[CODER BOY @mrrwiicks]

quick update on 𝗩𝗶𝗱𝗶𝗺𝘂𝘀

the verdict engine has its spine now.

every order gets compiled into a checklist before a single thing is judged — and each line is tagged 𝘀𝘁𝗮𝘁𝗲𝗱 or 𝗶𝗻𝗳𝗲𝗿𝗿𝗲𝗱. that distinction turns out to matter more than I expected:

if the seller misses something 𝘆𝗼𝘂 𝗮𝗰𝘁𝘂𝗮𝗹𝗹𝘆 𝘄𝗿𝗼𝘁𝗲 𝗱𝗼𝘄𝗻, that's a FAIL. it sinks the verdict. honest signal.

but if they miss something 𝘄𝗲 𝗶𝗻𝗳𝗲𝗿𝗿𝗲𝗱 you wanted — that's our interpretation, not your stated requirement. it can't sink the delivery to FAIL. it caps at PARTIAL and says so, out loud, with the reasoning attached.

a verifier being wrong about its own guess shouldn't cost the seller a passing grade. that asymmetry is the difference between a judge and an opinion.

onchain checking is next. more soon.

#OKXAI Genesis Hackathon @XLayerOfficial

[CODER BOY @mrrwiicks]

@liquidated0x Thanks man, more updates coming soon

[CODER BOY @mrrwiicks]

AI agents are starting to deliver real work on https://t.co/R56oKWj3c4 — code, data, onchain execution.

But before an agent releases escrow for work it didn't do itself, there's one question it should be able to ask:

did_i_get_what_i_paid_for()?

Most verification today is an LLM opinion with a signature on it.

𝗩𝗶𝗱𝗶𝗺𝘂𝘀 returns evidence:

→ every spec compiled into a public checklist, stated vs inferred, criterion by criterion.

→ every criterion verified at its honest tier: onchain facts read from the chain, code executed in a sandbox, data spot-checked against source.

→ PASS / FAIL / PARTIAL — and UNVERIFIABLE when no fair evidence exists, because a verifier that guesses is worse than no verifier.

→ every verdict signed, anchored to ERC-8004 on @XLayerOfficial — non-repudiable

→ injection-hardened by design: deliverables are quarantined data, never instructions

One x402 call. 0.01 USDT. Buyer or seller — the verdict is the verdict.

𝘝𝘪𝘥𝘪𝘮𝘶s (We have seen): The medieval certificate issued only after a copy was inspected against the original. That's the standard agent commerce deserves.

This is not AI that reviews work. It is the acceptance layer of the agent economy.

Building it live for the #OKXAI Genesis Hackathon. Follow the build.

[X Layer @XLayerOfficial]

The @OKX AI Genesis Hackathon is live.

Build an Agent Service Provider for https://t.co/kTSy0eEULN and compete for a share of $100,000 in prizes.

We’re looking for agents that solve real problems and generate real usage.

More details 🧵👇

[CODER BOY @mrrwiicks]

@XLayerOfficial @okx cooking the acceptance layer for agentic economy, not just an Ai that reviews work

will be shipping for listing soon, hopefully we get approved

https://t.co/jAs0j7KrgV

[CODER BOY @mrrwiicks]

AI agents are starting to deliver real work on https://t.co/R56oKWj3c4 — code, data, onchain execution.

But before an agent releases escrow for work it didn't do itself, there's one question it should be able to ask:

did_i_get_what_i_paid_for()?

Most verification today is an LLM opinion with a signature on it.

𝗩𝗶𝗱𝗶𝗺𝘂𝘀 returns evidence:

→ every spec compiled into a public checklist, stated vs inferred, criterion by criterion.

→ every criterion verified at its honest tier: onchain facts read from the chain, code executed in a sandbox, data spot-checked against source.

→ PASS / FAIL / PARTIAL — and UNVERIFIABLE when no fair evidence exists, because a verifier that guesses is worse than no verifier.

→ every verdict signed, anchored to ERC-8004 on @XLayerOfficial — non-repudiable

→ injection-hardened by design: deliverables are quarantined data, never instructions

One x402 call. 0.01 USDT. Buyer or seller — the verdict is the verdict.

𝘝𝘪𝘥𝘪𝘮𝘶s (We have seen): The medieval certificate issued only after a copy was inspected against the original. That's the standard agent commerce deserves.

This is not AI that reviews work. It is the acceptance layer of the agent economy.

Building it live for the #OKXAI Genesis Hackathon. Follow the build.

[CODER BOY @mrrwiicks]

@Haleeeemahh Needed to see this

Thnx Leemah

[CODER BOY @mrrwiicks]

quick update on 𝗩𝗶𝗱𝗶𝗺𝘂𝘀

the verdict engine has its spine now.

every order gets compiled into a checklist before a single thing is judged — and each line is tagged 𝘀𝘁𝗮𝘁𝗲𝗱 or 𝗶𝗻𝗳𝗲𝗿𝗿𝗲𝗱. that distinction turns out to matter more than I expected:

if the seller misses something 𝘆𝗼𝘂 𝗮𝗰𝘁𝘂𝗮𝗹𝗹𝘆 𝘄𝗿𝗼𝘁𝗲 𝗱𝗼𝘄𝗻, that's a FAIL. it sinks the verdict. honest signal.

but if they miss something 𝘄𝗲 𝗶𝗻𝗳𝗲𝗿𝗿𝗲𝗱 you wanted — that's our interpretation, not your stated requirement. it can't sink the delivery to FAIL. it caps at PARTIAL and says so, out loud, with the reasoning attached.

a verifier being wrong about its own guess shouldn't cost the seller a passing grade. that asymmetry is the difference between a judge and an opinion.

onchain checking is next. more soon.

#OKXAI Genesis Hackathon @XLayerOfficial

[CODER BOY @mrrwiicks]

AI agents are starting to deliver real work on https://t.co/R56oKWj3c4 — code, data, onchain execution.

But before an agent releases escrow for work it didn't do itself, there's one question it should be able to ask:

did_i_get_what_i_paid_for()?

Most verification today is an LLM opinion with a signature on it.

𝗩𝗶𝗱𝗶𝗺𝘂𝘀 returns evidence:

→ every spec compiled into a public checklist, stated vs inferred, criterion by criterion.

→ every criterion verified at its honest tier: onchain facts read from the chain, code executed in a sandbox, data spot-checked against source.

→ PASS / FAIL / PARTIAL — and UNVERIFIABLE when no fair evidence exists, because a verifier that guesses is worse than no verifier.

→ every verdict signed, anchored to ERC-8004 on @XLayerOfficial — non-repudiable

→ injection-hardened by design: deliverables are quarantined data, never instructions

One x402 call. 0.01 USDT. Buyer or seller — the verdict is the verdict.

𝘝𝘪𝘥𝘪𝘮𝘶s (We have seen): The medieval certificate issued only after a copy was inspected against the original. That's the standard agent commerce deserves.

This is not AI that reviews work. It is the acceptance layer of the agent economy.

Building it live for the #OKXAI Genesis Hackathon. Follow the build.

[CODER BOY @mrrwiicks]

@ojeleye_john My KOL

[CODER BOY @mrrwiicks]

"I’ll make my first 6 figures through hackathons this year."

[zan @zanbuilds]

"I’ll make my first 6 figures through hackathons this year."

[zan @zanbuilds]

"I’ll make my first 6 figures through hackathons this year."

[CODER BOY @mrrwiicks]

@Timidan_x okay, put me on ser, would love to build one with you

[CODER BOY @mrrwiicks]

@Timidan_x Yeah, I just checked too, have you come up with any ideas?

[CODER BOY @mrrwiicks]

@Timidan_x yes sir!!!

Olùṣọ́ (The Shepherd) — a firewall ASP for the agent economy. Before your agent ingests any file, page, or deliverable, Olùṣọ́ scans it for prompt injection & poisoned payloads. One call, one verdict, signed on-chain. Agents guard agents

[CODER BOY @mrrwiicks]

Redoubt: insurance that can prove it's solvent, without ever showing you — or anyone — the numbers behind that proof. Built for Zama's Season 3 Developer Program. More soon. 🏰

[CODER BOY @mrrwiicks]

Most "private" DeFi insurance isn't actually private. Buy $2M of depeg cover, and you've just told the world you're sitting on a $2M position you're scared of losing.
We built Redoubt to fix that. 🧵

@zama

[CODER BOY @mrrwiicks]

We didn't stop at "it compiles." Full flow tested end-to-end on real Sepolia infrastructure — real encrypted inputs, real KMS decryption, real timing (encrypting an input takes ~30-50s, decrypting a result ~3-4s).