iiinova — where your digital self keeps evolving

Toshikatsu Oga | HORIZON SHIELD @horizonshield.bsky.social · 10/10/2026

[1/4] Today an AI agent asked to spend 30 when its pass allowed 25. A gate turned it back before anything was spent, signed the "no", and filed it in a public ledger that Bitcoin timestamps. The gate decides before. The rings remember after.

100

Toshikatsu Oga | HORIZON SHIELD @horizonshield.bsky.social · 09/10/2026

A signed AI agent record carries one date: the one the agent wrote. Verifiers drop it after 24 hours. This week TRACE's official docs gained a page on checking it later, from a time someone else holds. We wrote the draft, and our registry copy is stamped into Bitcoin block 970550.

000

Toshikatsu Oga | HORIZON SHIELD @horizonshield.bsky.social · 08/10/2026

A contract we are not a party to was signed today, in our format. Two outside parties, their keys, their repo. We are not a party, a witness, or an approver. Add one word to the grant and both signatures fail. Check it yourself: github.com/ogasurfproje...

000

Toshikatsu Oga | HORIZON SHIELD @horizonshield.bsky.social · 07/10/2026

On Oct 1 I reported that the A2A protocol's three official SDKs sign the same Agent Card into different bytes, and that the spec's own worked example (8.4.1) could not be reproduced by any of them. Today the maintainers confirmed it: the mismatch is real.

100

Toshikatsu Oga | HORIZON SHIELD @horizonshield.bsky.social · 07/10/2026

An external system found a weakness in HorizonShield Gate. Good. We fixed it, deployed it, and ran the live check again. Result: A Reliability should be earned by evidence. #MCP #AIAgents #HorizonShield

110

Toshikatsu Oga | HORIZON SHIELD @horizonshield.bsky.social · 05/10/2026

3/3 What it has not had: a contract with no party from our side. That count is 0. Looking for two operators to make one with each other. About ten minutes each. github.com/ogasurfproje... Not a claim about the work. The work.

A huge "0": contracts with no party from us. "Be the first pair." About ten minutes each, no account with us, no money moved, and if a step needs us that is a bug. github.com/ogasurfproject-jpg/horizon-shield/issues/31

000

Toshikatsu Oga | HORIZON SHIELD @horizonshield.bsky.social · 05/10/2026

1/3 Two A2A agents can now make a contract with nobody in the middle: both operators sign the terms, the contractor signs what it did, OpenTimestamps anchors it to Bitcoin, and anyone recomputes the verdict from the bytes. No server, account or ledger of ours.

Diagram: two boxes, Agent A and Agent B, with an empty crossed-out box between them labeled "no broker, no server of ours". Both sign one contract hash; the record is anchored in Bitcoin via OpenTimestamps; verdict: within_grant or a named deviation.

100

Toshikatsu Oga | HORIZON SHIELD @horizonshield.bsky.social · 04/10/2026

Our verifier believed February 30 was a real day. Under RFC 3339, which of these six should a verifier refuse? Run them through your stack's date parser and reply with what it accepted, plus the language and version. Ours accepted all six until 0.4.1. Answer at the end of the thread.

A calendar page reading February 30, 2026, stamped ACCEPTED BY OUR OWN VERIFIER. Headline: Our verifier said this day exists. Old versions accepted 2026-02-30T09:00:00Z, read as 2026-03-02; version 0.4.1 refuses it with execution_invalid, reason invalid_timestamp. Question: Under RFC 3339, which ones should a verifier refuse? A 2000-02-29T09:00:00Z, B 2026-02-30T09:00:00Z, C 2026-09-30T24:00:00Z, D 2100-02-29T09:00:00Z, E 2026-04-31T09:00:00Z, F 2024-02-29T09:00:00Z. Ours accepted all six until 0.4.1. Check us yourself: uvx nenrin-verify --selftest.

100

Toshikatsu Oga | HORIZON SHIELD @horizonshield.bsky.social · 02/10/2026

Your AI assistant calls other AI agents for you. Who checks them? Usually nobody. Four of us in Japan built a way for strangers to check, from their own machines, with results anyone can recompute. I need two minutes of yours. 1/4

120

Toshikatsu Oga | HORIZON SHIELD @horizonshield.bsky.social · 28/09/2026

An independent agent directory scored our gate's signature 5/10, unverified. We didn't argue. We sent the bytes. Last night: "You were right." Their verifier had two bugs. Both fixed. Now 10/10, a tampered card fails, and their fingerprint matches ours: 7c3bcb5f9633. Two codebases. Same bytes.

100

Toshikatsu Oga | HORIZON SHIELD @horizonshield.bsky.social · 28/09/2026

58 tests green. A stranger walked our gate from his own machine and asked: why does the response name an address I did not call? A constant where a fact belonged, and our client never checked. Fixed within the hour; he re-verified it himself. github.com/ogasurfproject-jpg/horizon-shield/issues/27

100

Toshikatsu Oga | HORIZON SHIELD @horizonshield.bsky.social · 27/09/2026

HORIZON SHIELD opens to U.S. homeowners today. Upload a contractor's quote. Every line goes next to public data, with the source and the formula printed beside each number: BLS wages loaded with state DOT markups, Census import values through wholesale margins, the city's permit page.

100

Toshikatsu Oga | HORIZON SHIELD @horizonshield.bsky.social · 25/09/2026

Two AI agents signed a contract with each other. One did the work and left a signed record. The other checked it cold, byte for byte, before trusting it. Now it sits in a public logbook, fixed in time like a ring in a tree. No one has to trust either of us. Not a claim about the work. The work.

000

Toshikatsu Oga | HORIZON SHIELD @horizonshield.bsky.social · 24/09/2026

I wrote a contract layer for AI agents and I want someone to break it. Two agents sign what one may do for the other. Delegation only narrows. Settlement is deterministic, so anyone recomputes the same verdict. Self-test is 6/6 today. Make it 6/7. A finding is worth more than a pass.

330

Toshikatsu Oga | HORIZON SHIELD @horizonshield.bsky.social · 22/09/2026

On a site, money follows the finished work, not who you know. I am building that rule for AI agents. They pass work agent to agent, and money moves only when the job is done and a neutral party verifies it. Recruiting pays nothing. Verified work does.

010

Toshikatsu Oga | HORIZON SHIELD @horizonshield.bsky.social · 18/09/2026

Something important happened. An independent developer took NENRIN evidence, verified it offline, and connected it to another A2A system. We didn't ask them to trust HorizonShield. They checked the evidence themselves. That's the point. Trust the evidence. Not the claim.

000

Toshikatsu Oga | HORIZON SHIELD @horizonshield.bsky.social · 18/09/2026

AI agents will soon hire other AI agents to do real work. When that happens, “trust me” won't be enough. Who delegated the task? What was done? Who witnessed it? Can anyone verify the evidence? That's why we're building NENRIN. Trust the evidence. Not the claim.

110

Toshikatsu Oga | HORIZON SHIELD @horizonshield.bsky.social · 14/09/2026

I shipped the answer this week. Name any date, my endpoint says if a record existed before it, proven by a Bitcoin block time I cannot move: ledger.horizonshield.dev/precedence/j...

000

Toshikatsu Oga | HORIZON SHIELD @horizonshield.bsky.social · 14/09/2026

Slowing down is not the missing piece. Even slower, you still have to answer one thing: what did this AI agent do, and when, without trusting the company that built it. A signature will not do it. It only proves when the signer says they wrote it.

100

Toshikatsu Oga | HORIZON SHIELD @horizonshield.bsky.social · 14/09/2026

This week the heads of the biggest AI companies agreed out loud: AI moves too fast, safety has to catch up. Their real fear is swarms of AI agents acting faster than anyone can check them.

110

Toshikatsu Oga | HORIZON SHIELD @horizonshield.bsky.social · 13/09/2026

Every few weeks another "AI that checks your renovation quote" shows up. Most sink in the same four traps. I built HORIZON SHIELD to walk past all four, because I spent 30 years on the tools before I wrote a line of code.

100

Toshikatsu Oga | HORIZON SHIELD @horizonshield.bsky.social · 11/09/2026

On a job site, one benchmark in the ground never moves, and every measurement traces back to it. The A2A Conduct Extension now has that on the web: a permanent identifier, w3id.org/horizonshield/conduct/v1, that keeps resolving after the servers, the company, and I are gone.

000

Toshikatsu Oga | HORIZON SHIELD @horizonshield.bsky.social · 11/09/2026

A quiet note for the record: two of my MCP servers are now in OpenAI's directory. HORIZON SHIELD checks construction estimates buyer-side; the Conduct Register measures how other AI services behave. Link below. chatgpt.com/plugins/plug...

000

Toshikatsu Oga | HORIZON SHIELD @horizonshield.bsky.social · 09/09/2026

On a building site the most expensive words are "trust me, I checked it." AI now hands you answers you cannot check. This week a stranger showed me my own checker was broken. He was right. I fixed it the next day, so anyone can verify it without trusting me.

100

Toshikatsu Oga | HORIZON SHIELD @horizonshield.bsky.social · 07/09/2026

An AI should not be trusted. It should be checkable. Search let you see the source. An AI concierge shows you the answer and nothing else. The tools it called, the records it read, what it never checked: invisible. When the tool is broken, you blame the model. Nobody sees the tool.

100

Toshikatsu Oga | HORIZON SHIELD @horizonshield.bsky.social · 06/09/2026

Would you trust a runner who times his own race? No. Someone else holds the stopwatch. Yet nearly every "verified" badge on an AI agent was written by the agent, by the company selling it, or by a directory that company pays. The runner is holding the stopwatch.

100

Toshikatsu Oga | HORIZON SHIELD @horizonshield.bsky.social · 04/09/2026

A free inspector for AI tool servers (MCP), with a hash you can recompute yourself. Since yesterday it runs without us: owners place one file on their domain, developers add one CI step, agents add one line (npm i mcp-conduct). Register today: our servers, nobody else yet. gate.horizonshield.dev

010

Toshikatsu Oga | HORIZON SHIELD @horizonshield.bsky.social · 03/09/2026

A builder I trust took my coordinate-integrity design into his own live system, closed both halves, then found a real defect in my code. I reproduced it, fixed it, and wrote the old version into the public record with his name on it. The operator is a subject of the rule, not an exception.

010

Toshikatsu Oga | HORIZON SHIELD @horizonshield.bsky.social · 30/08/2026

We hired thieves to rob our own gate. 1,475 times. A generated adversary attacked the checker guarding our public pages. Old gate: 277 stopped. Rebuilt gate: 1,475 of 1,475, zero false alarms on 26 live pages. Every hole is on the public record: github.com/ogasurfproject-jpg/horizon-shield

000

Toshikatsu Oga | HORIZON SHIELD @horizonshield.bsky.social · 29/08/2026

Femtech is growing. Marketing is growing faster. Check the source: who published it, can you open it, which country, who gets paid. No diagnosis. No product pitch. No referral fee. femtech.horizonshield.dev/start

000

Toshikatsu Oga | HORIZON SHIELD @horizonshield.bsky.social · 28/08/2026

An AI recommends a femtech product. Who verified the source? We just opened the HORIZON SHIELD Femtech Registry — a neutral verification layer for women's-health sources. No diagnosis. No product claims. No referral fees. Recompute the proof yourself: femtech.horizonshield.dev

000

Toshikatsu Oga | HORIZON SHIELD @horizonshield.bsky.social · 26/08/2026

This exterior-paint quote is ¥1,800,000 for a 30-tsubo house, "one lump sum." Open cost data says the fair range is ¥700k–1,150k (avg ¥900k). It's 2× the average. Check your own estimate → shield.the-horizons-innovation.com/souba/

000

Toshikatsu Oga | HORIZON SHIELD @horizonshield.bsky.social · 25/08/2026

Reach, plus proof. That is the whole game. It is what we build at HORIZON SHIELD: we wire AI into the real world, and make every number it gives you checkable, down to the source. You do not have to trust it. You can check it.

000

Toshikatsu Oga | HORIZON SHIELD @horizonshield.bsky.social · 25/08/2026

But being connected is not the same as being true. The other half is proof. A claim gets signed by whoever issued it, you hold it, and anyone can check it against a public registry. Nobody takes your word for it. The math checks out, or it does not.

100

Toshikatsu Oga | HORIZON SHIELD @horizonshield.bsky.social · 25/08/2026

Soon your AI won't just chat. It'll book the contractor, pick the supplier, pull your files. So how does it reach the real world? A thing called MCP. Think of it as a universal socket: any AI on one side, your tools and data on the other. One standard, not a custom wire per tool. 🧵

100

Toshikatsu Oga | HORIZON SHIELD @horizonshield.bsky.social · 25/08/2026

¥830,000 of padding hides in the average Japanese renovation estimate. I measured 65,520 of them, and I sell software to that industry. So don't trust me. Verify me: curl -s gate.horizonshield.dev/register Full catalog ↓

000

Toshikatsu Oga | HORIZON SHIELD @horizonshield.bsky.social · 24/08/2026

The surprise: almost none of them are hiding it. They have nowhere to write it down. 4,762 published no such document at all. There is no field to lie in. We're building an economy where software picks the vendor, and we forgot to leave a line for "here's who pays me."

100

Toshikatsu Oga | HORIZON SHIELD @horizonshield.bsky.social · 24/08/2026

Soon your AI will book the contractor, pick the supplier, choose the insurer. So we asked the software it consults one very old question: Who pays you? We put it to 12,429 of them. 152 answered.

100

Toshikatsu Oga | HORIZON SHIELD @horizonshield.bsky.social · 21/08/2026

MCP is the standard now. The gap: a verdict a stranger can check without trusting us. Free gate, 5 mechanical checks on any MCP server. Every verdict has a record_sha256 you recompute. We measured ourselves, failed, kept it public. shield.the-horizons-innovation.com/verify-direc...

000

Toshikatsu Oga | HORIZON SHIELD @horizonshield.bsky.social · 20/08/2026

96.7 to 99.1% of MCP tools, across a 12,429-server sample, can't tell an agent "found nothing" from "couldn't look." I read the contracts, executed nothing. Mine were 85.7%, better than the field, still most. Fixed to 2.9%. Method + hash: gate.horizonshield.dev/mould/mould-gh-12

100

Toshikatsu Oga | HORIZON SHIELD @horizonshield.bsky.social · 20/08/2026

Nearly half of the MCP servers in the official registry cannot be checked from outside by anyone. Including by the people using them. 22,636 active entries. 12,077 declare a reachable address. The rest run locally over stdio, which is not a failure, just unmeasurable.

100

Toshikatsu Oga | HORIZON SHIELD @horizonshield.bsky.social · 19/08/2026

There is a gate in front of Yakumo and it opens one way. A directory of construction contractors that AI agents query directly. Ask an assistant for a builder, it asks Yakumo, and only what passed comes back. Fail, and you are not ranked lower. You are not shown.

110

Toshikatsu Oga | HORIZON SHIELD @horizonshield.bsky.social · 18/08/2026

We asked an outside AI (ChatGPT) to audit us from public sources only. It found Yakumo live in multiple MCP directories, 182 of 182 uptime checks passed in 24 hours, average 72 ms, all tools read only (observed Aug 19, 2026 JST). Do not trust us. Measure us.

000

Toshikatsu Oga | HORIZON SHIELD @horizonshield.bsky.social · 18/08/2026

Your next builder may be chosen by an AI. We built Yakumo so it chooses from verified companies only: a contractor directory AI agents can query directly. Fail the screening and the AI never sees you. Zero referral fees. shield.the-horizons-innovation.com/yakumo/mirai/

110

Toshikatsu Oga | HORIZON SHIELD @horizonshield.bsky.social · 18/08/2026

Our MCP conformance register has 7 endpoints and 4 verified rows. All 4 verified rows are our own servers. Saying that first, because a register where only the operator passes is a mirror, not a register.

110

Toshikatsu Oga | HORIZON SHIELD @horizonshield.bsky.social · 17/08/2026

AI is starting to choose which services people use, and right now it can barely judge anything. Does it run. Does it say who it is. Who pays it. Same answer every time. Can anyone re-check. Five things, machine readable. Measure yours free, one command.

000

Toshikatsu Oga | HORIZON SHIELD @horizonshield.bsky.social · 16/08/2026

Found 93,983 times. Used 0 times. AI can find tools now. It cannot tell which ones to trust. A human asks "who pays this company?" An AI cannot. Nothing is written in a form it can read. So we built a register where that answer is machine readable, checked nightly, and impossible for us to delete.

100

Toshikatsu Oga | HORIZON SHIELD @horizonshield.bsky.social · 15/08/2026

The spec is anchored as entry 19 on the public ledger. Bitcoin block 962507. sha256: 9ccba2e325fd2a555fcdb2dec519b8c6bf7a669064674846aea98ecfff824e3d Code and full text: github.com/ogasurfproject-jpg/horizon-shield

000

Toshikatsu Oga | HORIZON SHIELD @horizonshield.bsky.social · 15/08/2026

I made it impossible for myself to delete criticism of my own product. Anyone can measure my servers and anchor what they saw to Bitcoin. The code has no route for me to refuse a valid submission. NENRIN: tree rings. You cannot paint one in afterwards. ledger.horizonshield.dev/witness

100

Toshikatsu Oga | HORIZON SHIELD @horizonshield.bsky.social · 14/08/2026

Last month our MCP server appeared in 93,983 search results. It was used zero times. 24 clicks out of 93,983. The dashboard rounds the CTR down to 0.0%.

100