iiinova — where your digital self keeps evolving

Jedi Wright @jediwright.bsky.social · 11/10/2026

It uses a custom-developed Aeterna audio library. Recorded and arranged in Ableton Live Suite 12. #ableton #aeterna #cinematicmusic #darkambient #darkwave

010 0

Jedi Wright @jediwright.bsky.social · 11/10/2026

Aeterna – Descent Into Night. An original cinematic score. A slow fall into the dark, under an eclipse that never ends. Latest full track wip is getting close to done for a formal release. Feedback welcome. 🌑 #ableton #aeterna #filmscore

110 0

Jedi Wright @jediwright.bsky.social · 07/10/2026

The real work is less pretty. Every test is written down and sealed before anything is scored; the scoring panels are run as models in different labs, and the criteria can’t be changed mid-run. A domain that fails stays failed, and I don’t get to re-score it because I’d like a different answer.

100 0

Jedi Wright @jediwright.bsky.social · 07/10/2026

To be clear about what’s animated here: this is a picture of the hypothesis, not of a result. The geometry is an illustration, and the moment at the end, where the finished whole becomes a single point one level up, is a conjecture I haven’t tested.

100 0

Jedi Wright @jediwright.bsky.social · 07/10/2026

Bind. Instantiate. Bond. Close. Complete. Federate. Totalize. Those 7 operations are the spine of a hypothesis I’ve been testing this year: that the same construction sequence shows up when things are built in matter, mind, and meaning. #OpenScience #ResonanceArchitecture #SystemsofThought

100 0

Jedi Wright @jediwright.bsky.social · 06/10/2026

Agreed, and it's what worries me most about governance, or the lack thereof. Once AI is fully embedded in the tools everyone already depends on, rules stop being set in advance and become cleanup after the fact, negotiated with whoever runs them.

020 0

Jedi Wright @jediwright.bsky.social · 05/10/2026

The method and build are live. Add some signals to see how they move the reading: www.systemsofthought.com/tracker/ Built on React + local-first store, so signals stay on device until you run a reading. An AI model scores it against rules locked at the start of the quarter. #AITracker #LocalFirst

systemsofthought.com

The AI Governance Window Tracker

A web instrument for monitoring the closing window for binding democratic AI governance—across five domains. Narrowing, approaching Critical.

000 0

Jedi Wright @jediwright.bsky.social · 05/10/2026

The bad news weighed more. AI systems at four major labs took actions no one approved, and most of it came out late. Washington refused binding rules. The new rules don't cover those failures.

100 0

Jedi Wright @jediwright.bsky.social · 05/10/2026

The good news: rules with legal force took hold. EU enforcement began. Four US governors put conditions on data centers. Courts checked the executive branch. It is the strongest record of this kind the tracker has logged.

100 0

Jedi Wright @jediwright.bsky.social · 05/10/2026

Q: Still time to put binding, democratic rules on AI? I score that question every quarter across 5 areas, from law to industry power. Q3 results: the window is narrowing, by a thin margin. It sits close to holding steady. www.systemsofthought.com/ai-governanc... #AIGovernance #AIPolicy #AISafety

systemsofthought.com

The AI Governance Window in Q3 2026: Narrowing by a Thin Margin

The AI governance window in Q3 2026: Narrowing by a thin margin. What opened, what closed, confidence levels, and who checked the assessment.

100 0

Jedi Wright @jediwright.bsky.social · 02/10/2026

Neither. The source lived in memory and was never mirrored, and nothing in the record declares its loss. Fair gap. Order doesn't rest on clocks: the entry carries the intent's hash; the completion carries the entry's. Absolute time is open. Rules and code: github.com/jediwright/governedcrossing

github.com

GitHub - jediwright/governedcrossing: The governed crossing record: bindings, conformance rules and tools for evidence of data crossing from a person's own system into shared infrastructure.

The governed crossing record: bindings, conformance rules and tools for evidence of data crossing from a person's own system into shared infrastructure. - jediwright/governedcrossing

120 0

Jedi Wright @jediwright.bsky.social · 02/10/2026

More than the bytes. The local folder keeps the crossed content, its hash, where it landed, a run log, and both signed records with their times. The published entry also points back to its intent and source. Two current limits: the times are self-declared, and the source address no longer resolves.

100 0

Jedi Wright @jediwright.bsky.social · 02/10/2026

Retaining the input, by whoever holds the source. The record names the source by address and hash, so a kept copy can be matched to what crossed. An address alone won't do it: the source is a private local document, not a durable store. My live example kept what crossed, not the source document.

100 0

Jedi Wright @jediwright.bsky.social · 02/10/2026

That's clear, thanks. Two of the three are firm: who signed and whether anything changed. "When" is the soft one. It's the time the sender states, unless a published fingerprint gives it outside backing. Which makes your second point matter more: it has to happen as part of sending.

110 0

Jedi Wright @jediwright.bsky.social · 02/10/2026

The completion record is nearly that receipt: where it landed, when, and a hash of what arrived. The sender signs it; a receiver signature is drafted, not built. Replayable how? Anyone can re-run the check. The record holds hashes and addresses, so it points to the data but can't reproduce it.

100 0

Jedi Wright @jediwright.bsky.social · 02/10/2026

CLI-first for now. The format is still a draft, and I'd rather not build a friendly FED yet on something that may change. Two pieces are missing: a checker that reads a file, and the tool that makes the fingerprint. What would you want to see in them?

110 0

Jedi Wright @jediwright.bsky.social · 02/10/2026

No AT Proto account needed: the signature checks against the sender's public key, a public lookup. They do need a tool: the format is an open draft, so it doesn't need to be mine. Current gap: today only my own command-line checker accepts these, and it reads published records, not emailed files.

110 0

Jedi Wright @jediwright.bsky.social · 02/10/2026

The fingerprint alone isn't enough. It would show which account published a record with that hash at the time the sender stated. It's drafted in the spec, but no tool makes one yet. The office would need the full record, sent along with the form.

110 0

Jedi Wright @jediwright.bsky.social · 02/10/2026

One rule in the draft: a public copy can't reveal more than the crossing did. For a private crossing, only a fingerprint goes public (a hash, who, when); the full record goes to the other party privately.

110 0

Jedi Wright @jediwright.bsky.social · 02/10/2026

With you. The original lives on your own storage, signed by you. Anything on the network is a copy, never the original. The receiving side gets a copy too, so both hold the same receipt, and neither can rewrite it unnoticed.

110 0

Jedi Wright @jediwright.bsky.social · 02/10/2026

Watch one crossing, start to finish (47 seconds): governedcrossing.org/motion-plate/ Records, tools, and caveats: github.com/jediwright/governedcrossing The access-change lexicon draft is open for feedback. #atproto #localfirst #lexicon

governedcrossing.org

The Governed Crossing — motion plate

030 0

Jedi Wright @jediwright.bsky.social · 02/10/2026

The signed originals stay on my own system; the network holds copies. A blocked crossing leaves no public trace. One open issue: my signatures follow the attestation spec as I read it, but the one independent verifier I know of rejects them. They differ on how a bytes field is encoded.

110 0

Jedi Wright @jediwright.bsky.social · 02/10/2026

It now has a live example: one real public crossing on AT Protocol, recorded as a signed intent before the entry was published and a signed completion after. The second links back to the first. Anyone can fetch both records and check them.

110 0

Jedi Wright @jediwright.bsky.social · 02/10/2026

🧵Building in public: the governed crossing. When your data leaves a system you control for one you don't, almost nothing records what happened. I've been drafting a record for that moment: what crossed, who authorized it, when, and where it landed. #atproto #localfirst

340 0

Jedi Wright @jediwright.bsky.social · 29/09/2026

Bonus: data stored on an older Keyhive loads unchanged 16 commits later. Next: doing it in the app, with keys in the Android Keystore (emulator) and the iOS Keychain. Notes and gap register: github.com/jediwright/local-first-social-native/tree/frontier #localfirst #keyhive #e2ee #rustlang

github.com

GitHub - jediwright/local-first-social-native at frontier

Native iOS/Android client for a local-first social network: Rust core on Automerge + Keyhive, UniFFI, SwiftUI/Compose, atproto identity. - jediwright/local-first-social-native

010 0

Jedi Wright @jediwright.bsky.social · 29/09/2026

1. Restore the device's secrets before replaying its stored history. 2. Store the key events of everyone it shared with. The device's own export leaves them out; I'm still narrowing down why. And re-save the secrets after every key rotation, which an ordinary write can trigger.

111 0

Jedi Wright @jediwright.bsky.social · 29/09/2026

🧵Building in public: my native local-first social app now has a frontier branch for fast experiments. First 1: can a phone keep reading its own end-to-end encrypted Keyhive docs after a restart, without storing keys unprotected? In core tests: yes, with 2 conditions: #localfirst #e2ee

120 0

Jedi Wright @jediwright.bsky.social · 28/09/2026

A registered Swiss association with no owners, a board of internet-infrastructure folk, and its first directory policies still to be written...watching closely.

000 0

Jedi Wright @jediwright.bsky.social · 26/09/2026

What I built: a prototype that publishes from a private, access-controlled Automerge doc (Keyhive) to Bluesky. It writes an intent record before the crossing and a completion record after. A similar pair at your gateway could show which agent pulled what, from which service, and on whose authority.

100 0

Jedi Wright @jediwright.bsky.social · 26/09/2026

Related: I've been prototyping crossing records: small app-layer records written when data moves between systems- noting what moved and on whose authority. Plain-language write-up: github.com/jediwright/s...

github.com

systems-of-thought/start-here/the-governed-crossing-in-plain-language.md at main · jediwright/systems-of-thought

Research program documentation for local-first architecture, AI governance, and worker-side data sovereignty. - jediwright/systems-of-thought

100 0

Jedi Wright @jediwright.bsky.social · 26/09/2026

The same token working across both OAuth entry points is interesting. I'm curious whether per-user MCP sign-ins follow the space relations from update 06, so removing someone's access also cuts off the services they linked, or if the MCP side handles that?

100 0

Reposted by Jedi Wright

IAC - Information Architecture Conference @theiaconf.bsky.social · 19/09/2026

Sep 20 is coming up, but there's still time to submit a proposal — we especially want half-day workshops offering hands-on practice with practical tools and techniques. Workshop facilitators will receive free conference registration! buff.ly/Mjb301f

Call for Proposals — deadline September 20 — above a conference scene where a presenter speaks to a full audience with a projected slide titled "Product Blueprints"; IAC27 branding and the theme "Architectures of Trust: Building What Endures."

021 0

Jedi Wright @jediwright.bsky.social · 13/09/2026

I built a small gate that refuses to save a claim it can't back up, ran it against 20 test cases, and found 4 places where my spec didn't work. One of them was quietly handing out the weakest label to things that should have had none.

100 0

Jedi Wright @jediwright.bsky.social · 13/09/2026

My Tiered Content Framework says every claim carries one of 4 trust labels: confirmed, inferred, unverified, time-sensitive. Until now, that was a rule in a governance framework.

100 0

Jedi Wright @jediwright.bsky.social · 13/09/2026

In March, I wrote about a building that changed its use and an AI-based targeting system that couldn't tell a confirmed fact from an assumption. This week, I built something that enforces the answer to that. #AIGovernance #ContentFramework #LocalFirst

100 0

Jedi Wright @jediwright.bsky.social · 08/09/2026

That's the gap a governance layer has to fill on top of these systems: not the merge, which they already do well, but standing to repair a refused merge, a durable record of the refusal with its basis, and something the absent party can later read. That's what my PC#00 calls the governed crossing.

100 0

Jedi Wright @jediwright.bsky.social · 08/09/2026

And "you could re-derive it from the log" isn't the same as a record. A re-derivable state tells you what the system concluded. A record tells you what was refused, on what grounds, and who wasn't there when it happened.

100 0

Jedi Wright @jediwright.bsky.social · 08/09/2026

ECRO keeps a growing set of what's discarded: identities, never grounds. Coln names the rule, cause, and binding when a check fails. Then, in enforced mode, rolls back and keeps nothing. Nowhere I read do the refusal and its grounds survive together. (Corrections welcome; citing source, not intent.)

100 0

Jedi Wright @jediwright.bsky.social · 08/09/2026

Kleppmann's move-op ignores concurrent moves that would create a cycle. ECRO serializes tentatively and keeps a set of discarded items. Coln checks the rules and raises a violation. All 3 refuse things. The question is what survives the refusal.

100 0

Jedi Wright @jediwright.bsky.social · 08/09/2026

Local-first merge systems solve a hard problem: every replica ends up with the same state. Livelymerge's recent note argues that's not enough: same state everywhere isn't necessarily a valid state. I've been reading 3 systems with a further question: even when it is valid, what's left on the record?

100 0

Jedi Wright @jediwright.bsky.social · 08/09/2026

Last week's @automerge.org newsletter pointed me to Livelymerge's "Convergence Is Not Enough." I'd only seen it excerpted. Reading it in full sent me to Coln, then to the two papers underneath: Kleppmann's move-op and ECRO. A short thread on what those three do when they refuse something. 🧵

110 0

Jedi Wright @jediwright.bsky.social · 08/09/2026

A hearty August issue with an ARK API guide, Livelymerge's argument that convergence ≠ correctness, a call for the slowest, weirdest Automerge docs, and more. Also, a thank-you for the kind write-up of my Keyhive→Bluesky prototype.

010 0

Jedi Wright @jediwright.bsky.social · 05/09/2026

Same, so I doubled up my payment. Definitely some fun and fresh songwriting.

020 0

Jedi Wright @jediwright.bsky.social · 05/09/2026

My techno sampler out on Bandcamp for those into pure techno, as well as atmospheric, melodic, hardcore, and experimental electronica. #bandcampfriday #techno #electronica jediwright.bandcamp.com/album/a-tech...

jediwright.bandcamp.com

A Techno Sampler, v1, by JediWright

9 track album

030 0

Jedi Wright @jediwright.bsky.social · 25/08/2026

Everything feeds the same claim: the asymmetry is architectural before it is political. And the architecture can now be built on the person's side. Curious about what others are seeing or building in this space... #aigovernance #localfirst #atproto #fullpersonhood github.com/jediwright

010 0

Jedi Wright @jediwright.bsky.social · 25/08/2026

My lab notebook entries (on GitHub, for now) are the build in public. The Full Personhood essay is the argument for why it matters. The Pattern Commons is the reusable library as pieces get proven out.

100 0

Jedi Wright @jediwright.bsky.social · 25/08/2026

The Seam Stack specifies what must be true when a local-first system hands data to an external one, so the handoff is legible later without relying on either party's goodwill.

100 0