Open intelligence, built in the open. IIM is the house everything here ships under — OSINT research, reporting, and the software that makes both usable by someone who cannot buy a terminal.

What Wall Street hoards, we hand back.

That sentence is from tuck, but it is the whole account's thesis. Congressional trades, conflict data, surveillance contracts, crisis resources, film archives — all of it is information someone else charges rent on.

Some of the work below is done for VPDLNY. That part keeps its own counsel; the tooling is public even where the work is not.

100% Creative Clarity · osintnet.uk · Discord

ka-tet · orchestral agentic AI

Everything else on this page is built by this. Four AI agents and one human holding a single system upright. The architecture is engineered on the structure of Stephen King's Dark Tower novels — not as decoration, but because that cosmology is already a hub-and-spoke topology with guarded endpoints, an explicit integrity law, and a named failure mode, which is what a distributed agent system actually needs.

One writer, and that is the load-bearing constraint. Four seats research, draft, rank and sanitise. Exactly one may publish. That is a lock rather than a rank, and it was earned — two writers without one collided on a repository here, and a force-push dropped four files.

The Gunslinger — the forkable architecture. 24 chambers, exactly one mounted at a time, 241 dated action items across 15 numbered chapters. · The Second Gunslinger — 24 chambers for people in difficulty; seven of them fail closed and hand the action to a human rather than trusting the model to be careful. · The Archivist — the publisher, whose main page is a table of its own recorded mistakes. · The Gun — where the whole thesis started: the original 24-chamber routing design, drawn before any of it was running.

flowchart LR
 T["Target named<br/>at fork time"] --> R["Roster discovery"]
 R --> C["24 chambers<br/>one loaded at a time"]
 C --> B["Dated brief<br/>expires on schedule"]
 B --> W["Working agent"]
 W -. "findings ledger" .-> R

The creed is the architecture, not a mission statement bolted on. Five stanzas, each mapping to a component and the specific failure it forbids.

No text from the novels is reproduced; the creed is original and the avatars are original designs. Attribution.

warheatmap.app

Live global conflict intelligence, free. An interactive heatmap built on 4,000+ verified conflict events rather than headlines, with naval OSINT and autonomous posting on top.

The most-used thing on this page. No login, no paywall, no ads.

conflict data · naval OSINT · live map · verified events, not headlines

AXIOM

Autonomous Discord moderation intelligence. Reads a server in real time and catches trolling, harassment, bullying, political charging and spam with confidence-scored classification — while protecting good-faith debate and friendly chatter, which is the hard half. Graduated strikes, one-click enforcement.

The interesting part is the restraint: a moderator that cannot tell an argument from an attack is worse than no moderator. AXIOM publishes its discipline and keeps its calibration private.

Companion: axiom-scanner — a free, read-only web-security scanner you run from Discord.

Read off the running deployments via the Cloudflare API on 2026年09月03日, not copied from a doc.

real-time gateway · AI classification · graduated enforcement

vibemaestro.app · status

My own complete vibe-coding and orchestral AI studio. Describe an app in chat and VibeMaestro builds, ships, and publishes a real, live web app on the Cloudflare edge — tiered model routing, per-user spend caps, and a free build lane so anyone can create for 0ドル.

Conduct the code. Ship it or it didn't exist.

It publishes its own uptime. Eight public endpoints, monitored continuously, readable by anyone at status.vibemaestro.app — no login, and a JSON API at /api/status for anyone who wants to poll it themselves.

One conductor, 23 deployed workers. Enumerated live on 2026年09月03日.

The status worker in that table is the one serving the status page above — the platform monitors itself and publishes the result.

The three security and secrets workers are counted but not named. Everything else is listed by name; those three handle auth and secret material, and a public list of their hostnames would mainly be useful to someone probing the platform. The count is there so the picture is complete.

Kelvin — quant trading performance and risk observability. Four pillars: performance (indexed equity curve, daily P&L, expectancy); risk (Sharpe, Sortino, maximum drawdown, R-multiple, regime conditioning); attribution (per-strategy status, machine-decision log); and discipline, which benches negative-expectancy strategies and refuses trades in violent regimes. Machine in supervised training — figures are indexed demo units, equity indexed to 100, not account balances. Not financial advice.

Kelvin · kelvinquant.com

A trading machine under supervision, and the instrumentation that holds it to account. Indexed equity curve, expectancy, Sharpe, Sortino, maximum drawdown, R-multiple and regime conditioning, with per-strategy attribution and a machine-decision log.

The interesting behaviour is the refusal. Kelvin benches negative-expectancy strategies and declines to trade in violent regimes — a system whose most important capability is not acting. Same principle as AXIOM: publish the discipline, keep the calibration private.

The hub repo writes that boundary down — what is published, what is withheld, and why the equity curve is absent rather than estimated where no indexed export exists. It is also built to be read by machines: the front end ships an llms.txt describing itself, and the crawler policy admits AI agents deliberately.

Machine in supervised training. Figures are indexed demo units — equity indexed to 100, daily P&L in percent — not account balances. Not financial advice.

Free financial intelligence for everyone else: Tuck — congressional trades, sector heat, a geopolitical scanner and macro, in a single Cloudflare Worker. No login, no ads, no advice.

quant observability · risk analytics · regime conditioning

A wrong number in a crisis is worse than no number. That is why the Crisis Lifeline Bridge phone-verifies before it hands anything over, and why seven of the Second Gunslinger's chambers fail closed instead of improvising.

Shipping the tool is half of it. These are the patterns themselves, written down so someone else can rebuild them.