Contact us
Changelog

What shipped. What’s next.

Real entries only, derived from the actual build history — no invented milestones. The roadmap below is the build plan from the spec, and it is labeled as exactly that: not shipped.

Shipped
2026-06-11

Launch polish — the honest pass

Site-wide messaging reconciled to the build spec: the substrate story on every page. MCP & API docs rebuilt on the real tool surface — ~27 tools across memory, causal, decisions, and governance, each with its worked call. A real market-data layer behind the new Signals and Forecasts surfaces (live Polymarket readings, attributed and timestamped; causal overlays stay labeled illustrative). Agent surfaces shipped: llms.txt, AGENTS.md, /api/context, /api/tools.json. Try-it sandbox on the docs, mobile nav drawer, brand-law sweep, honest forms, and this page.

docslive dataagent surfacesnavigation
2026-06-11

cascadian.ai is live

First production deploy. The site ships as a static-first Next.js build — no backend pretending to exist; the alpha server stays gated behind paid pilots.

deploy
2026-06-11

The homepage sequence

Hero scroll-wipe rebuilt as a single timeline, with the real data-source marquee (Kalshi, Polymarket, SEC EDGAR, FRED). The shift, edge, and stack sections integrated from the design system; seven how-it-works animation variants behind a corner switcher; the agent-native convergence band.

homepagemotion
2026-06-11

Site scaffolded on Trace

Next.js app stood up on the Trace brand system — monochrome field, hairline structure, signal red reserved for live data. Pages: how it works, causal intelligence, architecture, concepts, who it's for, proof, services, investors, contact, and the MCP & API documentation.

foundationbrand
Roadmap — build plan, not shipped

The server, in five milestones.

The substrate itself — one MCP server in front of one Postgres — ships through the milestones below, each with a testable exit. Until an exit suite is green, nothing here is claimed as working product.

M0

Substrate

Temporal envelope, append-only + hash-chain triggers, the as-of repository, audit writer, approval CLI, MCP skeleton.

Exit criteriaInvariant suite green — triggers reject mutation, the hash chain verifies, and the approval token provably never crosses MCP.

M1

Honest memory

The full memory surface: ingest, facts, proposals, contradiction checks, hybrid search, entity dossiers, the context pack.

Exit criteriaDocuments become classified, provenance-carrying, as-of-queryable memory; a late-arriving fact is invisible before its available_at.

M2

Causal engine + knock-on

Governed graph with versioning and approvals, identification gate, simulate_intervention, the deterministic skeptic checks.

Exit criteriaThe engine recovers seeded known effects within tolerance, and the lookahead and post-treatment traps are caught by test.

M3

Decision ledger + compounding loop

Open decisions with pinned packs, frozen expectations, server-side approval gates, outcomes, learning proposals, reconstruction, leakage detection.

Exit criteriaOne end-to-end run — question → pack → identify → simulate → recommend → approve → outcome → learning — and the next decision's pack cites it.

M4

Estimation depth + calibration

Four real estimators with bootstrap CIs and dataset hashing, change attribution, calibration review, the complete capability registry.

Exit criteriaEstimators validated against simulator ground truth; calibration runs over ≥10 seeded decisions; capabilities() honestly partitions implemented vs interface_only.