Catalyst-Q

Ship Roadmap

A checkmarkable path from current Catalyst-Q workbench to paid-pilot-ready vertical agents and benchmark-backed production confidence.

Project phases.

These are the remaining product and engineering gates.

7/12 complete

Make the Cloudflare Worker deployable, observable, and protected across environments.

13/17 complete

Ship the first paid-pilot-ready vertical with public benchmark evidence.

DONE: Freight vertical spec includes TAM, freemium offer, guardrails, Catalyst-Q tools, Browser Run plan, and realtime data plan.
DONE: RouteOps local eval compares nearest-neighbor baseline to Catalyst candidate solver.
DONE: RouteOps live eval downloads and parses CVRPLIB A-n32-k5.
DONE: Prompt rubric checks constraints, KPI proof, data freshness, .rain, Catalyst-Q, and approval language.
DONE: Catalyst scoring weights can be tuned with npm run eval:routeops:tune.
DONE: RouteOps Proof Room shipped for human and machine reviewers with connector readiness and claim gates.
DONE: Route Decision Record field schema exposes inputs, constraints, baselines, candidate deltas, risk, approval, rollback, replay, signature, and export formats.

3/10 complete

Make every agent run replayable and useful for prompt/model/solver iteration.

DONE: Brain context injection exists for agent prompts.
DONE: Brain pulse hook exists for run summaries.
DONE: .rain event format stores observe, decide, simulate, correct, and commit events.
PENDING: Define tenant memory namespaces per vertical and customer.
PENDING: Add retention policies for free, pro, and enterprise tiers.
PENDING: Add replay endpoint for a specific .rain run id.
PENDING: Add redaction policy for sensitive operational data before Brain pulse.

5/9 complete

Connect agent evidence capture and realtime operational data into the production runtime.

DONE: Production runtime catalog names Browser Run, Live View, human handoff, Pipelines, Analytics Engine, and operational datasets.
DONE: Worker has Browser Run, Worker Loader, Pipeline, and Analytics Engine bindings.
DONE: Implement Browser Run tool adapter for approved customer portals and evidence screenshots.
DONE: Implement Pipeline event writes for eval runs, vertical runs, and model route outcomes.
PENDING: Add dashboard queries for ROI evidence.
PENDING: Add human-handoff state to vertical-agent outputs.
DONE: Browser Run can capture an approved staging dashboard screenshot.

6/11 complete

Make the project mergeable, deployable, supportable, and measurable.

DONE: CI workflow runs typecheck, RouteOps eval, Wrangler dry-run, and local smoke.
DONE: Live benchmark workflow runs RouteOps tuning and uploads report artifacts.
DONE: Staging deployment workflow exists.
DONE: Production deployment workflow exists with GitHub environment approval.
DONE: Smoke-test script exists for local and deployed routes.
DONE: Security and operations docs exist.
PENDING: Full public API/MCP schema docs are complete.