Platform Status
What’s stable today, what we’re building, and what’s still research. Every Stable layer has a passing, reproducible benchmark behind it — we don’t call something stable until it’s measured, then we freeze it.
Decision Platform — stable core, frozen
Decision Engine V1
Frozen — benchmarked for discovery quality
Decision IR (Representation)
Frozen — golden-replay validated
Decision Context Runtime
Frozen — knowledge routing
Knowledge Reliability
Frozen — per-item reliability contract (KRB-001→006)
Decision Contract
Understand the decision before the engine
Decision Support
Actionable options — Decision Actionability benchmark
Decision Chat
Grounded in the decision — continuity benchmark
Progressive Profile
User-provided facts, privacy-controlled
Knowledge Platform — the evolving layer
Provider Activation v1 (Public-Live)
Registry + Public-Live route the fetch behind a flag; 17/17 benchmarks + golden OFF≡ON regression; frozen (ADR-029)
Provider Registry
Selects who can answer, never what — explainable + abstains; frozen v1
Public-Live Provider
Best public knowledge today; OFF≡ON equivalence proven; routing by source + replay (not “future”)
Public-Historical Provider
Frozen v1 — replay keeps ≤-cutoff sources, drops/excludes the rest with an explainable trace; corpus 14/14 leak-free; retrieval leak solved (residual parametric leak is a documented model limit, not a provider defect)
Personal Provider
Your profile, history, preferences
Enterprise Provider
Internal documents, policies, processes
Financial Provider
Market, filings, valuation, macro
Knowledge Qualification Layer
Arbitrates across providers — agreement / freshness / authority / applicability. Only once ≥2 providers are live
Monitoring Platform
Monitoring
Watch a decision after it's made; re-evaluate when the world changes
Learning
30-day outcomes → prediction-vs-reality → decision memory
Applications
General Decisions
The shipping product today
Investment
Prototype — investment decision support
Enterprise Readiness
An active research initiative, not a product yet
Healthcare
Future application domain
Known limitations
We publish these on purpose — accurate expectations build more trust than polish.
- •Knowledge providers are still expanding — some decisions reason from the model's own knowledge.
- •Historical replay improves as dated providers are added; point-in-time correctness is verified, not assumed.
- •Confidence is shown only when it's measured against something reproducible — where we can't calibrate it, we say so instead of showing a number (ADR-031).
- •Enterprise Readiness is an active research initiative, not a certified capability.
- •Recommendations are designed to support, not replace, professional judgment.
Roadmap
Completed
- ·Decision Platform (Engine, IR, Context, Reliability)
- ·Decision Contract
- ·Decision Support
- ·Decision Chat
- ·Progressive Profile
- ·Provider Activation v1 (Public-Live)
- ·Historical Provider Activation v1
Now building — finish the Provider Platform
- ·Financial Knowledge Provider (next — knowledge, not prediction)
- ·Personal Provider
- ·Enterprise Provider
- ·→ then Freeze Provider Platform
Later
- ·Knowledge Planning Layer (what knowledge is needed — deterministic)
- ·Knowledge Qualification Layer (once ≥2 more providers exist)
- ·Monitoring
- ·Foundation Knowledge Store
- ·Decision Relevance Layer
- ·Learning Platform
How this is built: each layer ships only after passing a benchmark, then is frozen — we don’t reopen a working layer without measured evidence of a defect. That discipline is why this page is conservative about what counts as “stable.”