How we get to an answer.
You set the goal. We read your tools, weigh both sides, and hand you the call.
- Strategy
Write down what you want.
Your strategy is the workspace's constitution: the direction, who it is for, and what you will not do. One version is active at a time. Edit it and you get a new version, and old decisions keep the one they were made under.
- Goal
Break it into goals.
A goal is an owned, time-bounded outcome. You or an agent can propose one; only a human can accept it. Accepted goals share a hundred importance points per horizon, so the trade-off is explicit and on the record instead of relitigated every planning meeting.
- Signal
We collect the notes.
Every ticket, thread, metric and quote becomes a note with its source and time attached. Notes are never rewritten. If something turns out to be wrong we add the correction as a new note, and what a past decision stood on stays as it was.
- Claim
We find what the notes agree on.
When several notes point the same way we write it down as a claim you could prove wrong, with its evidence and its counter-evidence. Groupings are disposable and claims are versioned, so re-running the analysis cannot quietly change what a past decision relied on.
- Alignment
We check each claim against each goal.
A claim can support one goal and conflict with another, and we show both. The verdicts are supports, mixed, conflicts and unknown, with strength and confidence kept separate. A percentage appears only when it means something specific, and we say what.
- Recommendation
We rank the next moves.
Ideas are ranked by how far they move your goals, given the evidence, the cost, the risk and what you would learn. Each one names its goal, its evidence path, its alternatives, and what would change our mind. Confidence measures belief in an effect and priority measures strategic value, which is how a riskier idea can outrank a safer one and still be explainable.
- Decision
You make the call.
Accept, reject or defer. Your decision freezes as a snapshot: the idea, the strategy version, the goals, the evidence, and the alternatives you saw. Nothing that runs later can rewrite it.
- Outcome
We watch what happens.
After a decision, results come back as new evidence, with attribution kept explicit because other work may have caused them. When the strategy changes, we flag the reasoning that went stale.
Six months on, the reasoning is still attached to the decision that used it.
Rules the loop cannot break.
Each of these is enforced in the data model, not in a prompt.
- An agent-proposed goal cannot affect a ranking before a human accepts it.
- Editing the strategy creates a new version and marks the affected answers stale for review.
- Every claim links to the notes it was built from.
- The same claim can support one goal and conflict with another.
- Every idea names a goal, an evidence path, and what would change our mind.
- The engine is allowed to recommend stopping, or doing nothing at all.
- No unexplained percentages, anywhere.
The loop ends where a person starts.
Everything above narrows the question. None of it makes the call, and none of it can be configured to.