Hypothesis (Opportunity) Lifecycle

A hypothesis tracks an idea from initial signal to validated opportunity.

Every change to a hypothesis is recorded in an immutable changelog (before/after snapshots with timestamps), giving a full audit trail of how the hypothesis evolved.

Creator and assignee. Every hypothesis shows who created it - also when it came from converting a finding, which is marked as such - and the creator never changes. A shared hypothesis can additionally be assigned to one person in the workspace who looks after it; anyone who can edit the workspace can set, change or clear the assignee at any time, and only people who can edit the workspace can be assigned. The new assignee is notified, the change appears in the changelog, and someone who leaves the workspace (or can no longer edit it) is unassigned automatically. In the hypothesis list, people can filter by creator and assignee, with one-click "Created by me" and "Assigned to me" filters to find their own hypotheses.

Who is needed for each implementation step. When the AI drafts or adjusts the implementation plan of a shared hypothesis, every step names the business function needed to carry it out - for example Procurement, Engineering, Production or Logistics, at most two per step. It names functions, never individual people, so the team can see at a glance which departments to involve. Plans written before this was introduced are not changed retroactively.

The dashboard's hypothesis-flow view shows how ideas progress through validation, where rejected ideas left the process, and how value is split by origin. Users can switch between counts and summed potential, and follow links from status totals to the matching hypotheses.

Splitting a new hypothesis off a focused chat. Each hypothesis has its own chat focused on that one idea. If the conversation drifts onto a distinct, self-contained idea, the AI names the drift and offers to split it off as its own hypothesis rather than diluting the one the chat is about. Accepting the offer creates that new hypothesis immediately in the workspace - with its own detail chat to open and continue in - rather than leaving it as something the user still has to separately save.

Splitting a hypothesis that is too broad. Sometimes a hypothesis bundles several things that should be assessed and implemented separately - two materials, two plants, two different levers. When the user says so in the hypothesis chat, the AI works out a split plan: the new hypotheses, each with its own description, its share of the figures and its potential, and whether the original is discarded or kept with whatever is not split off - the user decides that. Figures are divided, never duplicated: an addressable spend is shared out between the new hypotheses, while a rate can apply to each of them, and the potentials add up to the original's. The method carries over, assumptions and findings are copied, and linked business data is narrowed to what each new hypothesis is about. The plan is shown as a card and can be changed in the chat; nothing is created until the user confirms it, and then everything happens in one step, so the dashboard never counts the same potential twice. The new hypotheses show where they came from, and a discarded original records which hypotheses it was split into. A split cannot be undone.

Linking a hypothesis to real business data. When a workspace has curated lists of real values - suppliers, materials, plants, and similar - a hypothesis can be tagged with the specific ones it concerns. The AI recognises these names as they come up in conversation and links the hypothesis to them automatically, matching case and spelling variations so a name typed slightly differently still resolves to the right entry. The user can also be offered a searchable pick-list to choose values directly, and that choice is saved immediately as a confirmed link - as well as being able to confirm, reject, or mark a suggested link as not applicable at any time. These links then power the knowledge graph below, clustering hypotheses that share the same real-world entity.

Asking the AI about existing hypotheses. In any chat, users can ask which hypotheses the workspace already has - on a topic, for a particular supplier or other linked value, in a given status or method, or above a certain potential - and get a list of matching hypotheses they can open directly. They can also refer to a hypothesis by its number (e.g. "HYPO-3") and ask the AI to summarise, explain or compare it; the AI reads the hypothesis' full content before answering rather than working from memory. When nothing matches, the AI says so instead of guessing.