Your first session
Build a first workbook from workspace context and a sourcing plan.
Your first session turns a new workspace into two inspectable things: Context that explains the business and a small, scored worksheet of the right people. It is for a founder or GTM operator who needs evidence that Dreamstate understands the business before they invest in outreach, publishing, or a larger data run. This is the day-one play. It connects Context with workbooks and worksheets, and it ends by recommending a later play rather than starting one.
The first-win skill keeps this session narrow: one accurate business description and one well-read sample beat a tour, a large list, or contact data. Architect reads the named skills, finds the needed tables and knowledge capabilities through the catalog, and follows their result state. The visible outputs are saved Context pages and a worksheet you can open from Workbooks.
Needs and produces
Begin with the company website and the workspace you intend to use. If more than one workspace is available, establish which one is the target before writing or running anything. The first session can still build Context and the worksheet when a later account connection is missing. That connection only blocks the run or outward action that actually needs it.
The play produces revisioned Context pages, a workbook named for the ideal customer profile, one worksheet with a row source for the scarcest criterion, an evidence column, and a fit score. It also produces a small sample with read rows and score reasons. It does not produce a started sequence, published post, sent email, DM, or contact-data lookup.
Build the business context first
Architect uses the knowledge skill to inspect the saved business information and determine which core pages are empty or incomplete. When the workspace needs the foundation, it creates or updates Product Information, Ideal Customer, Competitor Analysis, and Tone of Voice from the company website and competitor research, then reads those pages back. Context is the shared knowledge store, so later list qualification, message writing, and social drafts can refer to the same persisted facts.
Ask one question only when two readings of the ideal customer profile would return materially different people. Otherwise, state the interpretation in plain language and build from it. The useful check in the app is not that an update was requested. It is that the Context pages show their saved content and revisions, with the Ideal Customer page ready to ground the sample.
Build and read one small worksheet
Architect then loads the list-building skill. It chooses a row source that carries the scarcest qualification criterion, attaches that source to a single worksheet, and exposes the source fields a reviewer needs to see. A list lives in a workbook: one worksheet has one row source, columns enrich and judge its rows, and a score is optional. Nothing searches or previews people outside the worksheet.
The sample begins at ten to twenty rows in the first-win practice. State the row count and estimate, run the sample, wait for the run to finish, and read real rows rather than treating a returned total as proof. Add one evidence column and a fit score, but do not add contact data. The score should explain why a row does or does not fit. When a required answer is missing, it is unknown rather than false.
Review the top five rows with their score reasons in Workbooks. If fewer than half look right, adjust the source filter and run another small sample. Do not compensate by adding more columns to a weak source. A missing account does not prevent worksheet construction, but Architect should identify the account needed before a future provider-backed run.
- 1
Confirm identity and workspace
Architect confirms the active user and workspace, then uses theknowledgeskill to read the saved business facts before using them. - 2
Read and save Context
Create only the Context pages supported by research, then read their revisions back. Check the Context screen, especially the Ideal Customer page, before sourcing rows. - 3
Attach one source
Use thelist-buildingskill to choose a source for the hardest criterion and attach it to one worksheet. The workbook is the visible boundary for the list. - 4
Run a small scored sample
Add one evidence column and a fit score, state the sample size and estimate, then run the sample. Read the rows and score reasons after the run completes. - 5
Recommend, do not launch
Use theplaysskill to select three plays from the workspace state. Recommend one and name its prerequisites, but do not start it for the user.
Credits, leading metric, and next play
The leading metric is the proportion of sampled rows that look right when read against the saved ideal customer profile. The completion report names rows landed, rows that fit, top scored examples and reasons, credits charged, the source and filters, and the evidence column and score rule. It also identifies any account still to connect. This turns the result into a decision, rather than a generic onboarding summary.
The plays skill selects a next play from actual workspace conditions. Warm network first fits when LinkedIn is connected. A 50-lead outbound pilot fits when a sender exists. A founder posting system fits when Tone of Voice is strong. The first session should explain why one option matches the state and what it will produce, not automatically run an unrelated motion.
When it goes wrong
When the ideal customer profile has two plausible readings that would land different people, ask the narrow business question before sourcing. When the Context page lacks support for a claim, leave the claim out and record the missing evidence instead of filling the page from a similar business. A missing connected account is a prerequisite to record, not a reason to abandon Context or worksheet authoring.
If the sample is weak, change the source filter and resample. If a cell is empty or failed, distinguish the two states, read the column and its upstream inputs, and use Find out why something failed before attempting a larger run. If a run is queued or still working, follow its status and do not report the worksheet as complete from a request or an incomplete count.
If the requested work has already produced a suitable Context page or worksheet, read that result, revise it only where evidence or the operator's request warrants a change, and use the existing object in the report. The goal is a trusted first result, not duplicate workspace objects.