# Getting realistic results Misata's data holds up for two different reasons, and it helps to know which is which. - **What you state is exact.** Every number you give (a monthly total, a balance, a rate, a share, an average) is held in the rows and read back in SQL on the certificate. If one cannot hold, it is listed as not held, with why. It is never bent quietly. - **Everything else follows the business.** Orders cluster in the evening and in November, a few customers buy most of the volume, tickets are solved in office hours, a subscription never starts after its account churned, a refund follows its payment. These come from the business's own mechanics, not from a number you gave. So the way to realistic data is: **state the few numbers your story turns on, and let the business supply the rest.** ## 1. Choose the right way in | Your situation | Best way in | Why | |---|---|---| | A business Misata knows: B2B SaaS, e-commerce, banking and payments, a health system | Describe it in words | A tested built-in shape with its full answer key: the standard metrics, their SQL and their usual misreadings | | Schools, AML, machines and sensors, clinical trials, CRO finance | Describe it in words | Built-in shapes with a smaller answer key | | Your own tables (your product's schema, a customer's warehouse) | Import the structure, then state numbers on your columns | Your column names and types are kept exactly; the numbers are held on them | | Any other business | Describe it in words, or draw the tables | Designed from your description; read the plan carefully before building | The built-in shapes are the strongest: their stories, metrics and mechanics have been checked against what each domain's analysts look at first. ## 2. State the numbers that matter Write them the way you would say them to a colleague. Each of these is held exactly: | Kind | Examples | |---|---| | A total by month | "revenue from $40k to $95k over 2025", "about 8,500 orders in 2025", "signups double in March" | | A balance or level | "MRR grows from $50,000 to $200,000", "a loan book from $40M to $65M", "300 active subscriptions" | | A rate | "2% monthly churn", "refunds 4% of orders", "defaults at 1.5%", "fraud 0.3% of transactions" | | A rate that changes for a while | "2% churn, spiking to 5% in Q3 2025", "a card-testing attack in May" | | A share | "enterprise is 45% of MRR", "channels 50/35/15", "electronics 40% of revenue" | | An average | "average order value around $85", "CSAT from 4.1 to 4.4" | | A window | "over the last two years", "for 2025", "January 2024 to June 2026" | Tips that make the difference: - **Give the window.** "Over the last two years" or "for 2025" sets which months the data covers. Without one, a sensible default is chosen and shown on the plan. - **Name the months a spike covers.** "In Q3 2025" or "from July to September" is held month by month; "sometimes" is not a number. - **Keep the numbers consistent with each other.** MRR of $200,000 from 300 accounts means about $670 per account a month. If your other numbers say $50 plans, one of them cannot hold, and the plan screen says which. - **Leave out what you do not care about.** Every stated number narrows what the data can do. Three or four numbers that carry the story beat fifteen that fight each other. - **Rates round to whole rows.** 2% of 4,824 rows is 96 or 97 rows, so 1.99% or 2.01%. The certificate shows the figure reached. More rows give finer rates. ## 3. Read the plan before you build Nothing is made until you press Generate. The plan shows: - each table, its columns and roughly how many rows it will have; - every number Misata will hold, in your words, and anything it could not read as a number; - the window, and the story's events (a spike, a campaign, a dip). If a number is missing or misread, fix it there. The story editor on the plan takes the months, rates and events of a built-in shape directly ("50k", "1.2M" and "5%" all read as numbers). ## 4. Read what comes back Each dataset opens on its **Overview**. The parts to look at: - **The verdict.** "Verified" means every number you stated was met, every declared rule held, and no critical or major finding is open. Anything not met is listed by name. - **The Story tab.** Each stated number beside what the data holds, by month, with the SQL that reads it, then the standard metrics and the identities that tie them (billed revenue equals MRR, for example). - **Patterns.** The realism notes: what the data shows about seasons, hours, concentration and repeat behaviour, with the SQL behind each. These are observations, not guarantees, and they do not change the verdict. - **Explore.** Charts, profiles of every column and your own SQL over the tables. - **Why a build says "degraded".** If a model step failed, content was written by rules instead, or a number was withdrawn, the dataset says which. Rebuilding usually clears a failed model step. ## 5. Change it | You want to | Do this | |---|---| | Fix something in words ("far fewer accounts", "refunds rarer in December") | Overview: *What is wrong with it?* | | Set a monthly total by drawing it | The **Curve** tab | | Turn a cause up or down, change staffing, plant more or fewer problems | The **Adjust** tab (rebuilt by code in seconds, no model call) | | Change a built-in story's months, rates or events | The story editor, then rebuild | | The same data again, byte for byte | Rebuild with the same seed | | A different sample of the same story | A variation (a new seed) | Every change makes a new version, checked like the first. The first is never overwritten, and every version stays open in Studio. ## 6. Common mistakes - **A rate on very few rows.** 5% of 40 accounts is 2 accounts. Ask for more rows, or expect whole-row steps. - **A drawn curve for a rate.** Curves set monthly totals. A rate that changes over time is stated in words ("2%, spiking to 5% in Q3"). - **A size and a total that disagree.** 100 orders cannot make $1M a month at $40 each. The plan says when sizes are raised to fit a number; read it. - **Expecting live text.** Reviews, notes and ticket bodies are assembled from lines written for your business. In very large tables the same lines return. - **Reading a realism note as a failure.** Notes describe the data. Only the verdict counts what you asked for. ## A checklist before you show the data 1. The verdict reads Verified, or each item not met is one you can explain. 2. The Story tab's stated months match the story you will tell. 3. The window ends where you want it (the last month is the one your dashboard opens on). 4. You have run the three or four queries your dashboard runs, in Explore or the checker, and they read as you expect. 5. You have the answer key (Export, or the Story tab's download) if anyone will check numbers later.