DataMosaicMosaic · How it works
How Mosaic finds the reason.
A comparison stops at two numbers that differ. Mosaic keeps going: it lines your systems up every night, follows each difference back to the reason, and fixes what you have already taught it to fix.
01 · Connect
Connect the systems you already run.
A data source is one question put to one system: a SQL query against the tills, a REST call to the ERP, the file the warehouse drops at 02:00. Mosaic reads them where they are, from inside your network, and a data source never writes back.
Data sources
One query, every store. Where each shop runs its own till database, the query is written once and Mosaic runs it on every server — from a list, or from the hostnames on your business locations — with each row labelled by the store it came from. Open a shop in Mosaic and every check reads it.
Describe the rows, get the query. Say what you want — net sales per store and day — and Mosaic finds the tables, looks at the real values before it filters on them, writes the query and runs it. Every query it ran is listed under the answer.
Read-only, enforced. A statement that isn't a read is refused as it's typed and again on the server, and each run asks the database for a read-only session.
In the docs: Data sources · Running across many machines · A read-only login
02 · Understand
Teach it what your columns really mean.
The hard part is rarely the comparison. It is that two systems disagree about what a
value means — S014 on one side and 014 on the other, or a field
that was repurposed ten years ago and never renamed.
Normalise
S014
014
Drop first characters · 1
1240000
12,400.00
Multiply by · 0.01
What the column means
CUST_GENDERcustomer gender
- RETAIL
- WEB
- PARTNER
- OUTLET
MeansSales channel — read from its values, not its name
Normalise makes two systems read the same way before anything is compared: trim a prefix, pad a code, turn cents into units, move a clock an hour. It changes how a value is read, never the system it came from.
Meanings. Mosaic reads every table and column it can see and proposes what each one means — from its values, not just its name. Nothing counts until somebody agrees. Once agreed, every check built afterwards uses it.
That is the knowledge that usually lives in one person's head. After a few months, Mosaic knows how your business is wired.
In the docs: Normalise
03 · Check
Build a check: what to match, what to compare.
Choose what identifies a row — store and business date — and what gets added up — net sales, receipts, tax. Mosaic groups each side the same way, lines them up, and shows every row that differs, by how much, and which side is missing it.
Rules decide whether tonight passed.
A check with no rules shows the differences. Rules let it judge them, so that at 03:00 a run can fail without anybody reading it. Each rule is four fields and nothing more.
Add Rows compared, at least 1 as well: a query that quietly returns nothing has no differences, and would pass every other rule.
| What is counted | Must be | Threshold | On breach |
|---|---|---|---|
| Rows that differ | at most | 0 | Fail the run |
| Rows missing from a source | at most | 0 | Fail the run |
Last night, 00:452 of 2 rules failed
In the docs: Aggregations · Rules · Your first check
04 · Run
Run it every night, on every store.
A trigger runs the check on a schedule, in your workspace's own time zone, and sends the result where it needs to go. Nobody has to remember to look.
- Open
- Investigating
- Expected
- Resolved
- Ignored
In the docs: Triggers and schedules · Outputs · Exceptions
05 · Why
Follow a difference to its reason.
This is the step a comparison cannot take. An investigation is the runbook somebody would follow at seven in the morning, written down once: each step asks a system what it saw, a question splits the path on the answer, and it ends on a conclusion.
short in the ERP for ST-014
receipts rang at the tills
reached the integration layer — the whole day was delivered
refused by the ERP
A line with no tax codeT-014-0475 T-014-0480 · 480.00 between them
Steps query a system, run another check, or hand the row to a model that reads your systems with read-only queries. Questions and switches route on the answer. A conclusion ends it: answered, ruled out, or inconclusive.
On every disagreeing row. Switch it on and the investigation runs as part of the check, writing where it ended into a column at the end of the table — so the morning starts with the reason, not the difference.
Evidence, not guesses. Every answer shows the queries behind it. When the data cannot tell, the conclusion says so instead of inventing a cause.
Not every row needs a runbook. A drilldown opens the records behind any row — matching or not — with one query bound to it: the receipts behind a store's total, the lines behind a receipt.
In the docs: Investigations · Agents
06 · Act
Then do something about it, before anybody has to.
A workflow is a chain of steps on a canvas. It starts when a check finishes, when somebody sends it rows from a result, on a schedule, from a webhook or a call to the API — and then does what a person would have done at nine. You build every step; it does those and nothing else, and no AI decides what changes.
- Query a system
- Run a check
- HTTP request
- If and switch
- Filter, sort, sum
- Send an email
- Post to a channel
- Ask a model
Every execution is kept, step by step, with what went in and what came out. On your own server a workflow can also run a script or work with files — off until whoever runs the server switches it on.
07 · Nine o'clock
At nine, the answer is waiting.
The Overview is what you open first: how many checks agree, which rows are still out, and how last night's runs went against the fortnight before. It counts what disagrees; it doesn't chart your sales.
Your board. Each workspace arranges its own tiles — a check, a workflow, a monitor, or one figure set against the period before.
Monitors answer the question before the question: is each store's server answering at all? A workflow reads the hostnames from your business locations, knocks on each one, and the monitor shows who answered.
Store servers
Build one check and see what it finds.
Every feature is free while DataMosaic is in beta.