Find and map your source

Choose a practical read method for spreadsheets, tables, databases, business applications and files.

You can start with the source you have. Your agent should use approved read access, inspect a small sample and identify which fields are authoritative. The routes below are implementation patterns to assess, not a list of supported Middleman connectors.

Where your data lives	Start with	What to check
Google Sheets	A scoped Sheets reader or Apps Script job	Workbook/tab, headers, stable IDs, calculated values, owner and refresh timing
Excel in Microsoft 365	Excel tables through Power Automate or Microsoft Graph	Workbook/table, pagination, file locks, tenant permissions and available licensing
Local Excel, LibreOffice or CSV	A local file reader on an approved scheduled runner	Exact path, who updates it, atomic export completion, encoding and workstation uptime
Airtable, Notion or another table tool	Its current API or an already approved connector	Stable row IDs, linked records, computed fields and API access limits
SQL database or warehouse	A read-only view and scheduled query	Company/row scope, a stable key, consistent snapshots, update and deletion semantics
ERP, MRP, inventory, accounting or PIM	The vendor’s supported API or scheduled report export	Installed version/modules, permissions, stock meaning, units, currencies and paging
E-commerce platform	Product and location-level inventory APIs	Product variants, locations, available versus reserved stock and event delivery
Shared folder, SFTP or object storage	A manifest-driven file reader	Approved location, completed files, checksums, duplicate files and access expiry
Legacy system with report emails	A structured report generated from that system	Fixed columns, source export time and an approved future intake address
Custom software	Its HTTP API, database view or a small adapter	Existing language, deployment environment, authentication and owner
Documents, PDFs or screenshots only	A bounded extraction for a reviewed local draft	Uncertain values, document dates, stable identity and how new versions will arrive

If the source cannot be read or exported repeatedly, prepare the draft and state the limitation. Do not promise a maintained feed from an inaccessible system. Browser automation may help with an approved export when no stable interface exists, but it needs a reliable runner and a recovery plan for login and UI changes.

Spreadsheet setup

Keep one header row and a stable record ID column. Preserve identifiers as text, including leading zeros. Avoid relying on row numbers: sorting or inserting rows changes them. If no key exists, propose a durable source column or sidecar mapping; do not silently rewrite the workbook.

Read a named table or deliberate range. Define whether hidden/filtered rows are included. Resolve merged cells, totals, blank rows, duplicate headings, formulas with errors, locale-specific decimals and date/time formats. Preserve typed values and their meanings; a display string alone may hide precision or a unit.

A workbook edited by employees can be the authoritative source. Middleman should receive fresh readings automatically, so those employees maintain only that workbook. A one-time file can bootstrap mapping but does not establish recurring updates.

For Google Sheets, installable time-driven Apps Script triggers can run recurring work under their creator’s account; timing may be approximate. Use an account whose ownership and access will continue. Google trigger documentation.

For Excel, Power Automate can run Office Scripts and pass their results to other flow actions. Use the flow’s supported HTTP action or connector for delivery; do not assume an Office Script can make external calls when run by Power Automate. Check table pagination and connector limits. Office Scripts with Power Automate, external-call restrictions, Excel connector.

Map meaning, not just column names

Record the source field, destination field, transformation, null behavior, evidence and refresh rule. A small illustrative map might look like this:

Source column	Meaning	Mapping decision
SKU	Supplier’s listing reference	Preserve as text; do not assume it is the manufacturer’s part number
Qty	An unspecified stock quantity	Confirm whether sellable, on-hand, allocated or incoming before mapping
Price	Amount without a stated basis	Establish currency, sale unit, tax basis, audience and quantity tier
Updated	A source timestamp	Confirm timezone and what event it records; unknown remains unknown
Hours free	Potential capacity	Require a resource, unit and real start/end window
Need by	Buyer delivery target	Preserve as a buyer requirement; do not turn it into supplier availability

Use the supplier mapping guide for its preview record schema. Buyer requirements use the buyer workflow; there is no public buyer-write schema in this release. Do not force demand into a supplier record.

When data spans several systems, pick an owner for each field and a verified join key. Do not add together mirrored inventory. Extract only the approved business fields, excluding unrelated people, credentials and customer-specific confidential terms. The supplier preview supports public-price context only.