The full power of TDengine, now free forever for up to 5,000 tags.

Explore More

How Does Claude Data-Consolidate Join Manufacturing Exports?

Juno Qiu

November 18, 2026 /

The data-consolidate Skill joins available manufacturing exports and connected records into an XLSX working dataset and produces an HTML data-quality report. It matches records using identifiers such as part, production work order, sales order, lot, asset and supplier. The report classifies mismatches and proposes corrections for review; it does not make an uncertain match reliable merely by placing it in a workbook.

The Skill instructions describe file-only and connector-only operation. Files are a designed input, not proof that every plant export already matches the expected structure. The dataset contract defines what the workbook carries and where it is authoritative.

Worked example: reconcile a production export

The following small records are fictional and illustrate the review process. They are not a required upload schema or captured Skill output. Use your plant’s actual export headers and let onboarding identify their meanings.

ERP production orders

production_order,part_number,asset_id
PO-104,PN-4471-B,CELL-03
PO-105,AX-4471-B,CELL-03

Production export

production_order,part_number,asset_id,good_qty
PO-104,PN_4471_B,CELL-03,120
PO-105,4471B,CELL-03,80

Provide the two files and ask:

Consolidate these ERP and production exports for the stated production day.
Show the original identifiers, normalization rules, unmatched or ambiguous
records, and proposed corrections. Do not apply changes without my review.

The join-key reference describes case and punctuation normalization, while preserving the original identifiers and logging the rules. In this example, PN_4471_B can be compared with PN-4471-B after punctuation normalization. Looking only at the part number, 4471B has two possible ERP candidates after a prefix is removed, so the documented normalization rule does not choose one. Here, PO-105 also points to AX-4471-B; if the production order is verified, it can provide additional evidence to distinguish the candidates. Check that order-to-part relationship with the plant owner, and do not assume the Skill automatically resolves the ambiguity from that field.

When reviewing the returned workbook, trace the PO-104 quantity of 120 to the production export, and check the order, part and asset identifiers against the ERP row. In the HTML report, look for the normalization log and the unresolved candidate list. Do not assume the exact screen wording or a successful match: those depend on the files and the actual run. Keep maintenance ticket IDs separate from production order IDs even when both systems call them “work orders.”

A mapped asset identifier also needs its physical context. See industrial data contextualization for how asset and measurement relationships make plant data interpretable.

Decide what goes into the first run

Start with a bounded production period and a small set of source files: production and downtime records, an ERP work-order export, a maintenance asset/work-order export, and a planning file if the report needs plan comparisons. The exact set depends on the downstream question. Retain each file’s source, extraction time, covered period, time zone and owner; ask the owner about late corrections.

Check join identifiers before starting. For instance, a maintenance export may call an asset P-104, while the production file calls it Pump 104. Those values are not automatically the same asset. Provide an owner-approved cross-reference if available, and examine the unmatched rows if it is not.

After installation and plant setup, ask Claude to “consolidate the exports” and identify the input folder or supplied files. Inspect both outputs: the workbook for the usable records and the quality report for unmatched keys and proposed corrections. Approve only changes supported by source evidence. A proposed correction is a review item, not authorization to change the ERP or CMMS.

Read the mismatches, not just the totals

The Skill describes eight mismatch classes and suggests a correction for each. For a missing work-order key, check whether the order is absent from the ERP extract, uses a different identifier, falls outside the extraction window, or has not yet been created. Each explanation calls for a different owner and remedy.

A successful-looking join can still be wrong. Reused IDs, site-specific numbering, inconsistent units and late-arriving records can all produce misleading results. Pick several matched rows and trace them to the underlying source records. Keep unmatched and ambiguous records visible to the report reviewer.

What the workbook does not replace

The consolidated dataset is a shared working contract, not the system of record for every detail. In particular, the quality investigation Skill reads raw historian and inspection exports for high-frequency traces and measurement calculations; identity is still resolved against the shared dataset. Keep the original files and their timestamps so an engineer can reproduce a finding.

If process history resides in TDengine, its SQL query documentation describes querying through application interfaces. A plant can evaluate a controlled export into the expected input format, including asset IDs, timestamps and units. This is an implementation option, not a documented turnkey mapping or a tested Claude–TDengine integration.

Before using the workbook for a production decision, compare source totals and representative joins, resolve consequential mismatches, and disclose records that remain unmatched. Re-run the checks when the source schema or extraction schedule changes.

Related Articles