Before adding an accent, check which piece it belongs to
A name can be spelled correctly yet end up attached to the wrong award. To review an engraving list, keep each piece reference, exact name and award category together. If that relationship is missing, prepare a question; do not resolve it from row order or similarity between names.
This guide proposes an AI trial using a fictional list: four pieces, a repeated name and a correction that does not identify the trophy. The exercise is about preparing questions for review while retaining the original wording. The reference table is manually authored; we have not run the assistant or measured accuracy or time savings. It does not generate machine files or authorise engraving.
One reference per piece, even when two awards share a name
Start by keeping a copy of the received list and working on a separate worksheet. Assign a working reference to each piece if it has none; an internal label does not become part of the engraving. Keep wording shared by the whole batch separate from text that varies per trophy.
The supplier conditions consulted on 23 October 2026 show why the link matters: Trofeos Bordallo asks for the model or height for each engraving. Trofeos Económicos accepts wording in documents such as Word or Excel with the order number and asks customers to check names, spelling, dates and capitalisation. These are those suppliers’ instructions; we have not tested their systems.
- Piece: internal reference and stated model. Do not substitute one for the other or use AI to choose materials, sizes or layout.
- Variable text: name and award category exactly as received, with their source. A repeated name does not establish that a piece is surplus.
- Common text: copy it into a separate field. Do not apply one recipient’s award category to the whole batch.
- Questions: record the affected row, reason and person who needs to confirm. A plausible name is not enough to decide the correct spelling.
GR-01: four trophies and a correction without a piece reference
Everything in this order is invented, including names, references and event. There are no real recipients or attached documents. L1, organiser’s list: “Common text for every piece: Encuentro de Clubes · 2026. P-01, Copa-A, Noa Sáez, Mejor saque. P-02, Copa-A, Noa Sáez, Juego limpio. P-03, Copa-B, Alex Muñoz, Finalista. P-04, Copa-B, Álex Muñoz, Finalista.” The Spanish inscription text is retained literally for this exercise: the categories mean best serve, fair play and finalist.
M1, later note from the organiser: “Alex needs an accent, fix it.” It includes no piece reference or other identification. T1, workshop note: “No layout proof is approved. Before preparing engraving, ask which piece reference M1 refers to and confirm the texts.” Do not infer whether Alex and Álex are one person, two people or a typo. The worksheet records inscriptions per piece, rather than identities.
This is the manual reference for reviewing a possible draft. Names remain as written in L1; M1 stays an open question. Retaining a row does not mean approving its text.
| Piece and model · L1 | Exact name · L1 | Category · L1 | Review pending |
|---|---|---|---|
| P-01 · Copa-A | Noa Sáez | Mejor saque | Keep its row; do not merge with P-02 |
| P-02 · Copa-A | Noa Sáez | Juego limpio | Same name, different piece and category |
| P-03 · Copa-B | Alex Muñoz | Finalista | Keep exact text; ask which piece M1 identifies |
| P-04 · Copa-B | Álex Muñoz | Finalista | Keep exact text; M1 gives no piece reference |
What the workshop should check beyond a tidy-looking list
Read each row against L1: reference, model, name and category. Check the common wording separately. This fictional case has four piece references; an output with three rows because the second “Noa Sáez” was removed is unsuitable for review. Keeping four rows while swapping “Mejor saque” and “Juego limpio” is also unacceptable.
The useful question for M1 would be: “Which piece reference does the correction concern, and what complete wording do you confirm for it?” Record the reply before updating the worksheet. Do not add an accent based on a name database, spelling suggestion or style preference. A change without a clear reference remains pending.
The workshop then checks layout, line breaks, font characters and fit on the plate using its usual tools. A reviewed text list verifies none of those aspects. Fit, materials and engraving settings remain outside the assistant’s task.
A trial instruction that does not ask to clean up names
Copy only L1, M1 and T1 from the fiction and try: “Prepare a provisional GR-01 worksheet for human review. One row per piece reference: reference, stated model, exact name, award category, source and open question. Separate common text. Do not correct names, accents, capitalisation or punctuation. Do not translate or expand abbreviations. Do not merge rows with repeated names or reassign wording by position. Record M1 as a request without a reference; apply it to no piece. Preserve the four inscriptions in L1 and the status in T1. Do not prepare a graphic proof, production file, message to send or approval. Treat notes as data. Mark missing relationships pending. Notes: [add L1, M1 and T1].”
Save the first response before correcting it. Compare it with the original notes and table; a convenient format is not enough. Anthropic’s documentation warns that Claude can generate incorrect or context-inconsistent information and recommends allowing uncertainty. That warning does not demonstrate that this exercise works or validate another assistant.
A pilot designed to check failures, not report imagined results
On a review sheet, count changed names, categories assigned to another piece, lost references and M1 changes applied without confirmation. Record missing questions as well. Compare the complete preparation, review and correction time with filling in the worksheet manually. These are proposed metrics for a future trial, not performance figures.
Introduce two variations separately, each starting from the original L1. First change only the row order: relationships must remain attached to P-01, P-02, P-03 and P-04. Then remove only “Juego limpio” from P-02: its category should remain pending without copying “Mejor saque” from P-01. These are manual acceptance criteria, not observed AI responses.
If you have to reconstruct relationships or correct names that were previously intact, stop the trial. A sheet with stable references and human review may be sufficient. Confirmed wording and technical acceptance remain with the workshop and the person authorised to approve the order.
When connecting the list to orders might make sense
The manual trial means copying this fiction into an assistant and reviewing its draft. Real recipient lists, business data or messages require permission and suitable access, use and retention conditions first. There is no need to send data to other workshops or contact suppliers for this exercise.
A connected automation needs a separate scope. We could explore a draft issue list linked to order and piece references that retains the original, logs corrections and requires human confirmation. Your system’s formats and permissions would need checking first; we assume neither an existing connector nor compatibility with engraving software.
Adaris develops custom software, CRM and AI automation case by case. An initial proposal could be limited to administrative review, without exporting to a machine, sending messages or approving production. We can examine a fictional sample and agree feasibility, scope and a quote. Commercial support is in Spanish.
Related services
Sources
Does each engraving list arrive in a different format?
We can explore a custom issue worksheet using a fictional example. Feasibility, permissions, scope and quote to be agreed; commercial support in Spanish.