LOULOU / FIELD NOTES
Four real work moments.
Four decisions worth keeping.
No client or supplier identities, and no invented wins. These notes retain the problem, the checks actually made, the correction and the action you can reuse next time.
Start with the first noteFROM A REAL WORK RECORD · DE-IDENTIFIED
Catalog claims: “all models” still needs item-level proof.
In an artificial foliage selection, an overall capability statement did not fully match individual product pages.
The situation
The proposal needed to distinguish outdoor and fire-performance requirements by product. A broad supplier statement risked turning unconfirmed options into claims of compliance.
What was actually checked
I mapped quoted products one by one to their public product pages, checked the matching model and usage wording, and compared those statements with supplier material. The record showed three different states: both capabilities stated together, a choice between variants, and only one capability stated. Another item could not be matched in the public catalog. The client-facing selection and reference codes were then revised.
Separate “the supplier may offer a suitable variant” from “this exact model is confirmed.” Variant choices and contradictions need item-level written confirmation; a product page does not replace the applicable test report.
FROM A REAL WORK RECORD · DE-IDENTIFIED
One-stop sourcing: separate processes before shortlisting.
A multi-style apparel request crossed knit and woven construction. “One factory for all” was a condition to test, not a conclusion to assume.
The situation
Searching broadly for “garment factories” made many candidates appear suitable, even though knitwear and woven outerwear differ in equipment, pattern work and trims. A unified quote table could also hide differences in fabric weight, decoration and quantity tiers.
What was actually checked
The historical work separated styles by process, then used trade records, company pages and RFQ preparation material to shortlist candidates. The comparison kept quantity tiers and FOB terms visible, alongside non-capabilities and open questions.
A trade description saying “jacket” still needs clarification: is it a knit shirt-jacket or the required woven construction? Do not force unlike processes into one production story or treat the midpoint of a quote range as a confirmed price.
FROM A REAL WORK RECORD · DE-IDENTIFIED
STEP/BOM cost review: a large gap first calls for scope checks.
An engineered component estimate differed markedly from a factory quote. Louis challenged it; the model revisited units, finishing and omissions, then acknowledged that a major cost assumption lacked evidence.
The situation
AI produced a plausible materials, processing and total-cost range, but the quotation used a different unit basis and a different finishing stack. Treating the gap as excess margin would confuse unlike scopes.
What was actually checked
The review reconciled quote units with the system total, then checked material, welding, finishing, fasteners, packing and freight boundaries. The decisive challenge was a high-impact finishing allowance: the record did not contain an actual processing quote strong enough to support it.
Mark each cost item as original evidence, market-range estimate, or supplier quote still needed. If the largest variable remains estimated, the should-cost can guide questions but not anchor a confident negotiation.
FROM A REAL WORK RECORD · DE-IDENTIFIED
Supplier due diligence: facts, leads and inference are different.
A supplier comparison produced company reports, raw-source records and a risk sheet; the review also found some earlier interpretation too strong.
The situation
Registry pages, marketplace stores, trade records and websites can point to different names or entities. A low insured-employee count or import/export business scope can too easily become an unsupported conclusion about factory size or export performance.
What was actually checked
The historical work recorded entity fields, former names, operating status, affiliations and unresolved risk checks for four candidates. Raw records retained source and retrieval time; unanswered court or penalty checks remained tasks, not zero-risk labels.
Store “not checked,” “not found in the source,” “record found,” and “human assessment” separately. An old affiliate issue does not automatically become a current-entity risk; similar names do not automatically identify the same factory.
The reusable part is the review path, not the answer.
Source → claimWhich exact claim does a page, trade description or email support?
Gap → questionTurn a missing model, term or price basis into a specific question.
Correction → ruleMake the mistake caught by a person into a check for the next task.
This page is rewritten from historical work records. Identities, transaction details and distinctive product combinations are removed. Stages reflect those records, not current supplier status.