What SoulMen doesn't do.
A pre-upload document readiness check, not a guarantee of portal acceptance. Read this before relying on its output for a real tender submission — the same limitations SoulMen's own product documentation states internally, published here rather than left for you to discover after the fact.
What it is not.
Not legal advice
Findings describe document completeness, expiry, and consistency — they are not a legal opinion on tender eligibility, licensing law, or contract terms.
Not an official submission channel
SoulMen does not submit anything to DEWA, Dubai Municipality, ADNOC, Aldar, Emaar, or any other portal. You still upload through the buyer's own system.
Not a guarantee of portal acceptance
A "compliant" result means the document pack matches SoulMen's checklist for that buyer as currently configured — not that the buyer's portal will accept it. Portals can enforce additional, changed, or undocumented requirements at any time.
Not a scope/eligibility assessor
It does not evaluate activity-code eligibility, project experience adequacy, technical staff sufficiency, equipment lists, or bid scoring.
The built-in portal checklists, plainly.
Buyer checklists (DEWA, Dubai Municipality, ADNOC, Aldar, Emaar, and others) are seeded from a best-effort reading of publicly available procurement/supplier pages at a point in time — not a live feed from the buyer. Each requirement shows a confidence rating (verified against a cited official source, inferred from credible secondary sources, or unverified and carried over from an earlier pass) and the date it was last retrieved. Government portals change requirements without notice; a requirement accurate on its retrieval date may no longer be. See the full requirements dataset for the current rating on every tracked requirement.
Always cross-check the current buyer portal/checklist before final submission. If you upload your own client requirement pack, SoulMen uses only what's in that document — it does not fall back to or blend in the seeded buyer checklist.
What the AI model actually does, and where it can fail.
Document classification and field extraction (document type, expiry date, company name, license number) use a multimodal AI model. That output is treated as a hypothesis, not a fact — it's passed through deterministic validation (format checks, expiry-date math, fuzzy name matching) before it can appear as a finding. But extraction can still miss a field, misread a date, or misclassify a document, especially on low-quality scans. Low-confidence extractions are flagged for manual review rather than silently accepted — review those manually before treating a document as verified.
Very large documents.
Large PDFs — scanned, or digital with a text layer past the truncation cap — are processed in page chunks (60 pages/chunk, up to 10 chunks by default). Beyond that configured chunk ceiling (600 pages by default), only the first chunks are analyzed and the document is flagged as partially covered.