There is a version of tender review software that reads a document set and returns a clean, confident list. No caveats, no flags, nothing to reconcile. It demonstrates well. It is also the version that loses bids, because the confidence is manufactured. A model that reads four hundred pages of instructions, conditions, addenda and appendices will be certain about most of it and genuinely unsure about some of it, and a product that flattens that distinction has thrown away the most useful thing it knows. ## The failure mode we are avoiding An extraction that is ninety-five percent reliable and presented as certain is more dangerous than one that is eighty percent reliable and marked. The first invites a bid team to stop checking. The second tells them exactly where checking pays. That is the whole design position. Offra is not trying to be the last reader of your tender. It is trying to make the human reading productive by pointing at the parts that need a human. ## 1. Every extraction carries its source Each requirement, date, mandatory document and legal clause Offra surfaces is linked back to the document and page it came from. This is not a citation for appearance. It is the mechanism by which a claim can be checked in seconds instead of minutes, and it changes the economics of verification. A list of two hundred requirements you cannot trace is a list you have to re-derive. The same list with a page reference on every line is one you can spot-check and then rely on. It also makes disagreement possible. If the extraction is wrong, the source is right there, and you can see immediately whether the model misread the page or the page genuinely says something surprising. ## 2. Confidence is shown rather than smoothed Extractions carry a confidence indicator, and the low-confidence ones are meant to be seen. Low confidence usually means something specific and legible: the clause was ambiguous, the scan was poor, the requirement was split across a page break, or the same obligation is stated twice in terms that do not quite match. Those are exactly the passages a human should read, and they are a small fraction of the document set. Showing this costs us a tidier interface. It buys the bid team a triage order, which is worth more in the seventy-two hours before a close than a clean-looking screen. ## 3. Conflict is a state, not an error When Offra checks a tender's requirements against your company profile, past projects and certifications, each requirement resolves to covered, partial, conflict or missing. Conflict is the interesting one. It means your own records and the tender disagree — a certification that expires before the closing date, a stated capability that contradicts what a past submission claimed, an insurance limit below what is demanded. Offra does not resolve those. It routes them to a person with both sides and the source attached, because a conflict between two documents is a judgment about your firm, and that judgment is not the software's to make. ## 4. What we do not do with your documents The benchmarks behind Opportunity Radar and Win Intelligence are built exclusively from public procurement records — SEAO, CanadaBuys, SAM.gov, USAspending and state and municipal portals. Client documents are never part of those figures and are never used to build our benchmarks. That is a boundary, not a current limitation: a firm's tender documents, pricing and past submissions are its competitive position, and a product that quietly pooled them would be selling one client's advantage to the next. ## Where the boundary sits Offra reads the tender set, extracts what it obliges you to do, cross-checks it against what it knows about your firm, and flags what is outstanding, uncertain or contradictory. It does that quickly and with a reference for every claim. What it does not do is decide. What to bid, what to price, what to sign, and which interpretation of an ambiguous clause to carry are decisions that belong to the people accountable for them, and the system is built to hand those decisions over with the evidence attached rather than to make them quietly. A tool that shows you where it is unsure is more useful than one that hides it. It is also the only kind you should be willing to put in front of a public tender. Related reading: [the five documents that sink a public tender](/blog/five-documents-that-sink-a-public-tender).