Build a useful rejected-vessel explanation for a US desk
An exclusion should help a broker decide what to do
A useful rejected-vessel explanation identifies the requirement the candidate failed, the evidence used, and the change that would justify another look. It should not merely say low match score. For a US chartering desk, a rejection is valuable only if a reviewer can distinguish a genuine commercial mismatch from a missing answer, an outdated input, or a mistaken interpretation.
This matters because a shortlist hides more than it shows. The desk sees the surviving candidates but may never notice a viable ship excluded earlier. A persuasive explanation beside the winners cannot reveal that lost opportunity. Rejected candidates deserve a review route, particularly when the exclusion depends on an uncertain field or an assumption about the customer's flexibility.
The aim is not to make every user inspect every discarded record. It is to direct attention to exclusions that are consequential and reversible. A clear reason can tell a broker whether to make a call, correct a requirement, or move on.
Explain the actual process
NIST's explainable AI principles distinguish understandable explanations from explanations that accurately reflect how an output was produced. NISTIR 8312 abstract For a chartering application, that distinction means a polished commercial story is insufficient if it does not describe the real basis for the exclusion.
Suppose a system excluded a candidate because no equipment value was present. It should not explain that the vessel lacks the required equipment. Those are different statements. The first describes missing information; the second makes a factual claim about the vessel. An explanation that substitutes the second for the first can mislead the reviewer even if the system's developers intended to be helpful.
NIST describes its AI Risk Management Framework as voluntary guidance that includes evaluation considerations. NIST AI Risk Management Framework The commercial method below is original analysis, not a NIST certification scheme or an assertion that adopting it establishes compliance with any external standard.
A hypothetical New York desk loses a candidate
Imagine a hypothetical New York chartering desk seeking a vessel for an import movement. The customer's enquiry prefers shipboard handling equipment but permits a shore-equipment alternative if the receiver confirms it. Candidate Delta has no equipment value in the imported record. The assistant excludes Delta and explains that it does not meet the handling requirement. No actual vessel or customer is represented.
The explanation contains two errors. It treats an unknown value as an adverse fact, and it treats a preference with an allowed alternative as an absolute requirement. The desk may never contact Delta's operator or ask the receiver about the alternative. The search appears efficient because it has removed work, but the removed work may have contained a viable commercial option.
A better explanation says that equipment information is missing and that the candidate was held out pending clarification under the current screening setting. It also identifies the customer's permitted shore-equipment alternative. The reviewer can then decide whether to request particulars, confirm the alternative, or retain the exclusion.
If the customer later changes the instruction to require shipboard equipment without exception, the explanation should change accordingly. Delta still lacks evidence, but the permitted alternative has disappeared. The reviewer now understands both the missing fact and the stricter commercial boundary.
Separate four kinds of rejection
A confirmed mismatch means the available evidence conflicts with an applicable requirement. An information gap means the desk cannot yet determine whether the candidate meets it. A commercial decline means a relevant party has declined the proposed business on its stated basis. A search limitation means the system could not assess the candidate using its available inputs or methods.
These categories should lead to different actions. A confirmed mismatch may close the candidate unless the requirement changes. An information gap may justify a targeted question. A decline may close one enquiry without defining the vessel's general availability. A search limitation may require a human to assess the record outside the automated process.
The labels should remain subordinate to the explanation. A desk should be able to read the specific reason without memorizing a taxonomy. The practical sentence is what matters: the proposed loading window falls outside the operator's stated indication, or the record does not identify the relevant handling arrangement. Specificity makes review possible.
Show the boundary without inventing permission
A helpful explanation can state what change would alter the result, but it must distinguish a hypothetical change from an authorized one. If a vessel would become relevant with a later loading window, the assistant can identify that dependency. It should not suggest that the customer has already accepted the later window.
This is especially important in customer-facing summaries. An internal reviewer may explore a range of alternatives, while an external message should accurately represent the customer's current instruction. The assistant should avoid turning an explanation of the decision boundary into an unsolicited revision of the cargo enquiry.
Where several reasons apply, show the decisive ones first. A long list of minor issues can obscure a single requirement that makes further work pointless. Conversely, a strong-sounding primary reason should not conceal that it rests on a weak or stale source. The reviewer needs the evidence quality as well as the conclusion.
Test explanations by changing the evidence
An evaluation can start with a fictional rejected candidate and change only the input responsible for the exclusion. If supplying the missing equipment value does not change the explanation, the displayed reason may not reflect the process. If changing an unrelated field reverses the decision, the team needs to understand why.
Ask a broker to predict what the system will do after each change. Large gaps between that expectation and the actual result reveal where the explanation is unhelpful. This test examines operational usefulness, not merely whether the prose sounds reasonable.
Zeaclub's founder-described YC inspiration can inform a focused learning exercise here: investigate a small number of lost candidates before attempting a broad claim about matching accuracy. This is a proposed evaluation agenda for an AI-native direction, not a statement that Zeaclub already delivers every described function or has YC affiliation.
Bring a sanitized exclusion that a colleague later reversed. Identify the missing fact, the misunderstood requirement, or the changed commercial instruction. A productive Zeaclub discussion should turn that case into a better explanation and a clear review action, helping the desk recover opportunities without treating every rejection as an error.
Sources
- NISTIR 8312, Four Principles of Explainable Artificial Intelligence, inspected October 23, 2026.
- NIST AI Risk Management Framework, inspected October 23, 2026.
Related articles
- 0600: Keep minimum and maximum capacities distinct in US AI searches
- 0353: US Dry Bulk AI Matching: Keep Cargo Constraints Visible in the Shortlist
- 0642: Learn from Rejected Vessel Matches Without Blaming the Desk


