Proposal Evaluation

What a scorecard already does well.

Most of an evaluation is countable. Intake counts, response completeness, deadlines, raw scores tabulated. A scorecard handles all of it, and so does a general-purpose model. We concede every bit of it, without hedging. Nexonomy is not built to win the tabulation a spreadsheet already does.

The question with no correct weighting.

The award is the other kind of question. Which proposal wins when the cheapest bid is the riskiest and the strongest capability is the most expensive, and the criteria share no common unit. There is no correct weighting, and three evaluators will score the same response three ways. The output is not a total. It is an award someone signs and may have to defend to a challenger.

If summing the scores settled it, you would not need a committee. The award is not that sum.

One award, captured.

Nexonomy captured the evaluation as a typed, content-addressed Decision Record, sealed the moment it was saved: the criteria and their weights, each supplier's scores and the evidence behind them, the evaluator disagreement, the override of the top raw score and its rationale, the sign-offs, and the hash. One record, there on the day it matters. It runs watch-only inside your own environment, on the models you already trust, and nothing leaves the tenant.

How the call gets made.

Behind that record is the part a scorecard does not have. Separate from the evaluation committee, several frontier models weigh the same proposals in parallel, each one trading price, capability, and risk against the others. Where they agree, confidence rises. Where they disagree, the split is surfaced and kept rather than averaged into one number, so the recommendation that reaches the signer carries the dissent with it.

Everything that went into the call.

• The weighting, on the record

  Why each criterion weighed what it did, when no shared unit exists.

• Evaluator drift, kept

  Where three evaluators diverged, recorded as signal, not averaged away.

• The override, justified

  When the top raw score was passed over, the reason is on the record, not off it.

• Replayable under challenge

  Criteria, weights, evidence, and sign-offs, reconstructed if the award is protested.

When the award is challenged.

Six months on, a losing bidder files a protest: the award was arbitrary, the override unexplained. The committee has dispersed and the evaluation lives across a shared drive and a dozen inboxes. On a normal stack, defending it means reassembling who scored what, and why the top raw score was passed over, from people's memories, weeks of work with the award frozen. If the rationale cannot be produced, the award is vacated and the procurement starts over.

Nexonomy produces it in one lookup, because it never came apart. The award was reached by the model panel reconciling its own split, and the evaluators', against criteria with no shared unit, and that reasoning was sealed as a typed, content-addressed object: the criteria and weights, each supplier's scores and evidence, the model and evaluator disagreement, the override of the top raw score and its rationale, and the sign-offs.

Under challenge, retrieval is a lookup: the weights, the dissent, and the override rationale exactly as captured, addressed by the hash of their canonical content so any copy verifies against the original. The defense is not a claim that the file is unalterable; it is that the complete reasoning was captured when the award was made and reconstructs exactly, which is what a protest cannot break. The scoring store is the commodity part. What is not, and what re-fits every model cycle, is the parallel advisory and the calibration behind it, kept behind a schema stable enough that an award from two years ago still replays.