DIRECT ANSWER / 11
What makes this review deterministic, and why is there no model?
Every suggestion comes from documented numeric equality within the tolerance you set, every mapping is confirmed by a person, and every arithmetic trace shows its operands. The same inputs and the same decisions always produce the same output, and no language model is involved at any point.
Why this review exists
A reviewer has to be able to explain a finding to the person whose proposal it is. A score that cannot be traced back to an operand is not explainable, and in a funding decision that is not acceptable.
What deterministic means in practice
Given the same budget rows, the same narrative text, the same tolerance and the same human decisions, the tool produces the same claims, the same suggestions, the same traces and the same findings, every time and on every machine.
That is why the arithmetic traces show operands rather than only results. In the sample, the trace for the coordinator row reads twelve times 2000 equals 24,000 against an expected 24,000, so a reader can check it in their head.
Suggestions are equality within a tolerance and nothing else. There is no similarity score, no ranking, no confidence percentage and no partial match.
Why there is no model
A language model would produce plausible mappings, including for figures that do not correspond, and a reviewer would have no way to tell the two apart. In a proposal review that failure mode is worse than doing nothing.
It would also mean sending an unpublished proposal somewhere. The runtime makes no network call at all, which is a stronger privacy statement than any policy, and adding a model would remove it.
The tool therefore does the part that is mechanical and refuses the part that requires judgement, and it keeps the boundary between the two visible on every screen.
Practical sequence
- Read every suggestion as a numeric equality, not a judgement
- Confirm or reject each one yourself
- Read the operands on any trace you rely on
- Record your rationale so the next reader can follow it
What remains outside the app
Grantweft performs arithmetic and textual cross-reference on user-supplied material. It does not write proposals or determine eligibility, allowability, accounting treatment, compliance, completeness, accuracy, submission readiness or funding outcome.
The private desk keeps exact source labels, stable budget row identities, paragraph-based claims, confirmed mappings, visible arithmetic, findings, human decisions, review freezes and passive handoff files connected without an account or remote application database.