DIRECT ANSWER / 03
What does a complete grant budget cross-check look like?
It looks like the bundled Harbor Youth Lab review: four budget rows totalling 46,500 requested, three narrative paragraphs producing eight extracted claims, five confirmed mappings, four agreeing arithmetic traces, no open findings, and a thirty-file handoff packet.
Why this review exists
Reading one complete review end to end is the fastest way to see which steps the software does and which steps stay with a person, because every figure below can be checked in the sample in a few minutes.
The sources being compared
The review is named Harbor Youth Lab proposal review, for the fictional North Quay Learning Cooperative, against the Harbor Futures Mini-Grant with reference HYF-26-07, in USD, for a project running through 2027. The requested amount recorded on the brief is 46,500.
Two version labels are the anchor for the whole review: Budget v7 dated 3 September 2026 and Narrative v11 of the same date. Without those labels a finding is untraceable a week later, because both documents will have moved.
The scope note on the brief says the review covers figures and stated periods only, and that funder instructions remain a person's responsibility. That is the honest limit of what follows.
Four rows, eight claims, five mappings
The budget holds four rows. B-0001 is Personnel, a program coordinator over twelve months at 2000 per unit, 24,000 requested, with 6,000 recorded as other organisational funds. B-0002 is Program, workshop materials for four cohorts at 2500, 10,000 requested. B-0003 is Access, participant transit support at ten units of 750, 7,500 requested. B-0004 is Evaluation, an external learning review at 5,000.
Three narrative paragraphs produce eight bounded claims: twelve months, 24,000 and 6,000 from the personnel paragraph; ten months, 10,000 and 7,500 from the program delivery paragraph; and 5,000 and 46,500 from the evaluation paragraph.
A reviewer confirmed five mappings by hand and marked the other three claims as contextual or rollup evidence rather than line-item claims. The claim for 46,500, for example, is the proposal total restated in prose, not a claim about any single row.
The arithmetic and the findings
Four arithmetic traces are recomputed and shown: twelve times 2000 equals 24,000, four times 2500 equals 10,000, ten times 750 equals 7,500, and one times 5000 equals 5,000. All four agree at the tolerance of 0.01 recorded in the review settings, and each trace shows its operands, its operator, the calculated value and the expected value side by side.
The rollup is 46,500 requested, 6,000 in other funds, nothing in-kind, and 52,500 total project cost. No indirect rate is declared in the sample, so the indirect observations are empty rather than assumed.
With those five confirmed mappings and three ignored claims recorded, the review reports no open findings and no blocking findings. Change one budget amount and a finding appears immediately, naming the row, the claim and the difference, and leaving the decision to a person.
Download the sample and read it back
The two files below are generated by the cross-checker's own export code from this same review. The normalised budget CSV is the file the handoff packet contains, with one row per budget line carrying id, source row, category, description, role, period, quantity, unit rate, requested, other funds, match in kind and note.
Feed it back through the tool's own budget preview and column mapper and all four rows return with the same categories, descriptions, quantities, unit rates and amounts. The header names are the ones the mapper's own guesser recognises, so the mapping requires no manual work on this file.
The checkpoint is the durable backup. It carries the whole review with a SHA-256 digest, and the parser rejects a file whose digest does not match its payload rather than restoring part of it.
Method and limits
Working through this example shows that the arithmetic is exact and reproducible and that every mapping is attributable to a person. It shows nothing about whether the costs are eligible, allowable, reasonable or correctly categorised for this funder.
The tool never scores a proposal, never interprets a rule, and never converts an unconfirmed suggestion into a mapping. Those boundaries are what make a quiet review readable rather than reassuring.
Sample input files
Both files below were written by the cross-checker's own export code from the sample review described above. The budget table re-imports through the tool's own preview and column mapper, and the checkpoint is verified against its digest before it restores.
Practical sequence
- Open the sample and read the review brief
- Follow each budget row to the claim it was mapped to
- Read the four arithmetic traces
- Download the files below and re-import them
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.