DIRECT ANSWER / 08
How are the budget table and the narrative brought in?
The budget arrives as strict delimited text that is previewed and column-mapped before anything is committed, with formula-looking cells rejected. The narrative arrives as plain text or Markdown, split into paragraphs, and bounded numeric candidates are extracted from it. Both carry an exact version label.
Why this review exists
Version identity and a refusal to parse rich formats are what keep a review traceable. A finding against Budget v7 is useless if nobody can say which file v7 was.
The budget table
The table is previewed first: headers, a row sample and a row count, with nothing written into the review. Required columns must be mapped explicitly, though the mapper proposes a mapping from familiar header names, and cells that carry spreadsheet formula intent are rejected rather than evaluated.
The sample's normalised export uses the header names id, source row, category, description, role, period, quantity, unit rate, requested, other funds, match in kind and note, and re-importing that file produces the same four rows with the same values, which is the round trip the worked example offers as a download.
Each imported row keeps the source row number it came from, so a finding can point at a line in the file the reviewer actually has open.
The narrative and the source identity
Narrative text is accepted as plain text or Markdown and split into paragraphs, which become the unit that claims belong to. Rich formats are refused because a tolerant parser for them is the most likely place for a silent misreading to originate.
Extraction proposes bounded numeric candidates with their raw text, their kind, their numeric value, their character offsets and the surrounding sentence. The sample's three paragraphs produce eight such claims.
Source identity is recorded on the brief: organisation, opportunity, reference, currency, project period, requested amount, and the two version labels. That block is what makes the review reproducible and what appears at the top of the handoff packet.
Practical sequence
- Record the exact budget and narrative version labels
- Preview the delimited table and check the headers
- Map every required column explicitly
- Paste the narrative as plain text or Markdown
- Review the extracted claims before mapping any of 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.