nwhitehouse lets reviewers overrule the AI without starting over

A new manual edit path means a single wrong extraction no longer forces you to re-run the whole tabular review.

workflowcontract-review

In tabular review, documents get lined up against a set of columns and the AI fills in each cell - a summary, its reasoning, a flag. Until now those cells were AI-only: if the model got one wrong, your only option was to re-run the entire review and hope it landed better the second time. nwhitehouse adds a direct way for a user to edit a single cell by hand and mark it settled, with an ownership check so only people on the project can touch it.

The same release also rebrands this fork from Mike to Finch - new name, logos, and on-screen wording - but that part is cosmetic and easy to take or leave. There's also a new per-document detail view to see how a single file breaks down across the review.

So what Worth a look for anyone running document-extraction grids who needs a human in the loop to correct mistakes cheaply, instead of paying for a full re-run.

View this fork on GitHub →

Spotted something wrong? Or know the PR text has fresher detail than the writeup above?

Commits in this thread

1 commit from nwhitehouse/mike, oldest first. Source extracted verbatim from the harvested git log.

SHA Subject Author Date
fac10f42 Rename to Finch, ONIT logo, random greeting, doc detail view Nick Whitehouse 2026-04-30 ↗ GitHub
commit body
- App-wide rename Mike -> Finch in user-visible strings, system
  prompts, and tracked-change author. Internal types, file paths,
  and the mikeApi client name are unchanged.
- Brand logo swapped to assets/ONIT_Mark_Dark.svg (copied to
  frontend/public/onit-mark-dark.svg). Sidebar, header, and the
  initial assistant landing all use the static mark; MikeIcon stays
  in place for spinner / done / error status indicators.
- Initial assistant greeting now picks one of the work___ legal
  greetings at random per mount and substitutes the user's name.
- New 3-pane "doc detail" view in tabular review (TRDocDetailView):
  click a document filename in the table to drill into a 25/25/50
  layout - column list / cell value & explanation / DocView preview.
  Doc viewer wrapper is a flex container so DocView's flex-1
  scroll region resolves a real height (otherwise mouse-wheel and
  programmatic scroll were both dead).
- Inline cell editing in the doc detail middle pane: pencil-driven
  edit mode with textareas, Cancel + Save buttons, and a new
  PATCH /tabular-review/:reviewId/cells endpoint that stores the
  manually edited content and sets status to "done".
- In-doc search in the right pane: input + N/M counter + prev/next
  arrows. Multi-match highlighter walks the pdf-text-layer spans
  and wraps every occurrence in <mark.finch-search-match> with
  rgba background so the canvas glyphs underneath stay visible.
  Current match shown in orange, others amber. DocView's empty-quote
  effect now clears stale citation highlights so the search and
  citation paths don't fight over the same text divs.
- Top-bar prev/next document arrows so you can scan the same column
  across documents without going back to the table.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

Capture this thread into my fork

Download a single Markdown prompt that tells Claude how to port every commit above into your working tree — adapting paths and structure to match your repo. Run it via claude -p < capture-thread-112.md from inside the repo you want the changes in.

⬇ Download capture-thread-112.md