Review & release

After Document ingestion runs, some of what the pipeline produced is certain and some is not. Review is where the uncertain part goes. It is one prioritized queue, not a second view of Rent roll & financial data.

Questions this workflow answers

• What did extraction get wrong, or get twice, or fail to find at all?
• Which of those actually block a decision, and which can wait?
• Where did this number come from, and has a human agreed with it?
• Is this deal's data safe to build a memo on yet?

What lands in the queue

Review collects several different kinds of doubt into one place:

• Conflicts. Two documents state different values for the same field. The queue shows both, with their sources, and asks you to pick.
• Clarifications. The pipeline extracted something it could not fully interpret.
• Low-confidence values. Extracted, but below the confidence the pipeline needs to assert them.
• Missing data. A field the deal needs that no document supplied.
• Schema review and financial verification. Structural and arithmetic checks on what was parsed.
• Unverified sources. A value in use whose source has not been confirmed.

Every item carries its provenance, so a decision is made against the document that produced it rather than against a bare number.

Working the queue

Items are ordered by what matters to a decision, not by when they arrived. For each one you can accept the extracted value, correct it, resolve a conflict in favor of one source, or dismiss the item as not relevant to this deal.

Resolutions are recorded, not just applied. The queue keeps a resolved history with undo, so a decision made quickly under time pressure can be revisited without re-deriving it.

A correction you make by hand is marked as yours. That distinction survives into the IC memo and every export: a figure a person verified reads differently from one a model proposed, and the record keeps them apart.

Release

Deal data has two states. Preliminary means the numbers are usable but unconfirmed. Released means the deal's data has cleared its gate and downstream work can cite it.

The gate is deterministic. It does not ask a model whether the deal looks ready. It checks specific conditions and reports exactly which ones are unmet, so "not releasable" always comes with the list of what to do about it.

A cross-tab banner shows release readiness from anywhere in the deal, so you do not have to open Review to learn that something is blocking.

What depends on this

Releasing is what makes the rest of the workspace trustworthy rather than merely populated:

• Valuation assumptions inherit the release basis of the data behind them.
• IC memo sections disclose whether they were written on preliminary or released data.
• Deliverables carry the same disclosure into anything you send outside the firm.

Correcting data after a memo is written does not silently rewrite the memo. The affected sections are marked stale so you can see what changed and regenerate deliberately.