Unsaved suggestions at export

In the 16/10 alpha, 50 of 168 exported entries went to Suspense. 21 of them (₹9.80 lakh of ₹12.27 lakh) showed a suggestion in the Ledger box, but nobody saved it: 7 in Suggestions, 14 in Suspense. The Export step said only “50 to Suspense”. Choose an option below and try it in the mock. Use the keyboard or the mouse. Sample rows are based on the alpha imports (firm and ledger names only). Exact matches (78 entries) are left out of the tables.

Option

A · Check on Save B · List on Export recommended C · Signals only D · Export green ones

Sample data

Keys in the mock

Tab / Shift+Tab: step bar → Continue → table row → its Ledger box → export controls.
Step bar: ← → Home End; from the page Ctrl+[ / Ctrl+].
Rows: ↑ ↓ PageUp PageDown Home End; ↓ from the page enters the table.
Enter on a row opens its ledger list; Enter again takes the highlighted ledger (the suggestion, or Suspense).
Ctrl+Enter accepts the row’s suggestion. Typing (or /) searches ledgers.
Ctrl+Shift+Enter accepts every suggestion shown (Suggestions step; Export list in option B).
Escape closes the list, then goes back to the row; in a dialog it closes the dialog.
After a choice, focus goes to the next entry without a choice.

Comparison

Option	Stops the alpha mistake?	Pros	Cons	Effort
A · Check on Save	Yes at the last moment	Cannot be missed. Export looks as today when nothing is unsaved. One key (`Enter`) accepts all and saves.	A modal at the end. “Accept all” is blind: the user does not see the entries. “Show them” sends the user back to two steps for the alpha case. Red ones need a trip back.	Small ~1 day
B · List on Export	Yes before Save, in view	The user sees each entry, its ledger and its mark, worst first, and decides in place with the same keys as Suggestions. Both steps in one list. Not modal. Reuses `EntryTable` (items, `acceptAll`, `byConfidence`), so keys and focus rules come free.	The Export page gets long when many are unsaved. A second place to choose ledgers (but the same table).	Small–medium ~1–1.5 days
C · Signals only	Weak	Smallest change. Nothing in the way.	The alpha already showed “7 left” on Suggestions and it was ignored. A count is easy to skip. Overlaps with the “New suggestions in the Suspense step” thread (its line in Suspense counts the same rows).	Small ~0.5 day
D · Export green ones	Partly green only (13 of 21)	No user action for the sure ones. Yellow and red still go to Suspense.	Breaks the rule “recommendations are never saved by themselves”. A wrong green suggestion reaches Miracle without a look, and Suspense was at least visible. 8 of 21 still go to Suspense with no warning unless combined with A or B.	Medium ~1 day + rule change

Recommendation: B, List on Export.

It fixes the cause where it happens: at export the user sees which entries will go to Suspense although the app knows a ledger. The same table, keys and marks as the Suggestions step make it fast (Ctrl+Shift+Enter, then Enter on Save). Red ones sit on top, so they are seen before an accept-all. It needs no modal and no rule change. In the alpha case it brings Suspense from 50 to 29 entries with two keys.

Suspense-step accept-all and the “now suggested” line belong to the parallel thread for the Suspense step. B counts only what is still unsaved at export, so the two do not clash.