You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
[$250] Expense description reverts to old value after saving from the Reports/Expenses (Search) tab #101937
Version Number: Reproducible in staging?: Needs reproduction (Unable to reproduce) Reproducible in production?: Needs reproduction If this was caught during regression testing, add the test name, ID and link from BrowserStack: Email or phone of affected tester (no customers): Logs:https://stackoverflow.com/c/expensify/questions/4856 Expensify/Expensify Issue URL: Issue reported by:@Kel-Expensify Slack conversation (hyperlinked to channel name): #Expensify Bugs
Action Performed:
Go to the Reports (or Expenses) tab in New Expensify.
Open an expense from the list.
Click the Description field to open the full-page editor.
Type a new description and click Save.
Look at the expense's Description field.
Expected Result:
The Description field shows the newly saved text immediately.
Actual Result:
The page reloads and the Description field still shows the old text. In some cases the new description appears a few seconds later; in others it never appears until the page is refreshed or the expense is reopened from the Inbox.
Workaround:
Unknown
Platforms:
Select the officially supported platforms where the issue was reproduced:
changed the title [-]Tax amount change system message does not match the tax amount field when Taxes are disabled[/-][+]Expense description reverts to old value after saving from the Reports/Expenses (Search) tab[/+]on Sep 22, 2026
⚠️Heads up — the title and the body describe two different bugs. The title is about a tax-amount system message on tax-disabled workspaces; the body's repro steps are about the Description field not updating after save. I analyzed the body as the primary bug (it has the repro), and included a separate section on the title bug at the end. Please confirm which one this issue is meant to track.
I could not verify either one in a browser — browser control was unavailable for this run, so everything below is static code analysis.
What is the root cause of that problem?
Saving a Description edit writes the optimistic update to transactions_<id>, but the Reports/Expenses list renders from the Search snapshot key snapshot_<hash>. The save never patches the snapshot, because it never passes the hash argument that gates that write.
That matches the reported symptoms exactly: the row keeps the old text, then flips to the new text "a few seconds later" when the server response refreshes the snapshot, and a page refresh fixes it.
One caveat, stated plainly: this explains staleness on the Search list row. It does not explain a stale Description inside the opened expense detail view — I verified MoneyRequestView reads the live transactions_<id> key and sits outside SearchScopeProvider, so the snapshot redirect does not apply there. If you saw the old text in the expense's detail panel rather than in the list, this root cause is incomplete and we need an Onyx trace to find what reverts the value.
Investigation details — what I verified and what I ruled out
Read path.useOnyx.ts:75-106 silently rewrites a read of transactions_<id> to snapshot_<hash> when useIsOnSearch() is true. SearchScopeContext is module-private (SearchScopeProvider.tsx:9), and the only three render sites are the Search list, the Search chart, and EmptySearchView — so the redirect applies to list rows and nothing else.
Detail view reads live Onyx.MoneyRequestView.tsx:236 → getTransactionDetails → getDescription → transaction.comment.comment. Same key the save writes.
Ruled out — "the page reloads" is not a refetch. I initially suspected goBack produced a new route object, refiring the openReport effect in ReportFetchHandler.tsx. Traced it out and refuted it: goUp() emits exactly one StackActions.pop(1), StackRouter rebuilds by slice/concat so the surviving route keeps its object reference, and the effect does not re-run. The only escape hatch is a params mismatch forcing a REPLACE, which mints a new route key and remounts — I could not prove that fires here.
Ruled out — snapshot clobbering the live key. Both copy-back paths are closed on this flow. openReport's snapshot merge is gated on transaction && !parentReportActionID (Report/index.ts:1859), and no caller on the back-navigation path passes transaction. setOptimisticDataForTransactionThreadPreview only runs on row click.
Secondary gap. Even with hash passed, the snapshot write sits inside a block gated on group-policy, non-invoice expenses (UpdateMoneyRequest.ts:2344-2364). Personal-policy and self-DM expenses would still go unpatched.
Not verified: no browser repro, no Onyx trace, no production logs.
What changes do you think we should make in order to solve the problem?
In each of the three dynamic step pages, read the hash with const {currentSearchHash} = useSearchQueryContext(); and pass hash: currentSearchHash into the update call — mirroring exactly what DynamicIOURequestStepCategory.tsx:113,207 already does. No changes to the action layer are needed; hash is already an accepted parameter on all three functions.
Consider widening the snapshot-write gate at UpdateMoneyRequest.ts:2344-2364 so personal-policy and self-DM expenses also get the optimistic patch. I'd treat this as a follow-up, not part of the same fix.
Add a regression test asserting that a Description edit dispatched with a hash writes to snapshot_<hash>, so the next step migration can't silently drop it.
Before implementing, please confirm whether the stale field you saw was the Search list row or the field inside the opened expense. If it was the detail view, item 1 will not fix your repro.
What alternative solutions did you explore? (Optional)
Navigation-triggered refetch overwriting the optimistic value — traced and refuted; see the details block.
Snapshot-to-live copy-back clobbering transactions_<id> — both gates verified closed on this path.
Having the action read currentSearchHash itself instead of threading it — rejected; the action layer is deliberately hook-free, and the existing codebase convention is to pass it from the page.
Separate bug: the issue title (tax amount system message)
Since the title describes a different bug with no repro steps, here is a code-level read on it. Lower confidence — no repro to check against.
Two independent defects, both real in code:
Amount edits silently recompute the tax amount even when tax tracking is off.IOUAmountSubmission.ts:660-666 and TransactionInlineEdit.ts:338-356 recompute taxAmount from the policy tax rate with no isTaxTrackingEnabled guard — while the sibling create-flow path at IOUAmountSubmission.ts:589-595does guard. Meanwhile ReportUtils.ts:6486-6498 deliberately suppresses the tax fragment when the amount changed. Net result: the Tax field moves, the system message stays frozen at the old value. Example: $100 @ 10% with a manual $5.00 tax override, edited to $200 → field shows $18.18, message still says $5.00.
The message is never gated on tax tracking. The Tax rows are conditionally hidden (MoneyRequestView.tsx:605-607) but ModifiedExpenseMessage.ts:464-472 never reads policy.tax, so a "changed the tax amount" message can render for a field the user cannot see.
Suggested fix for both: guard the two recompute sites with isTaxTrackingEnabled(...) and keep the existing taxAmount/taxCode when it returns false.
Ruled out: sign mismatch (both sides Math.abs) and cents-vs-dollars (both sides use backend units). One thing I could not check — ModifiedExpenseMessage.ts:273 notes that ModifiedExpense::getNewDotComment in Web-Expensify must match this logic. If the backend does not apply the same suppression, the server-authored action would carry a taxAmount, and the App would render oldTaxAmount using the new currency. That needs someone with Web-Expensify access to confirm.
💡 Suggested AI code review rule
This looks like a recurring, mechanically detectable pattern — one PR dropped the same argument in three files at once — so it may be worth a coding-standards rule so the AI reviewer catches the next one. Proposed for Expensify/App:
Rule:CONSISTENCY-18 — Thread the Search snapshot hash through optimistic transaction edits
Condition: When a page or hook calls an updateMoneyRequest* action that accepts an optional hash, it must supply hash: currentSearchHash from useSearchQueryContext(). Without it the Search snapshot is not patched and the list row shows stale data until the server responds.
The full rule file would go in App/.claude/skills/app-coding-standards/rules/. If it's worth adding, a maintainer can open it there as a draft PR.
Next Steps for Contributor+ team: To accept:@MelvinBot implement [this](https://github.com/Expensify/App/issues/101937) to create a draft PR. To refine:@MelvinBot <your feedback> To reject: Explain why you are rejecting Melvin's proposal.
@MelvinBot to clarify, all of these steps are within the actual expense view itself. So it's not about the description field in the Spend > Expenses table not updating, it's seemingly within the actual expense.
Go to the Reports (or Expenses) tab in New Expensify.
Open an expense from the list.
Click the Description field to open the full-page editor.
Thanks — the title now matches the body, so the tax-amount half of my previous comment is out of scope here. Worth filing separately if it's a real report; the two defects I described there are still in the code.
On this issue, I dug into the one gap I flagged last time, and it resolved against my proposed fix. Headline:
The missing hash defect is real, but it does not explain the repro as written. Don't implement it as a fix for this issue yet.
What I confirmed
The Description field at step 5 is MoneyRequestView, and it reads live Onyx — not the Search snapshot. This holds in both tab flows (Expenses tab → SEARCH_REPORT RHP; Reports tab → SEARCH_MONEY_REQUEST_REPORT → MoneyRequestReportView). Both land on the same widget at MoneyRequestView.tsx:1289-1313, which reads transactions_<id> directly at MoneyRequestView.tsx:236.
This is now proven structurally rather than by grep: TAB_NAVIGATOR and RIGHT_MODAL_NAVIGATOR are sibling root-stack screens (AuthScreens.tsx:201 vs :312), every SearchScopeProvider lives under the Search tab, and React context doesn't cross that boundary.
After Save, you land back on MoneyRequestView, not on a list row. The dynamic-route suffix is stripped and popped within the same RHP stack, so the panel underneath was never unmounted. So the stale text you're looking at is the detail panel — a live-Onyx read — and patching the snapshot cannot change it.
What the missing hash actually explains
Not nothing — just not this. It explains two adjacent behaviors:
The Search list row behind the RHP staying stale after the edit.
Your "appears a few seconds later" symptom. useSearchHighlightAndScroll.ts:104-120 notices the changed transaction in live Onyx and refires search(), which rewrites the snapshot from the server. That's the delayed heal.
I'd still fix it — see below — but as its own defect.
What I could not determine
Nothing in the client code that I can prove reverts transactions_<id> on this path. The optimistic write is unconditional. I ruled out, with evidence: the navigation refetch (goUp emits a clean pop(1); route identity stable), openReport's snapshot→live merge (no caller passes transaction here, and the one that does is gated on the transaction thread not existing), search responses (they write snapshot keys only), and response-ordering via QueuedOnyxUpdates (OpenReport is a WRITE, so it serializes after the description write and its payload already has the new text).
That leaves two live possibilities I can't separate statically: the value is genuinely overwritten by something I haven't found, or it's never overwritten and the panel simply doesn't re-render.
Also relevant: the body says "Reproducible in staging?: Needs reproduction (Unable to reproduce)". Combined with the above, I don't think this is ready to hand to a contributor.
What would settle it
An Onyx trace of transactions_<id> across the save → goBack window. If the value is overwritten, the trace names the source; if it isn't, this is a re-render bug in MoneyRequestView, which is a completely different fix.
Whether [Report] Legacy transaction fallback: creating money request preview fires on this flow. If it does, Report/index.ts:1906-1910 is the clobber and I was wrong to rule it out.
Whether the backend's UpdateMoneyRequestDescription response echoes transactions_<id> with the old comment. Not visible from the App repo.
A screen recording that makes clear whether the stale text is the detail panel or the list row behind it.
Recommendation
Get a repro first. A video plus an Onyx trace would settle in minutes what I can't settle statically. Without it, any fix is a guess.
Fix the missing hash separately. It's a confirmed, mechanical defect from Migrate money request merchant, description and date steps to dynamic: Description, Merchant and Date all dropped the argument, while Category, Tag and the inline editors thread it. The fix is const {currentSearchHash} = useSearchQueryContext(); then hash: currentSearchHash in each of the three step pages. I verified SearchQueryProvider is an ancestor of those pages in both flows, so the hash resolves.
One caveat on that fix, which the sibling steps already have:currentSearchHash is never undefined — it falls back to -1 at SearchQueryProvider.tsx:60, and the snapshot helper only guards if (!hash …), so -1 passes. Editing from the Inbox would then create a bogus one-transaction snapshot_-1 key. Worth guarding while you're in there. Separately, the snapshot write is nested inside a group-policy, non-invoice gate at UpdateMoneyRequest.ts:2343-2365, so personal-policy and self-DM expenses stay unpatched regardless.
I could not verify any of this in a browser — browser control was unavailable for this run, so everything above is static code analysis.
Next Steps: Reply with @MelvinBot <your feedback> to refine this analysis.
⏱️ Ran out of time before I could finish. This usually means the task was too complex or I got stuck exploring. Please try again with a more specific request or post in #melvinbot-ai.
⚠️Low confidence — I do not have a root cause. Posting what I proved and what I eliminated, so the next person doesn't repeat it.
Understood: the money request view, not the Search table. I took that seriously this time and traced the whole path. My Search-snapshot explanation is wrong for this surface, and I'm retracting it as the fix for this issue.
Here's the uncomfortable result: I traced the save → Onyx → render path end to end and every link checks out. From client code alone, this bug should not happen. That means either the repro depends on something I can't see statically, or the value is being overwritten by the server.
What I proved (so nobody re-checks it)
Link
Result
The save writes the new description to transactions_<id>
✅ Correct — single Onyx.METHOD.MERGE of the whole updatedTransaction at UpdateMoneyRequest.ts:2174-2186; no second write to that key anywhere in the function
getUpdatedTransaction sets comment.comment and nothing clobbers it later
❌ Ruled out — nonTopScreenBehavior: 'freeze' is set only on the split navigator and keys off that navigator's state, so an RHP push never freezes the screen underneath; RightModalNavigator sets none
React.memo swallowing it?
❌ Ruled out — MoneyRequestView, MenuItemWithTopDescription and MenuItem are all plain exports, and MenuItem's title memos list title in deps
React Compiler holding a stale value?
❌ Ruled out — the memo key is transaction, a fresh object on every Onyx merge
OfflineWithFeedback / pendingFields.comment showing the old value?
❌ Ruled out — purely opacity/strikethrough, no previous-value fallback
Also eliminated, from the earlier round
Navigation refetch (goBack emits a clean pop(1); route identity stable, so the openReport effect never refires), openReport's snapshot→live merge (no caller passes transaction on this path), and response ordering via QueuedOnyxUpdates (OpenReport is a WRITE, so it serializes after the description write).
Also worth knowing: there is no test asserting the optimistic comment.comment. UpdateMoneyRequestTest.ts:1092-1126 only asserts a violation survives. So the write path is verified by reading, not by CI.
What's left — ranked
The save silently no-ops.DynamicIOURequestStepDescription.tsx:140-144 returns early without calling the action when newComment === currentDescriptionInMarkdown, where the right-hand side is Parser.htmlToMarkdown(transaction.comment.comment). If the form submits a stale input value, or the HTML→markdown round trip makes a real edit compare equal, you get exactly your step 5: back on the expense, old text, no pending state, no network request at all. I could not prove this fires — it needs a repro.
The backend echoes the old comment.UpdateMoneyRequestDescription is a Web-Expensify command I can't read from here. If its response or a Pusher update re-sends transactions_<id> with the old comment, it lands after the optimistic write and wins.
The repro is on a host that passes updatedTransaction. That prop wins over live Onyx at MoneyRequestView.tsx:237 and is only set by the merge and review-duplicates confirmation pages. Unlikely given your steps, but it would produce this symptom exactly.
The one observation that settles it
With the expense open and DevTools on, save a description edit and check whether a UpdateMoneyRequestDescription request fires at all.
No request → candidate 1. Client-side no-op, fixable in the step page.
Request fires, field still stale → read Onyx.get('transactions_<id>').comment.comment in the console. New value = a render bug I haven't found. Old value = candidate 2, and the fix is backend, not App.
That single check splits the remaining space and takes about a minute. Without it, any fix is a guess — and the body still says "Reproducible in staging?: Needs reproduction (Unable to reproduce)", so I don't think this is ready for a contributor proposal yet.
⚠️@open5588 Thanks for your proposal. Please update it to follow the proposal template, as proposals are only reviewed if they follow that format (note the mandatory sections).
Activity
[-]Tax amount change system message does not match the tax amount field when Taxes are disabled[/-][+]Expense description reverts to old value after saving from the Reports/Expenses (Search) tab[/+]MelvinBot commentedon Sep 22, 2026
Proposal
What is the root cause of that problem?
Saving a Description edit writes the optimistic update to
transactions_<id>, but the Reports/Expenses list renders from the Search snapshot keysnapshot_<hash>. The save never patches the snapshot, because it never passes thehashargument that gates that write.The snapshot write exists and is explicitly documented —
UpdateMoneyRequest.ts:121-124says "Search result rows render from snapshot data, so TRANSACTION writes alone do not refresh the list." It is gated behindif (hash)atUpdateMoneyRequest.ts:2418-2432, andDynamicIOURequestStepDescription.tsx:156-177never passes it — the file does not referencecurrentSearchHashat all.That matches the reported symptoms exactly: the row keeps the old text, then flips to the new text "a few seconds later" when the server response refreshes the snapshot, and a page refresh fixes it.
Merchant and Date have the identical defect. All three were migrated together in Migrate money request merchant, description and date steps to dynamic (merged 2026-08-17), and all three dropped the
hash. The sibling steps that were not in that PR thread it correctly:hash?One caveat, stated plainly: this explains staleness on the Search list row. It does not explain a stale Description inside the opened expense detail view — I verified
MoneyRequestViewreads the livetransactions_<id>key and sits outsideSearchScopeProvider, so the snapshot redirect does not apply there. If you saw the old text in the expense's detail panel rather than in the list, this root cause is incomplete and we need an Onyx trace to find what reverts the value.Investigation details — what I verified and what I ruled out
Read path.
useOnyx.ts:75-106silently rewrites a read oftransactions_<id>tosnapshot_<hash>whenuseIsOnSearch()is true.SearchScopeContextis module-private (SearchScopeProvider.tsx:9), and the only three render sites are the Search list, the Search chart, andEmptySearchView— so the redirect applies to list rows and nothing else.Detail view reads live Onyx.
MoneyRequestView.tsx:236→getTransactionDetails→getDescription→transaction.comment.comment. Same key the save writes.Ruled out — "the page reloads" is not a refetch. I initially suspected
goBackproduced a newrouteobject, refiring theopenReporteffect inReportFetchHandler.tsx. Traced it out and refuted it:goUp()emits exactly oneStackActions.pop(1),StackRouterrebuilds by slice/concat so the surviving route keeps its object reference, and the effect does not re-run. The only escape hatch is a params mismatch forcing aREPLACE, which mints a new route key and remounts — I could not prove that fires here.Ruled out — snapshot clobbering the live key. Both copy-back paths are closed on this flow.
openReport's snapshot merge is gated ontransaction && !parentReportActionID(Report/index.ts:1859), and no caller on the back-navigation path passestransaction.setOptimisticDataForTransactionThreadPreviewonly runs on row click.Secondary gap. Even with
hashpassed, the snapshot write sits inside a block gated on group-policy, non-invoice expenses (UpdateMoneyRequest.ts:2344-2364). Personal-policy and self-DM expenses would still go unpatched.Not verified: no browser repro, no Onyx trace, no production logs.
What changes do you think we should make in order to solve the problem?
const {currentSearchHash} = useSearchQueryContext();and passhash: currentSearchHashinto the update call — mirroring exactly whatDynamicIOURequestStepCategory.tsx:113,207already does. No changes to the action layer are needed;hashis already an accepted parameter on all three functions.UpdateMoneyRequest.ts:2344-2364so personal-policy and self-DM expenses also get the optimistic patch. I'd treat this as a follow-up, not part of the same fix.hashwrites tosnapshot_<hash>, so the next step migration can't silently drop it.Before implementing, please confirm whether the stale field you saw was the Search list row or the field inside the opened expense. If it was the detail view, item 1 will not fix your repro.
What alternative solutions did you explore? (Optional)
transactions_<id>— both gates verified closed on this path.currentSearchHashitself instead of threading it — rejected; the action layer is deliberately hook-free, and the existing codebase convention is to pass it from the page.Possibly related
Expensify/App#96982, open) — also a description edit not reflected in Search results. Same snapshot-staleness family; may share this root cause.Separate bug: the issue title (tax amount system message)
Since the title describes a different bug with no repro steps, here is a code-level read on it. Lower confidence — no repro to check against.
Two independent defects, both real in code:
IOUAmountSubmission.ts:660-666andTransactionInlineEdit.ts:338-356recomputetaxAmountfrom the policy tax rate with noisTaxTrackingEnabledguard — while the sibling create-flow path atIOUAmountSubmission.ts:589-595does guard. MeanwhileReportUtils.ts:6486-6498deliberately suppresses the tax fragment when the amount changed. Net result: the Tax field moves, the system message stays frozen at the old value. Example: $100 @ 10% with a manual $5.00 tax override, edited to $200 → field shows $18.18, message still says $5.00.MoneyRequestView.tsx:605-607) butModifiedExpenseMessage.ts:464-472never readspolicy.tax, so a "changed the tax amount" message can render for a field the user cannot see.Suggested fix for both: guard the two recompute sites with
isTaxTrackingEnabled(...)and keep the existingtaxAmount/taxCodewhen it returns false.Ruled out: sign mismatch (both sides
Math.abs) and cents-vs-dollars (both sides use backend units). One thing I could not check —ModifiedExpenseMessage.ts:273notes thatModifiedExpense::getNewDotCommentin Web-Expensify must match this logic. If the backend does not apply the same suppression, the server-authored action would carry ataxAmount, and the App would renderoldTaxAmountusing the new currency. That needs someone with Web-Expensify access to confirm.💡 Suggested AI code review rule
This looks like a recurring, mechanically detectable pattern — one PR dropped the same argument in three files at once — so it may be worth a coding-standards rule so the AI reviewer catches the next one. Proposed for
Expensify/App:CONSISTENCY-18— Thread the Search snapshothashthrough optimistic transaction editsupdateMoneyRequest*action that accepts an optionalhash, it must supplyhash: currentSearchHashfromuseSearchQueryContext(). Without it the Search snapshot is not patched and the list row shows stale data until the server responds.snapshot_<hash>is gated behindif (hash); Search list rows render from the snapshot, not fromtransactions_<id>. Introduced by Migrate money request merchant, description and date steps to dynamic.src/pages/iou/request/step/Dynamic*.tsxthat callupdateMoneyRequest\w+\(but contain nocurrentSearchHashDynamicIOURequestStepDescription.tsx:156-177DynamicIOURequestStepCategory.tsx:113and:207The full rule file would go in
App/.claude/skills/app-coding-standards/rules/. If it's worth adding, a maintainer can open it there as a draft PR.Next Steps for Contributor+ team:
To accept:
@MelvinBot implement [this](https://github.com/Expensify/App/issues/101937)to create a draft PR.To refine:
@MelvinBot <your feedback>To reject: Explain why you are rejecting Melvin's proposal.
view run
melvin-bot commentedon Sep 22, 2026
Triggered auto assignment to Contributor-plus team member for initial proposal review - @linhvovan29546 (
External)melvin-bot commentedon Sep 22, 2026
Unable to auto-create job on Upwork. The BZ team member should create it manually for this issue.
m-natarajan commentedon Sep 22, 2026
@MelvinBot Updated the title, Review the issue
trjExpensify commentedon Sep 22, 2026
@MelvinBot to clarify, all of these steps are within the actual expense view itself. So it's not about the description field in the Spend > Expenses table not updating, it's seemingly within the actual expense.
MelvinBot commentedon Sep 22, 2026
Review after the title update
Thanks — the title now matches the body, so the tax-amount half of my previous comment is out of scope here. Worth filing separately if it's a real report; the two defects I described there are still in the code.
On this issue, I dug into the one gap I flagged last time, and it resolved against my proposed fix. Headline:
What I confirmed
The Description field at step 5 is
MoneyRequestView, and it reads live Onyx — not the Search snapshot. This holds in both tab flows (Expenses tab →SEARCH_REPORTRHP; Reports tab →SEARCH_MONEY_REQUEST_REPORT→MoneyRequestReportView). Both land on the same widget atMoneyRequestView.tsx:1289-1313, which readstransactions_<id>directly atMoneyRequestView.tsx:236.This is now proven structurally rather than by grep:
TAB_NAVIGATORandRIGHT_MODAL_NAVIGATORare sibling root-stack screens (AuthScreens.tsx:201vs:312), everySearchScopeProviderlives under the Search tab, and React context doesn't cross that boundary.After Save, you land back on
MoneyRequestView, not on a list row. The dynamic-route suffix is stripped and popped within the same RHP stack, so the panel underneath was never unmounted. So the stale text you're looking at is the detail panel — a live-Onyx read — and patching the snapshot cannot change it.What the missing
hashactually explainsNot nothing — just not this. It explains two adjacent behaviors:
useSearchHighlightAndScroll.ts:104-120notices the changed transaction in live Onyx and refiressearch(), which rewrites the snapshot from the server. That's the delayed heal.I'd still fix it — see below — but as its own defect.
What I could not determine
Nothing in the client code that I can prove reverts
transactions_<id>on this path. The optimistic write is unconditional. I ruled out, with evidence: the navigation refetch (goUpemits a cleanpop(1); route identity stable),openReport's snapshot→live merge (no caller passestransactionhere, and the one that does is gated on the transaction thread not existing), search responses (they write snapshot keys only), and response-ordering viaQueuedOnyxUpdates(OpenReportis a WRITE, so it serializes after the description write and its payload already has the new text).That leaves two live possibilities I can't separate statically: the value is genuinely overwritten by something I haven't found, or it's never overwritten and the panel simply doesn't re-render.
Also relevant: the body says "Reproducible in staging?: Needs reproduction (Unable to reproduce)". Combined with the above, I don't think this is ready to hand to a contributor.
What would settle it
transactions_<id>across the save → goBack window. If the value is overwritten, the trace names the source; if it isn't, this is a re-render bug inMoneyRequestView, which is a completely different fix.[Report] Legacy transaction fallback: creating money request previewfires on this flow. If it does,Report/index.ts:1906-1910is the clobber and I was wrong to rule it out.UpdateMoneyRequestDescriptionresponse echoestransactions_<id>with the old comment. Not visible from the App repo.Recommendation
hashseparately. It's a confirmed, mechanical defect from Migrate money request merchant, description and date steps to dynamic: Description, Merchant and Date all dropped the argument, while Category, Tag and the inline editors thread it. The fix isconst {currentSearchHash} = useSearchQueryContext();thenhash: currentSearchHashin each of the three step pages. I verifiedSearchQueryProvideris an ancestor of those pages in both flows, so the hash resolves.One caveat on that fix, which the sibling steps already have:
currentSearchHashis neverundefined— it falls back to-1atSearchQueryProvider.tsx:60, and the snapshot helper only guardsif (!hash …), so-1passes. Editing from the Inbox would then create a bogus one-transactionsnapshot_-1key. Worth guarding while you're in there. Separately, the snapshot write is nested inside a group-policy, non-invoice gate atUpdateMoneyRequest.ts:2343-2365, so personal-policy and self-DM expenses stay unpatched regardless.I could not verify any of this in a browser — browser control was unavailable for this run, so everything above is static code analysis.
Next Steps: Reply with
@MelvinBot <your feedback>to refine this analysis.view run
trjExpensify commentedon Sep 22, 2026
@MelvinBot again, the bug report is entirely about within the money request view.
MelvinBot commentedon Sep 23, 2026
⏱️ Ran out of time before I could finish. This usually means the task was too complex or I got stuck exploring. Please try again with a more specific request or post in #melvinbot-ai.
Run details
MelvinBot commentedon Sep 23, 2026
Understood: the money request view, not the Search table. I took that seriously this time and traced the whole path. My Search-snapshot explanation is wrong for this surface, and I'm retracting it as the fix for this issue.
Here's the uncomfortable result: I traced the save → Onyx → render path end to end and every link checks out. From client code alone, this bug should not happen. That means either the repro depends on something I can't see statically, or the value is being overwritten by the server.
What I proved (so nobody re-checks it)
transactions_<id>Onyx.METHOD.MERGEof the wholeupdatedTransactionatUpdateMoneyRequest.ts:2174-2186; no second write to that key anywhere in the functiongetUpdatedTransactionsetscomment.commentand nothing clobbers it laterTransactionUtils/index.ts:846-851; every latercommentmutation is guarded on a key absent from a description editMoneyRequestViewreads that same live keyMoneyRequestView.tsx:236→displayTransaction→getDescriptionSearchScopeProviderhas exactly three mounts —Search/index.tsx:1221,:1359,EmptySearchView.tsx:97. None wrap report content, so the redirect atuseOnyx.ts:91is inert herenonTopScreenBehavior: 'freeze'is set only on the split navigator and keys off that navigator's state, so an RHP push never freezes the screen underneath;RightModalNavigatorsets noneReact.memoswallowing it?MoneyRequestView,MenuItemWithTopDescriptionandMenuItemare all plain exports, andMenuItem'stitlememos listtitlein depstransaction, a fresh object on every Onyx mergeOfflineWithFeedback/pendingFields.commentshowing the old value?Also eliminated, from the earlier round
Navigation refetch (
goBackemits a cleanpop(1); route identity stable, so theopenReporteffect never refires),openReport's snapshot→live merge (no caller passestransactionon this path), and response ordering viaQueuedOnyxUpdates(OpenReportis a WRITE, so it serializes after the description write).Also worth knowing: there is no test asserting the optimistic
comment.comment.UpdateMoneyRequestTest.ts:1092-1126only asserts a violation survives. So the write path is verified by reading, not by CI.What's left — ranked
DynamicIOURequestStepDescription.tsx:140-144returns early without calling the action whennewComment === currentDescriptionInMarkdown, where the right-hand side isParser.htmlToMarkdown(transaction.comment.comment). If the form submits a stale input value, or the HTML→markdown round trip makes a real edit compare equal, you get exactly your step 5: back on the expense, old text, no pending state, no network request at all. I could not prove this fires — it needs a repro.UpdateMoneyRequestDescriptionis aWeb-Expensifycommand I can't read from here. If its response or a Pusher update re-sendstransactions_<id>with the oldcomment, it lands after the optimistic write and wins.updatedTransaction. That prop wins over live Onyx atMoneyRequestView.tsx:237and is only set by the merge and review-duplicates confirmation pages. Unlikely given your steps, but it would produce this symptom exactly.The one observation that settles it
With the expense open and DevTools on, save a description edit and check whether a
UpdateMoneyRequestDescriptionrequest fires at all.Onyx.get('transactions_<id>').comment.commentin the console. New value = a render bug I haven't found. Old value = candidate 2, and the fix is backend, not App.That single check splits the remaining space and takes about a minute. Without it, any fix is a guess — and the body still says "Reproducible in staging?: Needs reproduction (Unable to reproduce)", so I don't think this is ready for a contributor proposal yet.
Unchanged from before
The missing
hashin the Description, Merchant and Date step pages is still a real, separate defect — it makes the Search table row stale after an edit, and it plausibly explains [$250] Spend-Expense remains in description filter results after its description is changed and [Exploratory] Spend - New group row is not displayed after editing merchant on an expense.. Worth fixing on its own issue. It is not this bug.I could not verify any of this in a browser — browser control was unavailable for this run, so everything above is static code analysis.
Next Steps: Reply with
@MelvinBot <your feedback>to refine this analysis.view run
95 remaining items
github-actions commentedon Oct 15, 2026
mvtglobally commentedon Oct 17, 2026
Issue not reproducible during KI retests. (Second week)
open5588 commentedon Oct 19, 2026
Contributor details
Your Expensify account email: open5588@gmail.com
Upwork Profile Link: https://www.upwork.com/freelancers/~0159c00273dd3bd644
melvin-bot commentedon Oct 19, 2026
✅ Contributor details stored successfully. Thank you for contributing to Expensify!
open5588 commentedon Oct 19, 2026
Contributor details
Your Expensify account email: open5588@gmail.com
Upwork Profile Link: https://www.upwork.com/freelancers/~0159c00273dd3bd644
melvin-bot commentedon Oct 19, 2026
✅ Contributor details stored successfully. Thank you for contributing to Expensify!
yusufdeveloper2903 commentedon Oct 22, 2026
The PR is ready for review. Thanks!