FMP-RESPONSE-001 · Application reliability · Repaired
When a malformed response erased good search results
A successful HTTP response is not necessarily a usable replacement.
My People displayed 12 fictional businesses. A later response with an invalid listing name left zero visible results and disabled the app’s history navigation. The saved shortlist remained intact. Validating the replacement before committing it now keeps the completed search available. [1]
This is a newly reproduced, low-severity defect in our own application. The checks did not establish a browser-engine vulnerability, an attacker’s ability to supply the response or a real customer incident.
The failure began after the network request succeeded
In a local copy, a fresh browser received a valid, fictional search response. The next HTTP 200 response contained a listing with a null name. The app committed that response before rendering, then cleared the result list and failed while creating a card. Its history-state validation also rejected the malformed data. [1]
| Test environment | Before repair | After repair |
|---|---|---|
| Microsoft Edge 157.0.4341.46 | 12 → 0 | 12 retained |
| Playwright WebKit 26.5 on Windows | 12 → 0 | 12 retained |
An ordinary HTTP 503 failure already kept all 12 results. That control isolated the difference between an unsuccessful request and a successful response with unusable content. Saved people survived both versions. [1]
Validate the candidate before changing the completed search
The repair checks the result collection, listing identifiers and names, optional text fields, directory links and supported serialized size before replacing committed data. A rejected candidate leaves previous results, filters and history available, with a message inviting another attempt. A valid empty response still replaces the old results; empty and malformed are different outcomes.
Six targeted regression cases cover five malformed response shapes and a valid empty replacement. The before-and-after browser checks additionally confirm that markup-shaped names remain text and non-HTTP website links are not offered. These are passing controls for the tested inputs, not proof that every injection is impossible. [1]
My People Release 5 also brings bottom navigation and full-screen Browse and Saved panels to narrow screens. Those interface changes are separate from the response-validation repair.
Check replacement behavior with fictional records
Use an authorized local application copy and intercepted test responses, keeping production services and real people’s data out of the exercise.
- Load a valid search, apply a filter and save one fictional listing.
- Return a successful response containing a malformed record. Compare the displayed results, filter, saved list and history with the starting state.
- Repeat with an ordinary failed request. A shared error banner does not prove both paths preserve the same state.
- Return a valid empty response. Confirm it clears obsolete results and remains a usable completed search.
The goal is to preserve completed work when a candidate fails validation, while still accepting an intentional, valid replacement.
Evidence and limits
Both reproductions used fictional responses, fresh profiles and an owned local server at a 390px desktop viewport. No malformed response was sent to the live service. The runs recorded no page errors or unexpected external requests. No incident frequency, customer loss, remote exploit path or browser-engine flaw was demonstrated. Playwright WebKit on Windows is not Apple Safari; physical iPhone, Android and native keyboard behavior remain untested.
[1] Original application evidence. Browser Busters evidence file, finding FMP-RESPONSE-001. The private test workspace retains original and repaired receipts and runnable fictional fixtures. The findings hub separates original application checks from historical vendor security disclosures.