FMP-UNICODE-001 · Application reliability · Repaired

How a split Unicode character cleared visible results

A valid long source name could be cut through a Unicode character, breaking contact-link rendering. The repaired link keeps the listing and saved data usable.

Owned local code, fictional data and fresh profiles. This finding does not establish a browser-engine flaw, attacker access, customer losses or incident frequency.

What failed, and what had to be true

My People showed one fictional search result. A later listing passed the string checks, but its fallback contact-link construction threw an encoding error and left zero visible result cards. The prior saved shortlist remained intact. A separate startup test with an already-saved malformed name showed zero saved cards and did not initialize history; the original storage bytes remained present. [1]

The demonstrated source path begins with a valid business name containing a character represented by a UTF-16 pair across the producer’s 180-code-unit cutoff. The fictional record passes the relevant source category and location filters and has no usable website, so the browser builds a Find contact link. Our own normalizer splits the character. We did not establish that a real provider currently supplies such a record or that any attacker controls it. The seeded saved-list case is a separate prerequisite, not proof that the baseline UI can normally save the already-failing record.

Transition traced: Valid source Unicode → bounded JSON decoding → source normalization truncates the name → accepted record and serialized cache payload → browser parses and validates the string type → replacement commits → list clears → fallback URL encoding throws. The independent producer-path check used the actual functions with an injected Response and mocked cache, not a deployed-worker request.

Make contact-link construction tolerate the string

Build the fixed HTTPS contact-search URL with URL and URLSearchParams. An unpaired surrogate becomes a replacement character only in that outgoing query. Keep the original listing name, identifier and saved record unchanged. The same card renderer now works during search and saved-list restoration. This is a consumer repair: upstream truncation remains unchanged, and the page does not claim that all Unicode source handling has been corrected.

The original review recorded 14 Unicode-related observations, including six failures. An independent review used the actual producer output at three adjacent cutoff positions: only the pair split by the cutoff failed. Valid ASCII, Arabic, CJK, combining marks and complete astral characters worked. A safe Website bypassed the failing fallback. Invalid raw UTF-8 was replaced during decoding and did not reproduce the escaped-surrogate failure. The repaired run passed all 14 Unicode-related observations, including ten actual Save-and-reload journeys; saved bytes were preserved. [1]

The browser repair ships in My People Release 7.

Use controls, not a dramatic payload

  1. On an owned local copy, supply a fictional valid source name whose complete astral character falls just before, across or just after the normalizer’s cutoff.
  2. Use the actual normalized output, keep category and location eligibility constant, and compare missing Website with a safe Website control.
  3. Record visible cards, status, navigation readiness and the unchanged prior shortlist. Test already-saved malformed text separately from ordinary saving.
  4. On repaired code, verify the fallback URL, valid multilingual text and actual Save-and-reload behavior without changing the original record.

Use synthetic records and an authorized local copy. Keep original failed receipts and distinguish a test-harness mistake from a product defect.

Evidence and limits

Reproduction and repair checks used Microsoft Edge 157.0.4341.46 and Playwright WebKit 26.5 on Windows. Physical Safari, iPhone, Android, native mobile keyboards and Firefox were not tested. Viewport checks do not establish physical-device behavior.

The independent producer trace exercised readBoundedJSON, normalizePhoton, normalizeOSM, cache canonicalization with mocked storage, mergeRecords and rankRecords. It establishes a code path, not production frequency. No network fault was injected into a public service.

No malformed request was sent to production. The evidence shows the tested inputs and transitions; passing controls do not establish universal safety or account isolation.

[1] Original application evidence. See Browser Busters evidence, finding FMP-UNICODE-001. Exact baseline and repaired hashes, bounded private reproducers, browser receipts and failures remain in the project’s operational records. The public evidence contains no customer records or working attack code.