Skip to content

iOS Location Forensics: Where Was the iPhone?

Every iPhone location source compared: Significant Locations, photo GPS, Wi-Fi networks, locationd caches, Biome and app data, with precision and retention.

Published on 9 min read

"Where was the phone?" is one of the most common questions put to an iPhone examiner, and one of the easiest to answer badly. iOS records location in many places, each with a different purpose, precision and lifetime. Some record where the device was. Others record where a Wi-Fi access point or cell tower is. Some are this device's data, others are synced from the owner's Mac or iPad.

This guide ranks the sources, explains which acquisition level each needs, and gives a method for presenting location evidence with honest uncertainty.

The sources at a glance

SourceWhat it locatesPrecisionRetentionMinimum level
Significant Locations (routined)The device: visits and raw fixesVisits: tens to hundreds of metres; fixes: metresFixes: days; visits: weeks to monthsFull file system, AFU
Photos.sqlite and EXIFWhere a photo or video was takenMetres when GPS was availableAs long as the asset existsUnencrypted backup
Wi-Fi known networksNetworks joinedNetwork range; BSSID geolocation is indirectUntil forgotten; last join onlyEncrypted backup
locationd cachesNearby towers and access points; harvested fixesArea levelDaysFull file system
Biome streamsVisits, Wi-Fi and Bluetooth connections, time zoneVariesDays to weeksFull file system, AFU
Third-party appsWhatever the app storesVariesVariesOften full file system
iCloud and providersFind My, Google, carrier recordsVariesProvider policyLegal process

Significant Locations: the best on-device record

routined learns the places the owner visits and keeps them in three Core Data databases under /private/var/mobile/Library/Caches/com.apple.routined/:

  • Local.sqlite: learned locations of interest and visits on this device, with entry and exit dates, coordinates, uncertainty and confidence.
  • Cloud-V2.sqlite: the same kind of data mirrored across the owner's devices through iCloud. Separate this device's rows from the others before reporting.
  • Cache.sqlite: raw location fixes (ZRTCLLOCATIONMO) with latitude, longitude, horizontal accuracy, speed and course, for the last few days.

The folder is a cache, excluded from backups and protected by class keys available only after first unlock. Plan for a full file system extraction of an AFU device, and collect the WAL files. Details, queries and timestamps are on the Significant Locations page.

Present visits as circles, not pins: plot the coordinate with its uncertainty radius, and state the confidence value. A visit to "a place within 150 m of this point between 19:02 and 21:40" is the honest claim.

Photos: precise, but about the photo

Photos.sqlite stores the GPS location of each asset along with capture time and, importantly, where the asset came from. A photo taken with the device's camera and carrying GPS places the device at that spot at that time, with metre-level precision when GPS was available. A photo saved from WhatsApp, AirDropped or imported from another camera does not: its location describes someone else's device.

Always check the asset's source (import, saved from an app, shared library) before using its location. Compare the database value with EXIF in the file using ExifTool; edits can change one and not the other. Photos reaches an unencrypted backup, which makes it the most accessible location source.

Wi-Fi: networks and access points

com.apple.wifi.known-networks.plist lists networks the device knows, with the date added, last user join and last automatic join, and the BSSIDs it associated with. A hotel network joined on a given day is strong evidence of presence, because Wi-Fi range is short.

Two limits: the plist keeps only the last join per network, not a history, and networks can arrive through iCloud Keychain sync from another device. For join history, look at Biome _DKEvent.Wifi.Connection and knowledgeC Wi-Fi events. Geolocating a BSSID through a public database such as WiGLE gives where others observed that access point, possibly years apart; treat it as a lead.

locationd: permissions and caches

/private/var/root/Library/Caches/locationd/ holds two different things. clients.plist lists apps and services that used Location Services, with authorisation and usage times, and reaches a backup. The cache_encrypted*.db databases hold cell and Wi-Fi location tables and, on some builds, harvest tables. See locationd caches.

The Wi-Fi and cell tables are the classic trap. Their coordinates are positions of infrastructure, often downloaded in bulk for the area around the device. Many rows share a timestamp for that reason. They can support "the device was in this area around this time", but they are not a track. Harvest tables, where present, are closer to device positions.

Biome, knowledgeC and other system data

Pattern-of-life stores add location context without coordinates or with coarse ones: Location.Visit and related streams, Wi-Fi and Bluetooth connections (a car stereo, home speakers), time zone changes (Device.TimeZone, ZSECONDSFROMGMT in knowledgeC). A time zone change is a surprisingly robust travel indicator. See the pattern of life guide.

Apple Maps, weather widgets and other system apps also keep recent searches and locations. Their content and paths vary by release; iLEAPP has modules for Maps history, weather locations and more.

Third-party apps

Ride-hailing, fitness, dating, mapping and social apps often store precise locations with timestamps in their own containers. Google Maps, Waze, Strava, Uber and Life360 are regular sources. Most need a full file system extraction; some reach a backup. Check the installed apps list early to know which containers to examine, and remember that the app's server may hold far more than the device.

Off-device sources

  • Carrier records: cell site data for calls and data sessions, with sector-level precision, through legal process.
  • iCloud: Find My locations are short-lived, but iCloud backups, Photos and synced routined data may hold history. Advanced Data Protection limits what Apple can provide.
  • Google Timeline and similar services if the user enabled them; Google moved Timeline storage to the device for many users from 2024, which changes where to look.

A method for location evidence

  1. Define the question. A place and a time window. "Was the phone near X between 20:00 and 22:00 UTC?" is answerable; "where was the phone last month?" is a project.
  2. Collect broadly. Full file system from an AFU device, encrypted backup, sysdiagnose, and requests to providers.
  3. Rank sources by type. Device positions (routined fixes, camera photos, harvest tables) before infrastructure positions (caches, BSSIDs), before indirect indicators (time zone, network names).
  4. Attach uncertainty to every point: accuracy radius, confidence, and the nature of the source.
  5. Check device attribution. Exclude Cloud-V2 rows and Biome remote records from other devices, and photos not taken by this device.
  6. Look for corroboration. A visit supported by a photo, a Wi-Fi join and a Bluetooth connection to a car is much stronger than any one of them.
  7. Present on a map with circles and times in UTC, with the local offset, and a table listing each point's source and accuracy.

Acquisition checklist for location

  • Keep the device in AFU state and isolated: routined data is only readable after first unlock.
  • Take an encrypted backup early (Photos, Wi-Fi settings, clients.plist).
  • Pursue a full file system extraction for routined, caches, Biome and third-party app containers; collect every -wal and -shm.
  • Take a sysdiagnose for recent unified logs and Wi-Fi scan data.
  • Record whether Location Services and Significant Locations were enabled (Settings, Privacy & Security, Location Services, System Services), photographing the screens if the device is accessible.
  • Request carrier and cloud data in parallel, because their retention clocks are running too.

A worked example

The question: was a phone near a warehouse (a 100 m radius around one point) between 01:00 and 03:00 UTC on a given night, three weeks before seizure? A full file system extraction in AFU state is available.

Cache.sqlite only covers the last four days, so raw fixes are out. Local.sqlite holds a visit to a location of interest whose centre is 60 m from the warehouse, with an uncertainty of 80 m and high confidence, entry at 01:17 and exit at 02:48 UTC. Cloud-V2.sqlite has a matching visit, attributed to this device's identifier. Photos.sqlite holds no photo in the window. The Wi-Fi known networks plist has no network near the warehouse. Biome shows a Bluetooth connection to the suspect's car ending at 01:14 and starting again at 02:51. The locationd cache tables have long been pruned for that date.

The report states that routined recorded a visit, with high confidence, to a place whose estimated centre lies 60 m from the warehouse with an 80 m uncertainty radius, from 01:17 to 02:48 UTC, consistent with the car Bluetooth disconnection and reconnection times. It shows the circle on a map, lists the sources checked without results (photos, Wi-Fi, caches), and does not claim that the phone was inside the building.

Reporting language

Location wording should carry its uncertainty. Prefer "the device recorded a visit to a place estimated at... with an uncertainty of..." to "the suspect was at...". Name the source for every point, give the accuracy value and its unit, and separate device positions from infrastructure positions and indirect indicators. When sources disagree, show both and explain the likely reason (different accuracy, a synced record, a photo from another device) instead of choosing silently. And say what was not available: a backup-only acquisition, pruned caches, or a disabled Significant Locations setting all limit the conclusion.

Timestamps

Most Apple location stores use Cocoa time in seconds (add 978307200 for Unix time). Plist dates display in UTC. Third-party apps often use Unix seconds or milliseconds. Convert everything to UTC before comparing, and watch for rows stamped at download time rather than observation time.

Common mistakes

  • Plotting WifiLocation rows as a route.
  • Using the GPS of a photo received through a messenger as the phone's position.
  • Including Cloud-V2 visits from the owner's iPad.
  • Presenting a visit centre as an exact address, without its radius.
  • Relying on a backup and reporting "no location history", when routined was never acquired.
  • Forgetting that the device can travel without its owner.

Related guides