Skip to content

LearnPhysical

How to report a physical pentest

Brief

A physical report describes doors, readers, procedures and ports, never the people who held a door, because the fix is always a change to a control and a named person changes nothing. Each finding pairs what the timed log shows with the client's own access and camera records, a fix with its cost class, and the walk that retests it.

Stage
Report
Reviewed
On this page
  1. Check first
  2. How do I report findings by control rather than by person?
  3. How are photographs handled under the agreed rules?
  4. How do I match the timed log against the client's access and camera records?
  5. How do I write the fix per control with a cost class?
  6. How do I plan the retest walk?
  7. The defender sees
  8. Done when
  9. Not covered
  10. Tools
  11. Related research
  12. Reviewed

All commands

Situation
A finished physical test under a signed authorization letter, with the timed log, the photographs taken under the agreed photography rules, and the client's access control export and camera timestamps for the test window. Everything here works on those records for the sites the letter names. Photographs and the raw log are held only for the agreed retention and are handed over or destroyed when the report is accepted.
Goal
One report with findings grouped by control, each with its evidence, the matching access or camera record, the fix and its cost class, plus a retest walk plan.
Outputs
Representative examples on placeholder hosts, trimmed to the lines that matter. Not captured from a public fixture yet.

Check first

  • The authorization letter's photography rules: what may be photographed, whether location data may stay in files, and how long photographs are kept.
  • The timed log, one line per attempt with time, door, control and result, written during the test.
  • The client's access control export and camera timestamps for the same window, requested before the readout.
  • The rule the letter sets on naming: no staff names, and no description that points at one person.
  • The site survey map, so every door in the report uses the same id.

Steps

How do I report findings by control rather than by person?

awk -F, 'NR > 1 { a[$3]++; f[$3] += ($5 == "succeeded"); c[$3] += ($6 == "yes") }
  END { for (k in a) printf "%s\t%d attempts\t%d succeeded\t%d challenged\n", k, a[k], f[k], c[k] }' timed-log.csv | sort

Expected output

badge reader	4 attempts	3 succeeded	0 challenged
network port	3 attempts	2 succeeded	0 challenged
tailgating	6 attempts	5 succeeded	1 challenged   # <- five of six followed through a controlled door
unattended area	2 attempts	2 succeeded	0 challenged
visitor process	2 attempts	1 succeeded	1 challenged

Each line of the log carries the control it tested, so the report's findings come out grouped by badge, tailgating, visitor process, unattended areas and network ports. A staffed post with one person on it is described by the process ("the lobby visitor sign-in did not check identification"), because naming the post would name the person.

Negative result. Every attempt on a control refused or challenged; the control is reported as held, with the count.

Fix. Agree the control list before the test, so every attempt is logged against a control and the report never needs a person to explain a result.

How are photographs handled under the agreed rules?

exiftool -q -r -if '$GPSLatitude' -p '$FileName' photos/ | wc -l
exiftool -q -r -p '$FileName $CreateDate' photos/ | head -n 2

Expected output

3   # <- three photographs still carry location, which the letter says must be removed before the report
IMG_0412.jpg 2026:10:21 10:42:07
IMG_0413.jpg 2026:10:21 10:44:51

Photographs show doors, readers, gaps, ports and screens, never faces; one that caught a face by accident is deleted, not blurred and kept. The capture time stays, because it is what the client matches against camera footage. Location data is removed when the letter says so, and the report captions each photograph with its door id instead.

Negative result. No file with location data, no face in any photograph, and every caption carries a door id and a time.

Fix. Put the photography rules in the letter: subjects allowed, metadata kept or stripped, storage, and the date photographs are destroyed.

How do I match the timed log against the client's access and camera records?

awk -F, 'NR == FNR { if (FNR > 1) seen[$1 "," $2] = 1; next }
  FNR > 1 && $5 == "succeeded" && $3 != "network port" { printf "%s %s %s\n", $1, $2, (($1 "," $2) in seen) ? "access event" : "no access event" }' \
  access-export-doors.csv timed-log.csv | grep 'no access event'

Expected output

10:42 D3-east-dock no access event   # <- an entry the access system never recorded: a tailgate or a propped door
11:15 D1-lobby no access event

The client exports access events as minute and door id for the doors in our log, so each entry is matched to the same door in the same minute. An entry with no access event is a tailgate or a door that did not latch; an entry with an event under a real cardholder's name is a cloned badge, and the report says which without giving the name. Camera timestamps are matched the same way.

Negative result. Every entry has a matching access event or a camera alarm that someone acted on.

Fix. Alarm doors held open past a set time, and review held-open and forced-door alarms daily.

How do I write the fix per control with a cost class?

Control           Finding                                    Fix                                         Cost class
Badge reader      125 kHz copies accepted at D1, D3           disable legacy technology per reader        configuration
Tailgating        5 of 6 entries at D3 unchallenged           turnstile or staffed lobby at D3           hardware
Visitor process   no identity check at sign-in               check identification, escort to host       procedure
Unattended area   print room open, screens unlocked          door closer, screen lock policy            hardware, configuration
Network port      DHCP lease on user network, room 2.14      802.1X, disable unused ports               configuration

The cost class tells the client which fixes are a setting, which are a procedure and which need hardware or building works, so the cheap ones are not held back by the expensive ones.

Negative result. A control with no fix the client can afford this year still gets a cost class and an interim procedure.

Fix. One owner and one date per row, agreed at the readout.

How do I plan the retest walk?

join -t "$(printf '\t')" round1-by-control.tsv round2-by-control.tsv \
  | awk -F'\t' '{ printf "%s\t%s of %s\t%s of %s\n", $1, $3, $2, $5, $4 }'

Expected output

badge reader	3 of 4	0 of 4   # <- no reader accepts the copy after the legacy technology was disabled
tailgating	5 of 6	2 of 6

The retest walks the same doors with the same controls and counts the same way, after the fixes are in. Both files are sorted by control for join, with attempts then entries per round.

Negative result. No change on a control means its fix did not reach the doors tested, and the row stays open.

Fix. Retest configuration fixes per reader or port, not per building.

The defender sees

Writing the report touches nothing on site. The matching step tells the client what their own systems saw: access events, alarms and footage. An entry nobody's system recorded, or an alarm nobody answered, is reported as a finding beside the control that failed.

Done when

Every attempt in the timed log maps to a control finding, no finding or photograph identifies a member of staff, each finding is matched against the client's access or camera record, each fix carries a cost class, and the retest walk lists the doors, the controls and the date.

Not covered

Running the test (see badge and physical access testing), disciplinary use of results, which the report rules out on its first page, and building works design; the outputs shown are representative examples on a placeholder site, not captured runs.