Multi Site Network Management: A Coverage Audit
Run a coverage audit across every site in your footprint to find the ones with no devices, no subnets, no circuit endpoint, or a contact who left two years ago.

On this page
- At a Glance: Three ways to find documentation gaps
- Decide the coverage rules before you query
- The site list is the only complete census
- Score every site from one endpoint
- Reconcile circuits from both ends
- Catch records that cannot be reached from a site
- Turn the output into a worklist
- Make the next run a diff
- Start with the two orphan checks
Every multi site network develops a second network on paper, and the two drift apart at different speeds. A switch gets replaced and the record keeps the old model. A site is closed and its subnets stay active in the plan. A circuit is canceled and the carrier reference survives in a spreadsheet that nobody opens until the next invoice dispute. None of it looks like a failure, because each record was correct on the day it was written.
The usual response is to audit the whole organization and report totals. Four thousand devices, ninety subnets, twelve circuits. Those numbers say nothing about whether a specific site is documented well enough to support it, and they hide the one thing worth knowing: which site would be hardest to troubleshoot this afternoon. A coverage audit inverts the question. It takes the site list as the census of what exists, then tests every site against the records it should own.
At a Glance: Three ways to find documentation gaps
| Spreadsheet sweep | Discovery tool | Coverage audit by site | |
|---|---|---|---|
| Starts from | Files somebody maintains | The live network | The site list |
| Finds | Missing columns, stale notes | Reachable devices and addresses | Sites with nothing attached, records with no site |
| Misses | Anything nobody wrote down | Planned sites, powered off hardware, the carrier side of a circuit | Whether the network is actually reachable |
| Best for | One or two owners | Confirming current state | Deciding where documentation work goes next |
The three are complementary. Discovery tells you what answered today, and a coverage audit tells you where the answer is missing a home.
Decide the coverage rules before you query
A rule you cannot test against a field is not a rule, it is an aspiration. Write each one so that a single API field proves or disproves it, and write down the failure signal at the same time. Six rules cover the questions that come up during an incident.
| Rule | Field that proves it | Failing signal |
|---|---|---|
| The site is grouped | site_group_name |
Empty, so it sits outside every rollup |
| Somebody can be called | contact_name |
Empty contact |
| Hardware is recorded | devices_total |
Zero devices |
| Address space is recorded | subnets_total |
Zero subnets |
| A WAN path is recorded | site_a_id or site_b_id on a circuit |
No circuit endpoint |
| The record was checked recently | updated_at |
Older than 90 days |
Exclude sites that are supposed to be empty before you apply the rules. A site in a planned state with no hardware is not a finding, and a closed site still carrying old records is a different problem from an active site with nothing recorded. Site status is free text, so standardize the values you use for planned, active, and retired before the audit, otherwise the exclusion itself becomes a discussion.
The site list is the only complete census
Everything downstream depends on one assumption: the site list is complete. It is also the assumption most worth testing first, because a missing site makes every later check silently wrong. Pull the list and its groups, then look for the two failure shapes that hide entire locations.
export OBELINF_API_KEY="your_read_only_key"
export OBELINF_ORG="acme-netops"
curl --fail --silent --show-error -H "Authorization: Bearer ${OBELINF_API_KEY}" \
"https://api.obelinf.com/${OBELINF_ORG}/sites?all=true" \
| jq -r '.sites[] | [.id, .name, .slug] | @tsv'
One parameter matters here. all=true returns up to 1000 rows in one call, and limit is capped at 100, so paging without all will quietly truncate a large footprint. The same endpoint accepts ungrouped=true, which is the fastest way to find sites that exist outside every regional rollup.
curl --fail --silent --show-error -H "Authorization: Bearer ${OBELINF_API_KEY}" \
"https://api.obelinf.com/${OBELINF_ORG}/sites?ungrouped=true&all=true" \
| jq -r '.sites[] | [.name, .slug] | @tsv'
lab-cage lab-cage
lisbon-pop lisbon-pop
The other half of the census is the group list. Each group comes back with site_count and child_count, so a group that exists but holds nothing is visible without counting sites yourself.
curl --fail --silent --show-error -H "Authorization: Bearer ${OBELINF_API_KEY}" \
"https://api.obelinf.com/${OBELINF_ORG}/site-groups?all=true" \
| jq -r '.site_groups[] | select(.site_count == 0) | [.name, .child_count] | @tsv'
An empty group is not automatically wrong, since a structure is often built before the sites land in it. Two empty groups after a year is a sign that the hierarchy was designed in a meeting and never applied.
Score every site from one endpoint
The reason this audit is cheap is that the site detail endpoint already returns counts. One call per site gives you racks, devices, subnets, the grouped name, the status, the contact, and the last update timestamp, so the whole scorecard fits in a single loop.
STAMP=$(date -u +%F)
mkdir -p "audit/${STAMP}"
OUT="audit/${STAMP}/site-coverage.tsv"
printf 'site\tgroup\tstatus\tracks\tdevices\tsubnets\tcontact\tupdated\n' > "$OUT"
for id in $(curl --fail --silent --show-error -H "Authorization: Bearer ${OBELINF_API_KEY}" \
"https://api.obelinf.com/${OBELINF_ORG}/sites?all=true" | jq -r '.sites[].id'); do
curl --fail --silent --show-error -H "Authorization: Bearer ${OBELINF_API_KEY}" \
"https://api.obelinf.com/${OBELINF_ORG}/sites/${id}/detail" \
| jq -r '[.site.name,
(.site.site_group_name // "ungrouped"),
.site.status,
.racks_total, .devices_total, .subnets_total,
((.site.contact_name // "") | if . == "" then "missing" else . end),
.site.updated_at] | @tsv' >> "$OUT"
done
The output has a fixed shape: one row per site, eight columns in that order, every run. Keep the order identical or the run to run diff becomes noise.
ams-dc1 Regions/EMEA active 8 96 14 Marta Silva 2026-10-21T09:14:02.311Z
fra-dc2 Regions/EMEA active 4 38 9 Jonas Weber 2026-10-18T16:40:55.008Z
lisbon-pop ungrouped active 0 0 2 missing 2026-03-02T11:02:19.774Z
A total count flattens this table into a number. The rows are what you act on, so filter for the ones that fail a rule.
awk -F'\t' 'NR > 1 && ($3 == "active") && ($4 == 0 || $5 == 0 || $6 == 0 || $7 == "missing")' "$OUT"
One caveat changes how you read the device column. A device is attributed to a site through its rack, so devices_total counts only racked hardware. A site reporting zero devices may own unracked equipment that the same query cannot see. That is not a flaw in the endpoint, it is the finding, and a later section catches those records directly.
Reconcile circuits from both ends
Circuits are where multi site documentation fails most often, because half the record lives with the carrier and the other half lives on your side. The circuit list has no site filter, which is a useful constraint rather than an obstacle: fetch every circuit and count how many times each site appears as an endpoint.
curl --fail --silent --show-error -H "Authorization: Bearer ${OBELINF_API_KEY}" \
"https://api.obelinf.com/${OBELINF_ORG}/circuits?all=true" \
| jq -r '.circuits[] | [.site_a_name, .site_b_name][] | select(. != null)' \
| sort | uniq -c | sort -rn
6 ams-dc1
4 fra-dc2
2 lisbon-pop
Sites missing from that list have no documented circuit endpoint. Most of them should, because a branch with no recorded WAN path is a branch whose connectivity plan exists only in someone’s memory. The exception is a site that genuinely has no wide area link, such as a single building served by local internet only. Record that decision in the site record rather than leaving the rule silently failed.
curl --fail --silent --show-error -H "Authorization: Bearer ${OBELINF_API_KEY}" \
"https://api.obelinf.com/${OBELINF_ORG}/circuits?all=true" \
| jq -r '.circuits[]
| select(((.cid // "") == "") or .bandwidth == null)
| [.name, (.provider_name // "unknown"), (.cid // "no cid"), (.bandwidth // "no bandwidth")]
| @tsv'
ams-fra-backup Zayo no cid no bandwidth
lon-ams-primary Colt ZAY-88421-EMEA no bandwidth
Bandwidth is stored in Mbps as a whole number, so 1000 means 1 Gbps and a null value means nobody entered the committed capacity. On a circuit row, cid is the carrier reference. It is the only field an escalation engineer can read back to a provider, and a circuit without one turns a two minute call into a contract search.
The pattern sits in the first row of that sample: the backup circuit has neither field, which is typical, because backup links get ordered in a hurry and documented later.
Catch records that cannot be reached from a site
The coverage rules assume every record has a path back to its site. That path is not always direct, and knowing where it breaks is what stops the next audit from blaming the wrong team.
Devices reach a site through a rack, so a device with no rack is invisible to every site filter. This query finds them.
curl --fail --silent --show-error -H "Authorization: Bearer ${OBELINF_API_KEY}" \
"https://api.obelinf.com/${OBELINF_ORG}/devices?all=true" \
| jq -r '.devices[] | select(.rack_name == null) | [.name, (.manufacturer_name // "unknown"), .status] | @tsv'
core-sw-04 Juniper active
pdu-b-r07 APC active
Subnets attach to a site directly, so an unassigned subnet is a single missing field rather than a missing chain. The ip_count column tells you whether it is worth chasing: forty recorded addresses belong to something somebody uses, while zero suggests the subnet was created in advance and abandoned.
curl --fail --silent --show-error -H "Authorization: Bearer ${OBELINF_API_KEY}" \
"https://api.obelinf.com/${OBELINF_ORG}/subnets?all=true" \
| jq -r '.subnets[] | select(.site_id == null) | [.cidr, .name, .ip_count] | @tsv'
10.42.0.0/24 legacy-users 61
192.168.250.0/24 lab-temp 0
Neither record should be deleted during the audit. An unassigned record is evidence of a link that was never made, and deleting it removes the only proof that the subnet existed. Assign it, or mark it for retirement, and let the deletion happen as a separate reviewed change. A site record that owns its racks, subnets, and circuit endpoints is what makes this audit answerable in the first place, and Obelinf’s multi site model keeps those links as records rather than conventions.
Turn the output into a worklist
A handful of queries produce a worklist that one person can own. Sort the findings by what an incident needs first, not by how many rows each finding produced.
| Priority | Finding | Evidence | Decision | Owner |
|---|---|---|---|---|
| 1 | Sites with no circuit endpoint | Endpoint count | Record the circuit, or mark the site as local access only | Network |
| 2 | Circuits without cid or bandwidth |
Circuit query | Pull the reference from the carrier portal, or mark the circuit retired | Carrier management |
| 3 | Active sites with zero racks and zero devices | Coverage rows | Import the hardware, or move the status to planned | Site lead |
| 4 | Devices with no rack | Device query | Assign a rack, or record the device as unracked with a reason | Field operations |
| 5 | Subnets with no site | Subnet query | Attach to a site, or schedule retirement | IP planning |
| 6 | Sites untouched for 90 days | updated_at |
Verify the three fields that change most: contact, circuits, subnet plan | Each site owner |
Priority one is the site with no recorded path to the rest of the network, because that is the record an engineer reaches for when the site is unreachable. Priority two is the field that lets somebody else act on a circuit without asking you. Contact details and subnet plans come last because they change slowly and are cheap to verify again.
Give every row an owner and a due date in the same table, then keep the table with the timestamped output. A finding without an owner is a note, and notes do not survive the next reorg.
Make the next run a diff
The value of a coverage audit is not the first run, it is the second one. Keep the column order fixed, write each run into its own dated directory, and compare them.
diff <(cut -f1-7 "audit/2026-10-29/site-coverage.tsv") \
<(cut -f1-7 "audit/2027-01-28/site-coverage.tsv")
The updated column is dropped from the comparison on purpose, because it changes on every run and would drown the signal. What remains is the shape of the footprint: a site that had twelve subnets and now has zero is a broken relationship, not a cleanup, and it should be treated as an incident in the inventory rather than a successful subtraction.
Stale rows deserve a string comparison rather than a date parser, since ISO timestamps sort lexicographically.
awk -F'\t' -v cutoff="$(date -u -d '90 days ago' +%Y-%m-%dT%H:%M)" \
'NR > 1 && substr($8,1,16) < cutoff {print $1 "\t" substr($8,1,10)}' \
"audit/2026-10-29/site-coverage.tsv"
That -d flag is GNU date. On macOS and BSD use date -u -v-90d +%Y-%m-%dT%H:%M instead, and keep the substr length at 16 so the comparison still lands on the minute.
Run the job with a key that has read permission. The audit only issues GET requests, and a scheduled job is the wrong place to hold a key that can rewrite the records it is measuring.
Start with the two orphan checks
Run the ungrouped query and the coverage loop today, on the sites you can name from memory. The rows that surprise you are the audit, and the first one to fix is usually a site that exists in the network and not in the site list, because every other check was quietly skipping it.
From there, work the worklist from priority one and let the dated output tell you whether the gap is closing. The network audit checklist covers the rest of the infrastructure review once the site level coverage is in place.
Frequently Asked Questions
What is a multi site network coverage audit?
How do I find sites with no documented network devices?
Should records that belong to no site be deleted?
How often should a multi site coverage audit run?
Can I run a coverage audit with a read key?
Free Tools
Put this into practice with the free calculators, reference tables, and checklists in the Obelinf toolkit.
Stop reaching for a spreadsheet
Obelinf keeps every subnet, device, circuit, and rack in one live source of truth, with audit logs and a topology view. Free for personal use.
Related Articles

OSPF vs Static Routes for Multi Site Networks
Static routes are fine until the second path, the first dead neighbor, or the twelfth site. A practical guide to choosing OSPF or static routing per site, with configs, cost math, and a cutover procedure.
Read more
ASN Planning for Multi-Site Networks
How to decide between one ASN and many, allocate prefixes that summarize cleanly, build the internal control plane, and verify a site's announcements before it leaks a route.
Read more
How to Build a Branch Office Network Inventory
Build a branch office network inventory that links sites, circuits, devices, subnets, and owners, with a rollout process your team can keep current.
Read more