Anti-Spoofing: M5 (Action 2)
The Anti-Spoofing score (metric M5) is built entirely from CAIDA Spoofer test results. The Observatory cannot see your filters directly. It only sees whether spoofed packets escaped your network during a test.
No tests run from your network during a month means no score (-) for that month. You must run tests regularly, at least monthly, to maintain a score.
How the score works
| Test results during the month | Score |
|---|---|
| No tests observed | - (no data) |
| Tests ran, no spoofed packets received | 100% |
| Spoofed packets received from 1 prefix | ~49% |
| Spoofed packets received from 2 prefixes | ~36% |
Each unique prefix that successfully sent spoofed traffic counts against you. The very common 49% score means exactly one prefix leaked spoofed packets during a test that month.
The percentages aren't flat penalties: the count of affected prefixes (the "raw" penalty) is mapped through an exponential decay curve calibrated so a raw penalty of 0.5 scores 60%. One affected prefix (raw 1) works out to ≈49%, two (raw 2) to ≈36%, and each additional spoofed prefix lowers the score further.
How to run a valid test
- Download the Spoofer client from spoofer.caida.org.
- Run it from a host inside your infrastructure, not behind NAT. NAT rewrites source addresses and invalidates the test.
- Test both IPv4 and IPv6 if you operate both.
- Enable "Allow anonymized test results to be shared publicly". Without this, MANRS cannot see your results and your score stays at no-data.
- Run tests from at least two different network segments.
- Repeat monthly. Scores reset each calendar month.
Check your recent public results at:
https://spoofer.caida.org/recent_tests.php?as_include=<your ASN>
My test failed, now what?
A "received" result means spoofed packets escaped from that prefix. To fix:
- Identify the failing prefix/segment in the test report.
- Apply source address validation there (uRPF strict mode, or ACLs at the customer/edge boundary, see BCP 38).
- Re-run the test from the same prefix to confirm.
A failed test counts against the whole month it ran in, so the score recovers once the calendar month rolls over with clean tests.