Verifiability

Trust is a control. So we removed it.

Every other “AI SOC” tool asks you to believe the vendor’s dashboard. Roxas Labs is engineered so that you, or your auditor, or your board, can check the work independently, offline, with no access to our systems.

The difference

Everyone says “it’s all auditable.”

The real question is whether you can check that, or whether you’re trusting that nothing changed between the vendor’s database and the export they hand you.

The question
Other vendors
Roxas Labs
Can you check it yourself?without asking the vendor
trust us
recompute it
Would a dropped or edited row show?a changed row among thousands
not visible
chain breaks
Where does the proof live?whose systems hold the anchor
their database
off our systems
Does it work offline?no login, no vendor in the loop
request an export
self-contained bundle
What does it prove?the result of the check
unverifiable
chain intact
The ledger

Hash-chained and append-only.

event
sha-256idle
row 
prev fingerprint 
entry hash 
ledger, newest first

Every material event becomes a row, locked to the row before it by a fingerprint. Change or drop a row and the chain breaks. Rows can only be added, never edited or deleted.

  • Sequential primary key: gaps are visible
  • Per-entry Ed25519 signatures
  • Concurrent writers can’t fork the chain
Anchoring

Signed checkpoints, anchored off-box.

On a schedule, the current chain head is signed and the checkpoint is written out: emailed to you, dropped in object storage, wherever you keep it. Because a checkpoint records where the chain was, a later chain that is shorter than a known checkpoint fails verification. Truncation isn’t just discouraged. It’s detectable.

checkpoint 2026-09-01T00:00Zvalid
seq
1043
head
9f55…7d0e
sig
3a90…c1, valid
stored
email + blob container
The part that matters

Your auditor verifies it without us.

We export a self-contained bundle. Your auditor runs the verifier against it with no database access, no API key, no call back to Roxas Labs.

auditor@laptop — offline
$ roxas verify roxas-q3.json
bundle: 1,043 rows · 12 checkpoints · public key loaded
re-checked every row’s fingerprint from the first entry; head matches
1,043 per-entry signatures valid
12 checkpoint signatures valid
no checkpoint beyond chain end; no truncation
RESULT: chain intact. No calls to Roxas Labs were made.
$
On top of the chain

The attestation report.

A framework-mapped summary for any period: the thing you hand to a board, a regulator, or a cyber-insurance underwriter.

Operations attestation

Q3 2026 · Acme Corporation

Period01 Jul – 30 Sep
312Investigations
18%True-positive rate
22%Human-review rate
2mMedian time to verdict
Verdict mix
  • True positive 18%
  • Benign positive 21%
  • False positive 57%
  • Undetermined 4%
MITRE ATT&CK tactic coverage

11 of 14 tactics seen and investigated

Response actions

18 taken · 18 approved by a person · 2 open in the backlog

Integritychain intacthead 9f55…7d0echeckpoint seq 1043 validpublic key a91c…e7a1
  • Verdict mix, true-positive rate, human-review rate
  • Time-to-triage and time-to-verdict, median
  • MITRE ATT&CK tactic and per-rule coverage
  • Response actions taken, open backlog, cost
  • An integrity anchor: chain-verify result, head, latest checkpoint, public key
Why anyone should care

Where “we checked our dashboard” isn’t an answer.

CMMC / DIB

Defense supply-chain assessors want evidence a control operated, not a screenshot. A signed, offline-verifiable chain is that evidence.

Cyber insurance

Underwriters and MGAs are pricing on whether monitoring is real and auditable. A verifiable operations record is a rate conversation.

Board reporting

When the board asks whether monitoring is real, “trust our AI” doesn’t survive a bad quarter. A tamper-evident record does.

We’ll run the verifier with you.

Walk a full investigation, then run the offline verifier against the exported audit bundle together, with no access to anything but the file.