Transparency

Every Stratum ETL and ingestion run appends a signed record to a hash chain, one chain per source and layer. Each record names the hash of the one before it, so changing, removing or reordering history breaks the chain. Once a day a verifier checks every chain and publishes a signed anchor: the head of each chain at that moment.

Because anchors are public and dated, anyone holding a Stratum audit bundle can check that the records in it were committed to here before they received it. Your browser checks the latest anchor below; nothing is sent anywhere.

Latest anchor

Loading the latest anchor…

Files

Verify it yourself

An anchor's anchor_hash is the SHA-256 of its fields anchor_version, stage, generated_at, previous_anchor_hash, public_key_sha256, chains as JSON with sorted keys, compact separators and ASCII escaping. The signature is ECDSA P-256 with SHA-256 over those same bytes, DER-encoded and hex. Each anchor names the previous one, so anchors form a chain too.

Anchors bound when a record could have been changed to the day before the anchor that first covered it. Timestamps come from Stratum's own clock; trusted third-party timestamps are on the roadmap.

Audit bundles, with a real FRED example

An audit bundle is a small zip that proves where one Stratum dataset came from and that its record has not been changed since it was anchored here. You check it on your own machine, with no account and no access to Stratum. Here is a real one, followed from the source.

The example is the yield curve signal. Among other series it uses T10Y2Y, the 10-year minus 2-year Treasury spread published by the Federal Reserve Bank of St. Louis (FRED). The values below are copied from records written in our sandbox on 10 October 2026, abridged.

Every run writes a signed record

  1. Source
    FRED API

    Series T10Y2Y, daily since 1976

  2. Bronze · raw copy
    bronze/fred #24

    06:02 UTC

    13,139 observations fetched, 13,139 stored

    API key recorded as REDACTED

  3. Silver · cleaned
    silver/fred #5

    07:54 UTC

    109,928 rows in (every FRED series), 106,813 out

  4. Gold · signal
    gold/fred #5

    07:58 UTC

    81,019 rows in, 26,399 out to the yield_curve table

Each record names what the run read and wrote, the row counts in and out, and a hash of the output schema. It is signed with a key that never leaves AWS KMS.

Records link into a chain, and the chain is anchored here

  1. gold/fred #4

    07:30 UTC · 26,399 rows

    prev 2280f2b9…

    hash ea129188…

  2. gold/fred #5

    07:58 UTC · 26,399 rows

    prev ea129188…

    hash 272835e5…

  3. Published on this page
    Anchor, 10 Oct 2026 10:00 UTC

    gold/fred head: #5 272835e5…

    plus the heads of 21 other chains, signed

Matching colours are the same hash. Record #5 contains the hash of #4, and the anchor contains the hash of #5. Change one number in #4 and its hash no longer matches the copy inside #5, the record the anchor names.

Checking the anchor published on 10 October 2026…

The bundle packs all of it into one zip

Suppose you used the yield curve from run #4 and want proof. We export the bundle for gold/fred #4: a 10 KB zip of 8 files.

Anyone can check it, offline

$ pip install cryptography
$ python verify_bundle.py verify-bundle stratum-bundle-gold-fred-4.zip
Bundle: stratum-bundle-gold-fred-4.zip
Target: gold/fred seq 4 (lineage_id eb376cb4-9d4b-48fd-b124-3173f94d9b36)
Records checked: 2
VERIFIED: hashes, chain links, signatures and anchor all check out.

It then helps to try to cheat. We edited copies of that same bundle, and this is what the verifier said:

What we changedWhat the verifier reported
Record #5's row count, 26,399 to 26,400 records/…005.json: SHA-256 does not match the manifest
seq 5: content does not match its entry_hash
The same, and the manifest patched to match seq 5: content does not match its entry_hash
Record #5 deleted records/…005.json: listed in the manifest but missing
the last record is not the chain head the anchor commits to

Recomputing the hash does not help: a new hash needs a new signature, and only our KMS key can make one. Someone could still forge a whole bundle with their own key and their own anchor. The verifier only checks a bundle against itself, so it would pass. The last check is yours: compare the bundle's public-key.pem with our signing key, and its anchor_hash with the dated copy published here that morning, 2026-10-10T100006Z.json. The box under step 2 does that comparison for this example.

What a bundle does not prove

Bundles are exported on request today; there is no self-service download yet.