Proof

What we have run, and what each record does not show.

Three records from Tioga's own systems. Each one is dated, each one says what kind of evidence it is, and each one says what it should not be read as.

Records

  1. 01 · Real system, synthetic data · 2026-07-31

    Governed write against a real Snowflake account

    What it shows
    3 writes to an open purchase order persisted and were confirmed by re-reading the state afterwards. 2 writes were rejected by policy (a vendor on hold, a closed purchase order). The gateway-to-Snowflake round trip was logged with policy-check and audit-trail data.
    What it does not show
    Not a client deployment and not a production system. The account was a sandbox created for the run, with seed data. Identifiers are removed from the public record.

    Read the run record →

  2. 02 · Real data, our own infrastructure · Captured Sep 8–9 2026

    Excerpt from Tioga's AI call ledger

    What it shows
    16 model calls from Tioga's own routing gateway, each with what was requested, what served it, token counts, cost and the pool it was charged to.
    What it does not show
    A dated excerpt, not a live feed. It excludes a separate private billing pool. It describes Tioga's own usage, not a client's.

    Open the ledger excerpt →

  3. 03 · Automated check · Runs daily, 13:00 UTC

    Daily production check of every live demo

    What it shows
    A scheduled job runs the functional test suite against the live site and calls each demo endpoint, so a retired model or a broken route is found by the job rather than by a visitor. The workflow file and every run are public.
    What it does not show
    A passing run means the tests passed that day. It does not measure answer quality, and it covers this website only, not any client work.

    See the workflow and its runs →

Ledger excerpt, in numbers

Imported from the same data file as the ledger page, so the two cannot disagree.

$0.356603
Excerpt subtotal
16 calls, Sep 8-9 2026 — not the same figure as the live window spend below
16
Calls logged
unsampled — every call, not a spot check
3
Backends in rotation
local free-tier → Google → OpenRouter, by policy
15 / 16
Paid vs. free-tier
call volume has grown past what the free tier absorbs — spend is still nowhere near the $30 cap

Read these records as follows

Questions about any record: contact us. The change history is in the build log.