Proof

We don’t sell vibes.
We sell receipts.

The posts that travel aren’t “our agent feels smart.” They’re checkable claims — timeouts caught, loops broken, a green test suite. ActTruth leans into that. Your live ledger stays yours until you export a receipt. The Tier 1 layer is free; the SOVERYN house and wiring are how we charge.

What we claim (and test)

  1. Quiet tool failures become visible FAIL rows — not silence.
  2. Soft {error:…} payloads still count as failures.
  3. Unprompted spend can exhaust → stand-down.
  4. Same tool FAIL ×2+ → soft LESSON (anti-loop).
  5. Lessons reach the agent (prelude + in-band tool result).
  6. Crew includes Kernel; earned-keep penalizes theater.
  7. Stats/proof exporters use the ledger only — no invented uplift.

Verify anytime:

python -m pytest tests/test_acttruth.py -v

Example receipt shape (illustrative)

Format only — not a live scrape. Operators generate real numbers locally:

ActTruth proof (ledger receipts, not vibes)

last 48h · N events
X ok · Y FAIL · Z timeouts · fail rate R%
lessons armed (anti-loop): L
proof suite: 12 passed

Quiet failures become rows. Repeat FAILs become lessons.
https://acttruth.com
acttruth proof
acttruth proof --style markdown
python -m soveryn.platform.acttruth proof   # house shim

Why autonomy makes this sharper

Interactive chat can ask “did that work?” Autonomous pulses cannot. The more the agent acts without you, the more silent failure becomes expensive — and the more a durable act ledger + anti-loop lessons matter.

Free core · paid house

Run the honesty layer yourself — Apache-2.0, no license key. Need a full crew, an install, or someone to read your ledger and kill quiet FAIL loops? That’s SOVERYN / [email protected].