AI execution · proof on every run

AI that ships real work, with a receipt on every run.

SourceA runs automations and agent workflows, then checks the result before you deliver. Try Forge Terminal free — or sign in to Cloud Dispatch.

Try Forge Terminal→ View live proof Cloud Dispatch

Proof

Don't take our word for it. See the receipt.

Every job on SourceA looks like this. Verifiable, every time.

What this proves

We shipped the work, ran quality checks, logged the result, and can replay delivery — not just claim success in chat.

Shareable client receipt Approved

JobSample job

Quality score98/100

Seal225d90aa…

Verify on the page yourself required

Receipts on disk — Governance receipts signed

Proof checks 4/4 Live sample · verify on this page

Rules check PASS PASS or BLOCK · four checks

Proof sample PDF bundle Status page →

sourcea-boot — ★ PyPI eval

request→ policy→ decision→ receipt→ replay→ tamper-FAIL

01

Brief

Scope the job and lock your rules.

02

Dispatch

Policy checked before any agent acts.

03

Run

Specialists execute under the gate.

04

Gate

The gate refuses a bad move — BLOCK.

05

Receipt

Signed evidence written to disk.

06

Replay

Client audits the whole run on site.

01

Something proposes action

Outreach, ops, research — your loop asks to run.

scoped inbox → role check

02

Your rules load first

Policy on disk, scope check, feasibility — before any model token spends.

rules check → PASS

03

Dispatch gate fires

This action, this moment — not governance you discussed at kickoff.

04

Verdict lands on disk

ALLOW or BLOCK — where replay can find it tomorrow.

PASS · BLOCK · tamper-FAIL

05

The loop runs — or stops

Blocked actions never touch production.

06

You export proof

Weekly bundle: HTML, PDF, replay script, handoff for your client.

export → replay → handoff

replay · sourcea-boot

Last factory boot · 2026-07-02 07:29:34 UTC · PASS

$ sourcea-boot --json
SOURCEA_BOOT PASS ok=true
  [SKIP] policy_version: no policy file (POLICY.md) (not configured)
  [PASS] provider: provider env present (ANTHROPIC_API_KEY)
  [SKIP] receipt_fresh: no prior receipt — first boot allowed
  [SKIP] queue_truth: no queue files configured (not configured)
REPORT=receipts/sourcea-boot/BOOT_REPORT.json▋

Try it yourself

W1 demo film. ALLOW · BLOCK · tamper-FAIL.

Governed execution on camera — ALLOW, BLOCK, and tamper-FAIL with receipts on disk.

W1 proof film · real UI capture · AI narration

Run in 5 minutes

pip install sourcea-boot or clone sourcea-boot on PyPI
sourcea-boot --json — four checks, PASS or BLOCK
Walk live scenario for ALLOW/BLOCK/tamper beats

Interactive film uses live boot proof — same terminal you verify in the browser

Play Governance Gauntlet → Try Forge Terminal

Self-serve walkthrough

What to verify. In what order.

Procurement closes when you skip the deck and walk the chain cold — verifiable on your screen in under five minutes.

01

Session gate PASS

Show stack version and role scope load from disk — not chat memory.

02

ALLOW on outreach

One governed send with scope, timestamp, policy version on the receipt.

03

BLOCK on ops

Cross-lane edit refused — no file touched, receipt says BLOCK.

04

Tamper-FAIL

Alter a hash live — spine catches it immediately. That is the wow moment.