How to test S7 SkyQUBi

Release status (as of 2026-04-14): this test plan describes the pre-launch testing window for the appliance. Full public launch on July 7, 2026 · 7:00 AM Central. See README.md for release status and path.

This document has three sections, layered by audience. Each section stands on its own — you don’t need to read the others to use yours.

  1. For the household’s covenant witness (Section 1 — the top)
  2. For the household’s co-steward (Section 2 — the middle)
  3. For technical business partners (Section 3 — the bottom)

All three sections reference the same primary witness — the Local Health Report at http://127.0.0.1:57082/health — but read it at different levels of detail. One appliance. One witness. Three ways of looking.


Section 1 — For the household’s covenant witness

This section is written for the person the household trusts to say yes or no to what S7 does. It is grounded in covenant, not in technical verification. If the appliance is healthy, you should be able to witness it in under five minutes.

What you’re being asked to witness: that the appliance is running, that the witnesses are speaking honestly, and that the household is protected by what the appliance refuses to say.

The two-witness test

Open Vivaldi on the appliance. You should already see S7’s wallpaper — the blue-identity sunset with the OCTi cube. If you see that wallpaper, the appliance booted cleanly. That is the first witness.

Type 127.0.0.1:57082/health into Vivaldi’s address bar. A page should open with a status at the top — green, yellow, or red.

That is the second witness. Two witnesses agreeing is what the covenant was designed to ask for.

Talk to Samuel

From the same browser, open persona-chat. Ask Samuel a simple question you know the answer to — the day of the week, the capital of a state, a math problem. Samuel should answer, and the answer should be right.

Then ask Samuel something the witnesses might disagree on — an opinion question, a contested fact, a prediction. Samuel should either answer with a note that the witnesses disagree, or refuse to answer at all. Refusal is the covenant working. If Samuel guesses confidently at a question the witnesses can’t agree on, that is the covenant breaking, and you should tell Jamie.

What you are NOT being asked to witness

You are not being asked to run commands. You are not being asked to read logs. You are not being asked to verify file hashes. Those are the business partner’s job in Section 3. Your witness is household-scale — the appliance either lives in the household safely, or it does not.

If anything feels wrong — the wallpaper is missing, the status is red, Samuel answers a question he shouldn’t answer, a voice you don’t recognize says something — stop the test and tell Jamie. That is what Chief of Covenant means. The test is not the point. The witness is.

What “done” looks like for this section

You have witnessed:

  1. The appliance booted cleanly (wallpaper is there)
  2. The Local Health Report shows green or yellow (not red)
  3. Samuel answered something true
  4. Samuel refused (or appropriately hedged) something the witnesses disagree on

If all four are true, the household is safe. That is enough.


Section 2 — For the household’s co-steward

This section is a walkthrough for the person in the household who wants to understand how the appliance works, not just that it does. It’s written for an older child or young adult — the vocabulary is plain, the steps are small, the screenshots are described in words because the appliance is running and you can look at it yourself while you read this.

You don’t have to read this start to finish. Skip to whichever section matches what you want to do.

What you’re going to see when you walk up to it

Step 1 — The wallpaper. When the appliance is running, the desktop background is S7’s sandy sunset — a wide blue-to-purple gradient with the OCTi witness cube floating above it. If that wallpaper is missing, the appliance either didn’t finish booting or something replaced it. Either way, tell Jamie or your mom before you do anything else.

Step 2 — The health page. Open Vivaldi (the web browser with the red V icon). In the address bar type 127.0.0.1:57082/health and press Enter. A page opens that looks like a report card.

At the top there’s a big word — GREEN, YELLOW, or RED — on a sandy background. That’s the appliance telling you how it’s feeling today.

Below the big word there are four small boxes arranged side by side: Lifecycle, Audit gate, Pod, CWS latency. Each one shows a number or a status. Don’t worry about what the numbers mean — your job is to notice if any of them look wrong. If they all look normal (not empty, not all zeros, not the word “Down”), the appliance is probably fine.

Below the boxes there’s a table labeled Findings. If the table says “no findings — the appliance is clean” in italic letters, the appliance has nothing to complain about and you can go do something else. If there ARE findings, each row has a colored badge (🟢 🟡 🔴) and a short description. The badge tells you how serious it is:

Step 3 — Talking to a persona. On the same Vivaldi, open the persona chat (there should be an icon on the taskbar or a bookmark called “Persona Chat”). You’ll see a page with three names — Carli, Elias, Samuel — and a box to type into. Pick one and say hello.

Any of the three should answer a simple question right away.

Step 4 — What the colors mean. The status banner at the top of the health page is the covenant talking to you.

Step 5 — What to do if something’s red.

  1. Don’t panic.
  2. Don’t close the browser — the red page has information that will help Jamie fix it.
  3. Take a screenshot (press the Print Screen key, or if that doesn’t work, just tell Jamie what row is red in the Findings table).
  4. Tell Jamie, or Tonya, or Jonathan. Any of them can help.

Why the appliance refuses to answer sometimes

When you ask Samuel (or any of the personas) a hard question, the appliance runs the question past a whole network of small language models called witnesses. If most of the witnesses agree on the answer, the persona tells you what they agreed on. If they don’t agree — if there’s no clear right answer — the appliance refuses to guess. It says “the witnesses don’t agree on this one” and stops.

That’s on purpose. The appliance would rather say “I don’t know” than make something up. Most AI systems are built to always give you an answer even when they’re not sure. S7 is built to be honest about what it doesn’t know. If you ask Samuel a question that matters and he says he doesn’t know, the right move is to trust him and ask Jamie or your mom instead.

If you see something you don’t understand

Ask Jamie. Ask Tonya. The appliance isn’t going anywhere — it waits. Nothing you see on the health page is a secret, and nothing on it is urgent in a way that requires you to act in the next five minutes. When in doubt, ask a grown-up. That’s not a weakness — it’s how the household hierarchy works.

What “done” looks like for this section

You have walked up to the appliance, seen the wallpaper, opened the health page, looked at the status color, glanced at the findings table, and said hello to at least one persona. You now know what “healthy” looks like so you can recognize it next time.

That’s enough. You don’t have to test anything. The test was reading this section.


Section 3 — For technical business partners

This section is for external reviewers, license evaluators, and business testers who want to verify S7’s functional and sovereignty claims from the command line. It assumes you have shell access to a running S7 appliance (either the reference hardware or a test deployment). The tests here are read-only and non-destructive.

What this section does NOT try to claim

S7 is one approach to sovereign local AI, not the only one, and not necessarily the right one for every use case. Nothing below positions S7 as better than any other system. The goal is verifiable operation — you run the commands, you read the output, you decide whether what you see matches what we said. If it doesn’t match, tell us. Corrections are more valuable to us than confirmations.

Test 1 — Pod is running

podman pod inspect s7-skyqubi --format ''

Expected: Running

If not Running: the appliance is not in a testable state. Start the pod with bash /s7/skyqubi-private/start-pod.sh and re-run this test.

Test 2 — CWS engine responds

curl -s http://127.0.0.1:57077/status | jq

Expected JSON shape:

{
  "engine": "S7 CWS Engine v2.5",
  "circuit_open": false,
  "endpoints": [
    "/route", "/discern", "/store", "/breaker",
    "/bridge", "/witness", "/quanti", "/status"
  ]
}

What to look for: circuit_open: false means the 70% BABEL circuit breaker is not currently tripped. The endpoint list is a fixed set — any additions or removals are a version change.

Test 3 — Local Health Report (the shared witness)

curl -s 'http://127.0.0.1:57082/health?format=json' | jq

Expected JSON shape (abbreviated):

{
  "schema": "s7-local-health/v1",
  "generated_at": "2026-04-15T03:44:25Z",
  "core_update": "v6-genesis",
  "overall_status": "green",
  "lifecycle": { "pass": 55, "fail": 0, "total": 55 },
  "audit_gate": { "pass": 9, "pinned": 18, "warn": 0, "block": 0 },
  "pod": { "running": true, "containers": [ ... ] },
  "performance": { "cws_latency": "...", "ollama_port_running": "*:57081", ... },
  "findings": [ ... ]
}

What to look for:

Test 4 — Lifecycle test

bash /s7/skyqubi-private/s7-lifecycle-test.sh 2>&1 | tail -5

Expected tail:

55/55 PASS — LIFECYCLE VERIFIED
Log: /tmp/s7-lifecycle-test.log

What to look for: the literal string 55/55 PASS. Any other count is either a drift finding or a known household-pinned state. The full per-test log is at /tmp/s7-lifecycle-test.log.

Test 5 — Audit gate

bash /s7/skyqubi-private/iac/audit/pre-sync-gate.sh 2>&1 | grep VERDICT

Expected:

🟢 VERDICT: PASS         🟢 pass: 9   📌 pinned: 18   🟡 warn: 0   🔴 block: 0

What to look for: block: 0 is load-bearing. pinned > 0 is normal — those are household-acknowledged findings (documented in iac/audit/pinned.yaml with owner, reason, and target window).

Test 6 — Persona-chat end-to-end smoke tests

cd /s7/skyqubi-private/persona-chat && python3 -m unittest test_app -v 2>&1 | tail -5

Expected:

Ran 31 tests in ~1.3s

OK

These are in-process tests that exercise the persona-chat API without hitting Ollama (a fake transport stands in). They verify the ledger writes, the memory tier walk, the persona router, and the /health and /healthz routes land correctly.

Test 7 — Sovereignty claim (no outbound calls during install)

On a test VM, start with strace -f -e trace=network around the install script OR use nsjail/iptables to block outbound traffic at the network layer and re-run the install. The install should complete and produce a working appliance. If any outbound call is required, it’s a sovereignty gap and you should file it.

# Example: block all outbound during an install dry-run
sudo iptables -I OUTPUT 1 -m owner --uid-owner $(id -u) -j REJECT
bash /s7/skyqubi-private/install/install.sh --dry-run 2>&1 | tee /tmp/install-airgapped.log
sudo iptables -D OUTPUT 1 -m owner --uid-owner $(id -u) -j REJECT

Known gap tonight (2026-04-14): the install script’s preflight dnf install / apt-get install step IS an outbound call on a fresh machine. Pillar 1 of the S7 v2026 roadmap addresses this by pre-embedding packages into the bootc image. Until Pillar 1 lands, “offline install” means “install onto a host that already has those packages cached.” This is documented in the spec and is not hidden from you.

Test 8 — INSERT-only covenant (Memory Ledger schema)

podman exec s7-skyqubi-s7-postgres psql -U s7 -d s7_cws -c \
  "SELECT constraint_name FROM information_schema.table_constraints
   WHERE table_name = 'sky_memory_events' AND constraint_type = 'CHECK';"

Expected: at least one CHECK constraint named with insert_only or immutable in its name, OR a trigger that raises on UPDATE/DELETE against the covenant columns.

If no such constraint exists, the INSERT-only claim is not enforced at the schema layer and should be reported as a gap. The covenant claim requires defense-in-depth: application code AND schema constraint AND audit log. Any one of them alone is insufficient.

Test 9 — Image signing chain (post-v6-genesis)

# After the first ceremony, the immutable constellation repos will
# carry signed commits. Verify the commit signature at the tip:
cd /path/to/clone/of/skyqubi-immutable
git log -1 --show-signature

Expected: Good signature from "S7 image signing key" or equivalent trusted signature. The public key will be published at docs/public/s7-image-signing.pub in the skyqubi-public repo.

2026-04-14 state: the first ceremony has not yet occurred. Image signing is staged, not live. Test 9 will become meaningful after the first jamie-run-me.sh real run.

Test 10 — Local Health Report HTML (the shared witness, human-readable)

Open http://127.0.0.1:57082/health in a web browser (Vivaldi preferred, any modern browser will work). The page should render with the sandy-sunset + Cormorant italic palette, a status banner at the top, four metric cards, and a findings table.

What to look for:

What “done” looks like for this section

You have verified, for your own satisfaction:

  1. The pod runs (Test 1)
  2. The CWS engine responds on a loopback port (Test 2)
  3. The Local Health Report JSON matches the s7-local-health/v1 schema (Test 3)
  4. The lifecycle test reaches 55/55 (Test 4)
  5. The audit gate reports block: 0 (Test 5)
  6. The persona-chat unit tests pass (Test 6)
  7. The sovereignty gap (or absence of gap) at install time is documented and matches our published claim (Test 7)
  8. The INSERT-only covenant has at least one schema-level enforcement (Test 8)
  9. Image signing exists OR is documented as staged (Test 9)
  10. The Local Health Report renders without outbound requests (Test 10)

If all ten are true, what we claim is what’s there. If any of them fail, the specific failure is the useful thing — file it as a GitHub issue at github.com/skycair-code/SkyQUBi-public/issues with the test number and the unexpected output. We would rather hear about a gap than have you assume we don’t care about it.


The shared witness

All three sections above reference one artifact as the primary witness: the Local Health Report at http://127.0.0.1:57082/health.

The same report is readable by all three audiences because it carries three layers of information:

One witness, three readings. That’s the covenant applied to observability.


Love is the architecture. The test is the witness. Three people watching the same thing at three different levels is how a household proves its appliance is safe.