The receipt · snapshot October 8, 2026 at 3:21 PM ET
NOBODY'S WATCHING.
THE LOGS ARE.
This site has run on its own since Aug 22, 2026. A script is scheduled to check it every 30 minutes, a watchdog every 15, and an AI repair pass gets called in when something breaks. The name isn't a vibe. It's the ops model.
Every number below is counted out of this site's own ops logs. Nothing is estimated, rounded up, or borrowed from the rest of the portfolio.
The numbers
- days live
- 48
- Aug 22, 2026 to today, counting both
- of 2,265 scheduled engineer sweeps ran
- 2,168
- scheduled every 30 minutes; 3 days ran well short
- sweeps that used zero AI
- 99.4%
- a bash script checked everything and went back to sleep
- of 10,840 page renders passed
- 10,832
- real headless browser, real pages
- watchdog probes
- 4,208
- scheduled every 15 minutes
- incidents opened
- 8
- site down, deploy stuck, bad build, bad deps
- fixed by the watchdog
- 4
- build gate green, and no deploy failed right after. 1 more passed its gate, then the next deploy failed anyway
- times a human got pinged
- 2
- when the machines were actually stuck
every single day…
48 DAYS, ONE SQUARE EACH
- All green
- Had work
- Partial coverage
- Had an issue
- Aug 22: 26 of 26 scheduled sweeps, 25 green, 1 with work
- Aug 23: 48 of 48 scheduled sweeps, 48 green
- Aug 24: 48 of 48 scheduled sweeps, 48 green
- Aug 25: 48 of 48 scheduled sweeps, 48 green
- Aug 26: 48 of 48 scheduled sweeps, 47 green, 1 with work
- Aug 27: 48 of 48 scheduled sweeps, 48 green
- Aug 28: 48 of 48 scheduled sweeps, 48 green
- Aug 29: 48 of 48 scheduled sweeps, 48 green
- Aug 30: 41 of 48 scheduled sweeps, 41 green
- Aug 31: 48 of 48 scheduled sweeps, 48 green
- Sep 1: 48 of 48 scheduled sweeps, 48 green
- Sep 2: 48 of 48 scheduled sweeps, 48 green
- Sep 3: 48 of 48 scheduled sweeps, 47 green, 1 with an issue
- Sep 4: 48 of 48 scheduled sweeps, 48 green
- Sep 5: 48 of 48 scheduled sweeps, 48 green
- Sep 6: 48 of 48 scheduled sweeps, 48 green
- Sep 7: 48 of 48 scheduled sweeps, 48 green
- Sep 8: 48 of 48 scheduled sweeps, 48 green
- Sep 9: 48 of 48 scheduled sweeps, 48 green
- Sep 10: 48 of 48 scheduled sweeps, 48 green
- Sep 11: 48 of 48 scheduled sweeps, 47 green, 1 with work
- Sep 12: 48 of 48 scheduled sweeps, 48 green
- Sep 13: 48 of 48 scheduled sweeps, 47 green, 1 with work
- Sep 14: 48 of 48 scheduled sweeps, 48 green
- Sep 15: 48 of 48 scheduled sweeps, 48 green
- Sep 16: 48 of 48 scheduled sweeps, 48 green
- Sep 17: 48 of 48 scheduled sweeps, 48 green
- Sep 18: 48 of 48 scheduled sweeps, 48 green
- Sep 19: 48 of 48 scheduled sweeps, 48 green
- Sep 20: 47 of 48 scheduled sweeps, 47 green
- Sep 21: 48 of 48 scheduled sweeps, 48 green
- Sep 22: 48 of 48 scheduled sweeps, 48 green
- Sep 23: 48 of 48 scheduled sweeps, 48 green
- Sep 24: 48 of 48 scheduled sweeps, 47 green, 1 with work
- Sep 25: 48 of 48 scheduled sweeps, 48 green
- Sep 26: 48 of 48 scheduled sweeps, 48 green
- Sep 27: 48 of 48 scheduled sweeps, 48 green
- Sep 28: 48 of 48 scheduled sweeps, 48 green
- Sep 29: 48 of 48 scheduled sweeps, 48 green
- Sep 30: 39 of 48 scheduled sweeps, 38 green, 1 with work
- Oct 1: 48 of 48 scheduled sweeps, 48 green
- Oct 2: 48 of 48 scheduled sweeps, 48 green
- Oct 3: 29 of 48 scheduled sweeps, 28 green, 1 with work
- Oct 4: 19 of 48 scheduled sweeps, 19 green
- Oct 5: 18 of 48 scheduled sweeps, 16 green, 2 with work
- Oct 6: 48 of 48 scheduled sweeps, 48 green
- Oct 7: 47 of 48 scheduled sweeps, 44 green, 1 with work, 2 with an issue
- Oct 8: 30 of 31 scheduled sweeps, 30 green
2,155 of 2,168 sweeps came back green. 10 found work to pick up and 3 flagged an issue. On 3 days the engineer ran fewer than three-quarters of its scheduled sweeps. Those are marked partial coverage: the monitor was down, so green isn't a claim we can make for them. Hover a square for the day's count.
HOW THE LOOP WORKS
01 · Every 30 min
The engineer
A bash sweep renders the key pages in a real browser, checks the git tree, checks Cloudflare, and reads the task queue. If it is all green it logs a line and exits. An AI model only gets called when the sweep finds actual work.
02 · Every 15 min
The watchdog
Cheap probes for a dead homepage or a stuck deploy. A failed probe opens an incident and spins up an AI repair pass. Nothing it writes ships unless the full production audit and build pass. Attempts are capped, with a cooldown.
03 · Every 5 min, if queued
The deployer
No AI. When a change is queued it runs the production audit, the dependency pins, the build and a live smoke test. Any gate fails, nothing ships. Five failures in a row halt it until a cooldown passes, and the halt itself becomes a watchdog incident.
04 · Every 5 min, if paged
The principal engineer
A script reads every warning any role posts to Slack. A new one sends an AI pass to find the root cause and fix it in a throwaway checkout, behind the same build gate. "Root-caused and closed" in the timeline is this pass.
05 · When it is stuck
The human
Past the attempt caps, the incident becomes an escalation with the evidence attached. The count is on this page.
THE INCIDENT LEDGER
8 opened. 4 fixed by the watchdog, 1 patched, then the next deploy failed a gate anyway, 2 gone on the next probe, 1 ran out of attempts and went up the chain.
- High-severity dependency CVE3
- Deploy stuck2
- Site down2
- Build failed1
THE DEPLOY GATE
Every deployer run has to clear a security audit, the dependency pins, a production build and a live smoke test. A run that fails a gate publishes nothing.
- Deployer runs12
- Shipped, smoke test passed2
- Nothing new to ship3
- Stopped at a gate7
straight from the logs…
THE LAST 20 THINGS THAT HAPPENED
- Task
Task closed
Add WebPage JSON-LD to /privacy/ and /terms/
- Fixed
Root-caused and closed
sharp CVE-2026-96889 via @astrojs/cloudflare→wrangler→miniflare→sharp<0.35.5 triggered production audit failure; watchdog repaired by bumping wrangler to 4.144.0 but dep-pins exemption wasn't atomic — 4 retries failed…
- Deploy
Deployed
Audit, build and live smoke test all passed.
- Incident
Incident opened: deploy-stuck
the deploy queue unprocessed for 22m — deploys not completing
- Human
Escalated to a human
Re-escalation (same issue, nothing new): deployer failed 5x 06:15–06:40 ET
- Didn't stick
Patched, then the next deploy failed: npm-audit-high
sharp override to 0.35.5 + npm audit fix; prod audit 0 vulns, build 7 pages clean
- Incident
Incident opened: npm-audit-high
production npm audit remains high+; update dependencies in a reviewed commit
- Task
Task closed
Failure diagnosis: Recover failed dependency audit prerequisite
- Human
Escalated to a human
Two dependency-audit tasks parked to hold/
- Task
Task closed
Failure diagnosis: Repair dependency audit release prerequisite
- Fixed
Auto-repaired: build-fail
build passes clean — tailwindcss present in node_modules, both audit and build gates green (7 pages)
- Incident
Incident opened: build-fail
deploy failed — astro build error
- Deploy
Deployed
Audit, build and live smoke test all passed.
- Task
Task closed
Add visible keyboard focus states across primary site interactions
- Task
Task closed
Add AboutPage JSON-LD schema to /about/ template
- Task
Task closed
Add ContactPage JSON-LD schema to /contact/ template
- Fixed
Auto-repaired: npm-audit-high
upgraded wrangler 4.138.0→4.144.0; undici vuln cleared; prod audit + build green
- Incident
Incident opened: npm-audit-high
production npm audit remains high+ after npm audit fix; repair requires a dependency upgrade
- Task
Task closed
Run unsupervisedmedia.com audience and monetization readiness baseline
- Fixed
Auto-repaired: npm-audit-high
bumped wrangler 4.128.0→4.138.0; audit clean (0 vulns), build green
HOW THESE ARE COUNTED
A script walks this site's ops logs (live and archived), its incident files and its closed task queue, counts what's there, and strips paths, hostnames and internal ids from the text. The numbers are one site's: this one. The rest of the portfolio keeps its own logs.
The page is a snapshot, not a live feed. This one was taken October 8, 2026 at 3:21 PM ET.
SAME INFRASTRUCTURE.
EVERY BRAND WE OWN.