Internet Outage Warlord

Overview

Home internet goes down. The household revolts. Someone (you) is deputized to fix it. The failure can be in seven different places — your device, Wi-Fi radio, router LAN, router WAN, DNS resolver, ISP last-mile, or a regional/platform outage — and the difference between them is the difference between "reboot the router" and "call the ISP with evidence", or "do nothing, it's Cloudflare".

Internet Outage Warlord does three things:

1. Scopes the fault — a layered probe sequence that isolates WHERE connectivity died, from the laptop outward: link → gateway → DNS → public endpoint → ISP headend, plus known-platform-outage heuristics. The output is a scoped verdict (WIFI, LAN, ROUTER-WAN, DNS, ISP, PLATFORM, UPSTREAM-COMPLETE), not vibes.
2. Prescribes — the right first action per verdict: which box to power-cycle and in what order, what NOT to touch (don't factory-reset at 9PM), when to switch to mobile data or a phone hotspot, how to ride out an ISP outage with a downtime SLA credit claim.
3. Builds the evidence log — timestamped probe results persisted to disk. When the ISP says "we see no outage in your area", you paste the log: 47 minutes of 100% packet loss to the gateway at 19:02–19:49, with model/serial of your equipment, traceroute capture at fault time, and a ready-to-send complaint with statutory citation where applicable.

The core principle: an outage you can't scope, you can't fix — and an outage you can't document, you can't claim.

When to Use

• "The internet is down" / "Wi-Fi is down" / "websites won't load"
• "Internet is slow / videos buffer constantly" → run diagnose in degrade mode (it logs latency/jitter/loss, not just up/down)
• "Is it just me or is everyone down?" → platform-outage heuristics + optional online check
• Preparing an ISP complaint / SLA credit claim after repeated outages
• Setting up the monitor before problems start (proactive monitoring on a home server / Raspberry Pi / NAS cron)
• Deciding whether to work from a café or wait it out (verdict + ETA heuristics)

Don't use for: enterprise networks with NOC teams (different tooling), fiber cuts you can physically see (call the ISP immediately), or suspected compromise/hacking of your network (different threat model — this is availability, not security).

How It Works — Numbered Workflow

1. Run diagnose the moment connectivity degrades. The script probes in order, stopping at the first hard failure layer: link (interface up? IP?) → gateway ping → DNS resolution of a stable domain → ping a public anycast IP (1.1.1.1, 8.8.8.8) → HTTP HEAD to 2–3 majors. Each hop gets logged with a timestamp to outage_log.jsonl.
2. Read the verdict — a scoped fault label plus the confidence and the immediate prescription. The script prints the action sequence (power-cycle order: modem → wait 2 min → router → wait 2 min → re-probe), never a factory reset.
3. If DNS: the script tests alternate resolvers (1.1.1.1, 8.8.8.8, 9.9.9.9) and prescribes the temporary switch (and how to revert).
4. If ISP/platform: check the heuristics — can you reach anycast IPs but not HTTP? Are specific platforms failing while others work? The script optionally checks a public status aggregator (if you have any connectivity at all, e.g. via mobile data) and logs the result.
5. After the outage, run report — the evidence package: outage windows extracted from the log (≥2 consecutive failed probe sets), total downtime, per-window scope, equipment inventory, and a filled-in ISP complaint letter (with EU/UK/US regulatory hooks where relevant).
6. Optional monitor mode — run from cron every 1–5 min; silent while healthy, logs and (optionally) runs a notification command on state change. This is how the warlord catches the 3AM outages you sleep through and the intermittent degradation that never shows up when the technician visits.
7. Restore your normal config after any temporary DNS/workaround changes (the log records what changed).

Verdicts & First Actions

Quick Start

python3 scripts/outage_warlord.py diagnose          # scope the fault now
python3 scripts/outage_warlord.py diagnose --degrade   # slow-but-up mode: latency/loss stats
python3 scripts/outage_warlord.py report         # evidence log → ISP complaint package
python3 scripts/outage_warlord.py monitor --interval 120   # cron mode, logs state changes
python3 scripts/outage_warlord.py report --sla-credit --contract "100Mbps, SLA 99.5%"   # credit claim

Log: outage_log.jsonl beside the script (override --log). Network probes use only stdlib (socket, subprocess ping) — works on Linux/macOS; on Windows use WSL or the --no-ping fallback (HTTP-only probes).

The ISP Complaint Package

report generates:

• Outage table: start, end, duration, verdict at each window, probes that failed
• Equipment inventory: modem/router model + serial (from setup, entered once) — proving it's not "your router" before the ISP blames it
• Complaint letter (plain text): dates/durations, statement that customer equipment was power-cycled per ISP instructions during windows (from your action log), reference to contractual availability (e.g. "99.5% monthly availability" = max 3h39m downtime/month; your log shows 11h), and the remedy requested (credit, escalation, fix commitment).
• Where relevant, cites: US — FCC complaints; EU — Art. 22 Universal Service Directive availability/quality obligations... see references/isp-evidence-playbook.md for jurisdiction specifics.

Common Pitfalls

1. Rebooting everything at once. It "fixes" ~60% of outages and destroys the evidence: you lose the fault state the ISP needs to see (line stats at fault time), and if it's an ISP-side fault you've learned nothing. Scope first with diagnose, act second.
2. Factory-resetting at night. The 9PM panic reset wipes your PPPoE credentials, Wi-Fi names, port forwards. The warlord never recommends factory reset as a first-line action — power-cycle only, and only the scoped layer.
3. Trusting the router's "Internet: Connected" light. DHCP-lease-OK but no actual upstream happens constantly; the script probes public endpoints, not indicator LEDs.
4. Blaming DNS for everything. "Internet down" that's actually DNS is common but over-diagnosed; the layered probe distinguishes "names fail, IPs work" (DNS) from "everything fails from the gateway out" (ISP).
5. No log = no claim. ISP support scripts are designed to wear you down ("unplug for 30 seconds"). A timestamped log with durations and a letter citing the SLA percentage changes the conversation from support-tier roulette to a documented claim.
6. Testing only one endpoint. google.com down ≠ internet down. The script probes anycast IPs + multiple HTTP endpoints + DNS to separate platform vs general failures.
7. Ignoring the modem's line stats. Coax SNR/fiber RX power visible in modem UI are the ISP's own telemetry; the reference explains how to capture them into the log (manual entry command) as fault-time evidence.

Verification Checklist

• diagnose on a healthy connection prints UPSTREAM-OK and appends to the log
• diagnose with network disconnected (or --simulate isp) prints a scoped verdict + prescription
• report after ≥2 logged failures prints outage windows + complaint letter
• monitor --interval N writes only on state change (healthy → failing → healthy = 2 lines)
• Simulate modes work: --simulate dns|isp|platform|wifi (for demos/tests without breaking anything)
• No non-stdlib imports (grep: requests|urllib3|scapy → empty; stdlib urllib.request IS used for HTTP probes)

One-Shot Recipes

"Internet is down" at home:

python3 scripts/outage_warlord.py diagnose
# → VERDICT: ISP (gateway ok, 100% loss to 1.1.1.1 and 8.8.8.8)
# → actions: 1. mobile-data check of ISP status page 2. power-cycle modem only ...
python3 scripts/outage_warlord.py monitor --interval 120    # passive until it recovers
python3 scripts/outage_warlord.py report --sla-credit       # when it's back

Chronic evening slowdowns:

python3 scripts/outage_warlord.py diagnose --degrade --label evening
# latency/loss table for the evening window; congestion pattern = ISP capacity issue
python3 scripts/outage_warlord.py report   # trend: 20:00-23:00 latency 4x baseline

Is Discord down or is it me?

python3 scripts/outage_warlord.py diagnose --target discord.com
# PLATFORM verdict if everything else probes fine

License

MIT (see LICENSE). No warranty; network probes run on your own equipment against your own connection.