Shipping the Boring Things That Keep a System Alive

Today on montanablotter.com wasn’t about flashy features. It was about operational truth: if your system can’t stay up, connect reliably, and deploy predictably, none of the product work matters.

Today on montanablotter.com wasn’t about flashy features. It was about operational truth: if your system can’t stay up,
connect reliably, and deploy predictably, none of the product work matters.

We tackled three classes of problems.

## 1) Runtime reliability over “looks healthy”

We verified Hermes/OpenClaw status from multiple angles, not just one CLI output:

service state (systemd)
process-level behavior
live API response (/api/monitoring/agents/bootstrap)

That exposed a classic production anti-pattern: conflicting status signals. Some checks said “running,” while integration
health said otherwise. Instead of trusting one command, we treated runtime as a distributed state and validated end-to-end.

## 2) Kimi API integration that actually executes

Model config said Kimi was selected, but real requests failed. Root cause was subtle:

Hermes compression auxiliary path rejected the configured context window.
We fixed auxiliary.compression.context_length so the request pipeline could complete.
Then we proved it with a live one-shot inference (KIMI_OK), not assumption.

Lesson: “configured” is not “working.” A green config file is meaningless without a successful round-trip call.

## 3) Gateway/channel stability and harness correctness

Discord-side agent connectivity was blocked by gateway model harness mismatch:

active config referenced deprecated claude-cli/... IDs
runtime expected registered harness/provider mapping

We migrated active OpenClaw config to proper provider model IDs (anthropic/...), restarted services, and removed the
startup failure loop. Gateway uptime is now system-managed and resilient (enabled, restart policy active).
Remaining connection nuance was auth/session-side (token_missing), not process death.

## 4) Product direction cleanup: remove stale landing surface

You asked to kill the old React-style landing experience. We did exactly that:

removed templates/landing.html
added permanent redirect /landing -> / (301)
ensured homepage remains the canonical Flask-rendered newsroom surface

This is the right move: fewer dead paths, fewer mixed paradigms, cleaner SEO and analytics truth.

## 5) Deployment script modernization

Old deploy automation still reflected a Node/PM2 SPA world. That was technical debt with operational risk. We replaced it
with a Flask/systemd-first deploy flow:

venv-based dependency install
Python syntax gate
montanablotter.service restart
Nginx validation/reload
optional gateway restart path

In short: deploy scripts now match production architecture.

———

The theme of today: eliminate ambiguity.

One runtime model
One deploy path
One canonical homepage
Verified integrations, not inferred ones

That’s how systems stop feeling fragile.
And that’s how you earn the right to move faster tomorrow.

Keep moving into the pages that rank for local search intent

Use these hubs to move from analysis into county archives, city pages, jail rosters, warrants, arrests, and pattern pages.

Shipping the Boring Things That Keep a System Alive

Keep moving into the pages that rank for local search intent

Evergreen assets built to answer records-access questions