Why a fixed audit menu
You can ask any read-only question of your Singlewire system in plain English. So why does mcsinglewire ship a fixed menu of fourteen ready-made audits on top?
This page explains what the menu buys you that free-form questions don’t.
What a menu entry is
Section titled “What a menu entry is”A menu entry is a phrasing that’s been written once, vetted, and stored. When you pick Offline devices from the menu and fill in “24” for hours back, what really happens is the same as if you’d typed:
“Find IP speakers and other endpoints that haven’t checked in within the last 24 hours. For each stale device, list its name, how long since it was last seen, and which site it belongs to. End with a one-line summary of stale vs. healthy.”
The menu doesn’t do anything different — it just types the question for you, exactly the same way every time.
The cost of asking it yourself every time
Section titled “The cost of asking it yourself every time”You can absolutely ask “find offline speakers” in plain English and Claude will figure it out. So can your colleague next Tuesday. Probably the same way. Probably.
The trouble is “probably”. Asking the same question in your own words has at least three failure modes:
- Drift. Your second time around it’s “show offline speakers” — same intent, slightly different phrasing. The answer comes back in a slightly different shape. Comparing this week to last week is harder than it should be.
- Implicit assumptions. “Offline” without a window means Claude has to pick one. Usually it’s 24 hours, but not always. Across daylight-saving transitions, time-zone differences, or weekend coverage gaps, that ambiguity bites.
- Drift across people. Two people on the same team will phrase the same audit differently. The system answers correctly each time. But the records of those calls don’t line up cleanly.
None of these are bugs. They’re just how language works. The fix isn’t “phrase more carefully” — that’s been tried for fifty years and it doesn’t scale.
What the menu buys you
Section titled “What the menu buys you”A menu entry fixes the phrasing once, then lets you tune just the parts that should vary (the time window, the site name, etc.). The result is:
- The same answer shape every time. Run Offline devices today and three weeks from now: same structure (stale list, healthy summary, follow-up question). You can paste either result into a weekly report template.
- Records that compare cleanly. Same audit name → same kind of log entry. Counting “how often did we check offline speakers this quarter?” is a single query, not a fuzzy text search.
- Less room for surprise. The phrasing explicitly says “list only, no changes”, “end with this summary line”, “if the data isn’t there, say so”. The AI has less wiggle room to do something unexpected.
- A surface anyone on the team can hit. It’s a menu. You don’t need to remember how the question is phrased — you pick it and fill in one value.
When asking plainly is still right
Section titled “When asking plainly is still right”The menu isn’t the right move for everything. Reach for a free-form question when:
- The question is new — nobody’s asked it this way before, so there’s no menu entry for it. Asking Claude directly costs nothing extra.
- The question is exploratory — “what kinds of devices does this system know about?” The right answer is whatever Claude finds, not a fixed shape.
- You’re investigating — the Investigating an incident tutorial shows the pattern: anchor with a menu audit, drill into unexpected results with a plain question.
Why we don’t ship hundreds of menu entries
Section titled “Why we don’t ship hundreds of menu entries”Singlewire’s API has a few hundred read endpoints. A reasonable next thought is: “if menu entries are so good, why not have one per endpoint?” Two reasons:
- A menu of fourteen is something you can scan. A menu of three hundred is a wall.
- Each new entry has to be vetted. That’s an explicit choice, not an accident — every entry on the menu has been written to produce a stable, useful answer, with explicit read-only intent where it matters. Auto-generating hundreds would either lose that care or make every spec update a re-audit.
The current count is fourteen because that’s how many distinct audit needs the team has identified as worth pre-writing. It will grow when new patterns emerge.
