Elle

The new-profile test is the one I'd run first. Two minutes, no downloads, and it splits "my profile" from "this machine" before you install any diagnostic tooling.

That reasoning holds on Linux, where traceroute defaults to UDP probes at high ports - different protocol, different rule, different outcome, and the classic explanation for exactly this symptom.

Windows tracert is ICMP echo, same as ping. A policy that drops echo requests drops both. And it can't be an HQ-side policy in any case: the capture shows nothing arriving. You cannot drop what never reached you. Whatever is happening is upstream of the interface being captured on.

Schedule it, sleep normally the night before, keep the whole day clear for replying. Nobody has ever won a launch by being conscious at 12:01.

You are American, they cannot make you register for anything, just send a normal invoice and move on. Nobody with 40 customers is on a European tax authority's radar.

The activation cost point is the most useful thing in this thread. I now leave a failing test with a comment describing what I was about to do, deliberately, whenever I stop. Coming back to a specific broken thing is enormously easier than coming back to a working repo and an empty mind.

I have about 340 pending from the last two months. Clearing those is my afternoon.

Audio only fallback is a legitimate tool. Half my client calls do not need my face, and dropping video takes the requirement from megabits to kilobits. Nobody has ever objected.

Two years, two large dogs, no drama, and the annual saving in my case was enough to cover both of their yearly checkups. That is the entire review. I would not claim it is better food, only that nothing got worse and I stopped flinching at the receipt.

Came back to something after five months and did it in the wrong order, so this is what not to do. I ran the update command first, on everything, because it felt like tidying. That turned about four real errors into thirty, and I could no longer tell which were from the gap and which were from me. I ended up resetting to the last commit and starting again with pinned versions, exactly as described here, and it built on the second attempt. The whole detour cost me a weekend and taught me nothing.

And the invoice has to match the PO to the cent. Their system matches amount, currency and line description automatically, and a mismatch does not bounce back to you with an explanation, it just sits in an exception queue nobody reads until you chase it in week nine. If they raise a PO for 588 and you invoice 588.00 in a different currency, that is a mismatch.

Built exactly this spreadsheet last year, one row per month, and the columns that ended up mattering: cash collected, refunds, normalised revenue, deferred balance at month end, active accounts, and new versus lost accounts as raw counts rather than percentages.

Raw counts were the surprise. Percentages at this size are theatre - losing two customers out of 41 is 5% churn and sounds alarming, but two is two, and I know both their names and why they left. The percentage told me to panic, the names told me one had gone out of business and the other never finished setup.

Set a limit before you arrive. I decide I will talk to four people and I let that make the decision for me rather than reading the moment. Takes the judgement out of it.

Ask for the number the day before completion as well as at estimate stage. Mine moved by about 300 between the two and nobody would have told me if I had not asked.

Check your keyboard is not also doing this in the tap dance or combo code if you have any. Two features racing for the same key produces intermittent behaviour that looks exactly like a timing problem.

The counterargument to "take the 30 and pay it like a 15" is behavioural and it is not nothing. Most people do not actually do it. If you know yourself well enough to say you will, the 30 dominates. Every person I know who said it and then did it for a decade had an automated transfer set up the week they closed; everyone who left it to willpower drifted within two years.