Home › Topics › Uptime & Availability › Uptime Evidence

Building the Uptime Evidence File


Every IT department eventually meets the sentence. It arrives in a meeting, delivered with total confidence by someone from another department: "the network is always down." You know it is not true. You were there for both actual outages this quarter, you fixed them, and they totaled less than an hour. But you know it the way you know what you had for lunch last Tuesday, vaguely and from memory, and the sentence lands anyway. In a room where nobody has data, the most confident memory wins.


The antidote is not a speech. It is a file. I lost that argument twice, with the better memory and no paper. I won it the third time with one table I had been keeping for four months and had not yet shown anyone. What the table held was a maintained, boring, factual record of what actually happened. It showed how available each service really was, every incident with its duration and cause, and the trend across quarters. It also showed the part almost nobody keeps: the failures users never felt because your redundancy quietly absorbed them. Call it the uptime evidence file. It is not a marketing document and it is not a diary. It is a ledger, and its power comes precisely from how unexciting it is.


This guide covers what goes in the file and how to assemble it in about thirty minutes a month from data your monitoring already collects. It covers the three formats for presenting it. It also shows how to use it: to correct the mythology, anchor budget requests to specific outages, and hand a defensible record to whoever asks. None of it requires new tools. It requires a ritual and a folder.


In this guide

  • Why memory loses to a ledger
  • What goes in the file
  • Collecting it without heroics
  • Presenting it: three formats
  • Using it: receipts, budgets, handoffs
  • Tone rules
  • Start this month

Why memory loses to a ledger


Human memory is spectacularly biased about outages, in ways that all point the same direction. People remember the two mornings email was down vividly — frustration writes durable memories. They remember none of the two hundred and fifty ordinary days surrounding them, because nothing writes a memory of a normal Tuesday. Worse, unrelated problems get filed under "the network": a slow website, a cloud provider's bad afternoon, a laptop with a dying disk. Each becomes another anecdote in the case file titled always down. Infrastructure work has a cruel asymmetry: your failures are public events and your successes are the absence of events. Nobody experiences the outage that redundancy prevented.


You cannot fix the bias with arguments, because an argument between two memories is just two confidence levels colliding. You fix it by being the only person in the room with a written record. Not a better story — a ledger. When someone says "always down" and you can say "three incidents this quarter, forty-one minutes total, here is the list," the conversation changes shape instantly. Nobody argues with a boring table, precisely because it is boring: it sounds like accounting, not like a defense.


What goes in the file


Five sections. Together they answer every question anyone ever actually asks about reliability. How good is it? What went wrong? Is it getting better? What did maintenance cost? What almost happened?


1. Monthly availability, per service

One number per key service per month — the percentage of measured time it was working — computed the same way every month, using rules written down in advance. The two guides that make this number defensible are this pillar's foundation. One covers what counts as "down" in the first place (see Defining "Down"). The other covers how to do the arithmetic without fooling anyone, including yourself (see Honest Availability Math). The critical property is consistency. A number computed one way in March and another way in April is not a record; it is two anecdotes wearing a spreadsheet. Track your top handful of services — internet access, sign-in, email, the main application, file shares — not all fifty things you monitor.


2. The incident ledger

The heart of the file: one line per user-visible incident, seven columns, written within a day of the incident while the facts are fresh. Date, service, duration, scope (who felt it), cause, fix, follow-up. One line — the discipline of the format is the feature. A one-line entry always gets written, while a five-paragraph postmortem waits for a free afternoon that never comes. A filled example:


DateServiceDurationScopeCauseFixFollow-up
Mar 4Email22 minAll staff, inbound onlyFiltering service hung after automatic updateService restartedService now checked directly; updates moved to monthly window
Mar 18Internet51 minWhole siteISP fiber cut nearbyISP repair; nothing on our sideLogged for ISP accountability; second-line quote refreshed
Apr 9File server12 minFinance shareDisk filled by runaway application logLog cleared, growth cappedDisk warning threshold lowered to 80%
May 22Wi-Fi38 minEast wingFailed power injector on access pointSwapped from shelf spareSpare restocked; injectors added to spares list

Notice what the columns quietly document: the fix column shows competence, and the follow-up column shows learning. The same cause should not appear twice, and when it does not, the ledger proves it. Write causes in neutral, blameless language: "configuration applied without review," never a name. The ledger records what the system did, not who to blame. The moment it becomes a blame instrument, people stop feeding it, and it dies.


3. Trend lines, quarter over quarter

Individual months wobble; the trend is the story. Keep a small running table — incidents, total user-visible minutes, availability for the critical tier — and let it accumulate:


QuarterIncidentsUser-visible minutesCritical-tier availability (business hours)Note
1st421099.55%Two traced to the single ISP line
2nd38599.82%No repeat causes
3rd34199.91%Longest incident 22 min

Three quarters of this and "always down" simply has nowhere to stand. If the trend ever points the wrong way, you want to be the one who notices first, with the table in hand and causes attached. (For translating the percentages into plain downtime-minutes for a non-technical reader, the tables in The Nines, Translated do the work.)


4. Planned windows: used versus budgeted

If you run scheduled maintenance windows — and you should — record each one: date, purpose, announced duration, actual duration. Two lines per month for most small shops. This section proves two things at once. Planned work is disciplined (windows announced in advance, finished inside their slot). And planned time is honestly separated from unplanned time in your numbers, per the rules your outage definition sets. It is also where the difference between "IT took the system down again" and "the third-Thursday window ran twenty minutes, as announced" gets settled by a row in a table.


5. The near-miss log

The section nobody keeps and everyone should: failures that genuinely happened but that users never felt, because something you built absorbed them. Every one of these is an outage that did not happen — and by definition, nobody experienced it. Redundancy doing its job is perfectly invisible work. Without a log, the money spent on resilience produces no visible evidence of ever having been worth it. One line per event fixes that:


DateWhat failedWhat absorbed itWithout it
Feb 12Primary DNS server crashed at lunchSecondary DNS answered all lookupsSite-wide "internet is down" for an hour-plus
Apr 3Ninety-second utility power cutUPS carried the server room throughHard shutdown of every server; hours of recovery
May 30Disk died in the storage arrayHot spare rebuilt automaticallyFile server restore from backup; most of a day
Jun 17Primary firewall hung during a firmware crashStandby unit took over in secondsWhole-site outage until manual recovery

The log also doubles as an early-warning list. A component that keeps almost failing is telling you something. The near-miss rows are where that pattern becomes visible before it graduates into the incident ledger.


Collecting it without heroics


Here is the good news: if you run a monitoring system, most of the file already exists. The availability numbers are in its history. The incidents are in its alert log, with timestamps for failure and recovery. What the tool cannot know — cause, fix, follow-up, scope — is exactly what fits in the one-line ledger format. It takes minutes to write while the incident is fresh. (If you have no monitoring at all yet, that is the prerequisite; start with Network Monitoring Fundamentals — a file without measurements behind it is just opinions in a folder.) Mapped out, the sourcing looks like this:


File sectionWhere it comes fromMonthly effort
Availability numbersMonitoring history, computed by your written rulesMinutes — look up and record
Incident ledgerAlert-log timestamps plus the columns only you can fillOne line per incident
Trend tableThe file's own earlier monthsOne new row
Window ledgerYour maintenance calendar and notesA line or two
Near-miss logFailover events and hardware warnings in monitoringOne quick sweep

The whole discipline is a monthly ritual of about thirty minutes, same calendar slot every month, first week for the month just ended:

  • Pull the availability numbers for your key services from monitoring history, computed by your written rules. (5 min)
  • Reconcile the incident ledger against the alert history: every user-visible incident has a line, every line has all seven columns filled. Anything missed gets written now, while some memory remains. (10 min)
  • Sweep for near-misses: failover events, hardware warnings, disks replaced, UPS activations. One line each. (5 min)
  • Update the window ledger: planned windows held, announced versus actual duration. (2 min)
  • Extend the trend table with the new month. (3 min)
  • Write the one-page snapshot (next section) and file everything in one folder, one subfolder per month, named consistently. (5 min)

Tip: put the ritual in your calendar as a recurring appointment and treat it like patching — non-negotiable, boring, done. The file's entire value comes from being complete. An evidence file with holes in it stops being evidence, because the first gap anyone finds discredits every other page.


Presenting it: three formats


The raw file is for you. What others see is one of three renderings, each matched to a moment.


The monthly one-pager

Send one page, once a month, whether or not anyone asked. Include availability per key service with a simple met-target mark, plus incident count and total minutes with one-line summaries. Include windows held and one near-miss if there was an instructive one. No graphs of CPU, no jargon — a page a manager reads in ninety seconds. This is the uptime-focused sibling of the broader monthly report pattern covered in Building a One-Page Network Health Scorecard. If you already produce a scorecard, the availability lines of it can simply be this, backed by the file.


The annual review pack

Once a year, staple the trend table to the twelve one-pagers. Add a short cover note: the year in three sentences, the causes that repeated, what was fixed so they stop repeating. Attach the budget asks that the evidence supports. This is the document that turns review season from vibes into arithmetic, and it assembles in an hour because the twelve months already exist.


The incident brief: the awkward Monday

After a visible outage there is always a Monday, and on that Monday someone senior will want to understand what happened. The move that changes everything: the brief is already written and already sent — before being asked. Four short labeled sections, plain language, no acronyms:


WHAT HAPPENED   Email delivery paused for 22 minutes
                Tuesday morning (9:41 to 10:03).

IMPACT          Incoming messages queued and arrived after
                recovery; none were lost. Outbound mail
                was unaffected.

CAUSE           The mail filtering service hung following
                an automatic vendor update.

WHAT WE CHANGED The service is now monitored directly, so
                a repeat is caught in about a minute, and
                its updates now run in the monthly
                maintenance window instead of automatically.

Ten minutes to write, because the ledger line already contains every fact. The difference in how the room treats you when the explanation precedes the question is difficult to overstate. Asked-for explanations read as excuses; unasked-for ones read as command of the situation. Same words, opposite effect.


Using it: receipts, budgets, handoffs


Correcting perception. When "always down" surfaces, do not argue the adjective — produce the specifics, gently. "This quarter: three incidents, forty-one minutes total, none with a repeated cause; here is the list." Then stop talking. The file does not win by persuading; it wins by being checkable. Often it also surfaces the real complaint hiding underneath: one slow application or one bad Wi-Fi corner. That is an actual, fixable ticket instead of a mythology.


Anchoring budget asks. The evidence file converts requests into conclusions. "We need a second internet line" is an opinion and gets opinion-grade responses. "Both extended outages this year trace to the single ISP (internet provider) line. See rows two and seven in the ledger: ninety minutes of company-wide downtime. Here is the quote for a second line with failover." That is an argument that arrives pre-won. The decision-maker can fund it or decline it, but nobody can claim the problem is imaginary. Your near-miss log argues the other side of the same coin — it shows the redundancy already bought doing its job, which is what keeps it funded.


Settling disputes with a provider. Redbrook Engineering's internet provider sent a quarterly report showing the line at 99.9%, which under the service level agreement meant no credit was owed. Redbrook's incident ledger had three provider outages that quarter, each with start and end timestamps from checks run inside the building, totaling just under four hours. The admin replied with one email: the three ledger rows, with the alert-log timestamps attached. A sentence asked which of the two records the provider would like to reconcile against. The account manager came back the next morning with a corrected report and a service credit. Nobody at Redbrook had to raise their voice or their ticket priority. The email was eleven lines long, and nine of them were a table.


Reviews and handoffs. At performance review time, the file is the difference between "things seemed fine, I think?" and a documented record of availability delivered, incidents resolved, and causes eliminated. And when you eventually hand the network to a successor — or inherit one yourself — the file is the handoff. It holds every past incident with its cause and fix, every known weak point, every almost-failure. An inherited evidence file is the difference between starting from institutional memory and starting from zero. Be the admin who leaves one behind.


Tone rules


The file has exactly one voice: flat. Never triumphalist — no "another perfect month!" This month's victory lap becomes next month's hostage the first time the number dips. It also teaches readers that the file is advertising. Never defensive — no pre-emptive justifications attached to bad numbers, no adjectives doing damage control. A bad month is recorded in the same dry format as a good one: the number, the incidents, the causes, the follow-ups. That flatness is not a style choice; it is the mechanism of credibility. A record that reads the same in good months and bad is a record people believe in both. After enough months of that, your numbers get accepted without argument — which was the entire point.


And sometimes the ledger will side with the critics. The trend table points the wrong way; a cause repeats that should not have; a service really is having a bad year. Resist every instinct to soften it. Record it flat, bring the fix plan in the same breath, and let the file's history work for you. A record that has been honest about good months is believed when it reports a bad one. That means your diagnosis and your fix plan get believed too. An evidence file that only ever exonerates you would be worthless — its willingness to convict is exactly what gives its acquittals weight.


Start this month


One warning about timing, and it is the reason to act now rather than someday: the file is only powerful if it predates the complaint. A ledger that has been quietly accumulating for months is evidence. A ledger hastily assembled the week after an accusation is a defense exhibit, and everyone in the room can tell the difference. You cannot retroactively grow credibility — the record's age is its authority.


So start ugly and start now. A spreadsheet with seven columns is a fully legitimate evidence file. This month: write the ledger lines for any incidents you still remember clearly (mark them "reconstructed" — honesty about the seam matters). Pull this month's availability numbers, book the recurring thirty-minute slot, and create the folder. That is the whole launch. In one quarter you will have a trend table; in a year, an annual pack. At some meeting in between, someone will say the sentence — and for the first time, you will simply open a file.


Where to go next


The evidence file sits on top of everything else in this pillar. That includes availability numbers computed honestly, an outage definition everyone accepted in advance, and checks that test real services rather than mere ping. It also includes targets your architecture can actually meet. Each of those has its own guide, collected at Uptime & Availability. If you want the broader monthly reporting habit the file feeds into, the scorecard guide linked above is the natural companion. The ritual is thirty minutes a month, which is probably less time than the sentence has already cost you this month. It is the only reply to the sentence that does not sound like one.


Put this into practice with SysPeep Monitor

An evidence file is easiest when the measurements collect themselves. SysPeep Monitor watches your devices and services around the clock and keeps the history. It turns that history into reports and dashboards you can pull from one web console - most of your monthly ritual, already assembled. The free trial installs on a Windows machine in minutes.

Download Free TrialLearn About SysPeep