What a Baseline Is Worth and When to Take It

Why this matters

Every shop has been told to take baselines and most of them do, in a fashion. The numbers go in a field somewhere and nobody looks at them again, and when somebody finally does look, two years later on a hard fault, the baseline turns out to be unusable. Not wrong. Unusable, because nobody can say what the machine was doing when it was taken.

A baseline is not a number. It is a number plus the conditions that make a future number comparable to it. Its value as evidence is fixed at the moment of capture by what you recorded alongside it, and no amount of later effort can raise it. That is the whole idea, and it decides both which baselines are worth taking and how much time each one deserves.

What a baseline actually is

The point of a baseline is to convert a later single reading into a difference. A single reading has to be argued against a published specification, which brings in every condition the specification was written under. A difference against your own earlier reading on the same machine sidesteps most of that, because the machine is its own reference.

That only works if the two readings were taken under conditions close enough that the difference means what you want it to mean. So a usable baseline carries, at minimum:

• The value, its units and its reference. Pressure with no statement of gauge or absolute is two different numbers. Temperature difference with no statement of which two points is a number nobody can retake.
• Where exactly it was taken. Not "at the unit." The port, the terminal, the conductor, the side of the component.
• The operating state. Running or off, how long it had been running, what stage or speed, what the load was doing.
• The conditions that move the number. Ambient, entering conditions, filter or screen state, supply voltage, whatever the physics says the reading depends on.
• The instrument and its last check date. Which is what lets a future reader decide whether a difference is real or is two instruments disagreeing.

Anything the future reader could derive from those, leave out. A baseline record padded with computed values is harder to read and gives a second place for an error to live.

The one gate

Before you spend time on a baseline, ask a single question:

Can a technician who has never been to this site reproduce these conditions from the record alone?

If yes, the baseline will still be answering questions years from now. If no, you are recording a number that will be argued about and then ignored. The gate has nothing to do with how careful the measurement was. A meticulously taken reading with no conditions attached is worth less than a rough one with the conditions written down, because the rough one can be repeated and the meticulous one cannot.

The two cases below are both real shapes, both from routine work, and they resolve in opposite directions.

Case one: the baseline that still works two years later

A commissioning visit on a residential air handler. The tech records external static pressure across the unit, notes the filter as new and names its type and nominal rating, records the blower speed tap, notes all supply registers open and the outdoor temperature, and stamps the manometer's identity and its last check date.

Two years later the customer complains of poor airflow. The tech re-measures with the same instrument type, which is not the same instrument and is why what follows is read as materially higher rather than as a number, sets the same blower tap, opens the registers, waits for the same running state, and finds static pressure has risen by roughly a third above the recorded baseline.

That difference is worth something specific. It is not a comparison against a published maximum, which would drag in duct design assumptions nobody has. It is the same machine against itself with the two conditions that move static pressure most - blower speed and filter state - held or accounted for. The filter is now loaded, so the tech pulls it, re-measures, and finds the rise only partly recovers. The unrecovered part points past the filter into the duct system, and that is a finding the baseline made available in about ten minutes.

Note what did the work. Not the number. The blower tap and the filter description, which cost about half a minute each to write down and which turned a static pressure reading into a two-term comparison.

Case two: the baseline that was worthless in six weeks

Same shop, different tech, a heating appliance. The record reads: temperature rise, one value, dated. Nothing else.

Six weeks later there is a complaint and a re-measurement comes in noticeably lower. Is that a real change? Nobody can say. Temperature rise moves with firing rate, with airflow, with how long the unit ran before the reading, and with the state of the filter and blower. The baseline record fixes none of those. The two numbers are not comparable, which means they cannot even establish that nothing changed.

The failure was not laziness in the measurement. The reading itself was probably fine. The failure was that the record captured only the output and none of the inputs, so the entire value of the exercise evaporated at the moment the conditions drifted, which for a heating appliance is roughly the next filter change.

What the two cases share and where they part

Both took a competent reading with a working instrument. Both went into the same system. The difference is entirely in the fields alongside, and it produced two outcomes that are not close: one baseline carried a real diagnosis two years out, the other could not answer a question six weeks out.

The same failure has a shape in every trade. A pressure baseline on a pump that does not record which valves were open. An insulation resistance baseline that does not record the temperature. A vibration baseline that does not record the speed. In each case the missing field is one that the physics of the measurement says the value depends on, and in each case it costs seconds to record and cannot be recovered later at any price.

When a baseline is not worth taking

Not every reading deserves one, and pretending otherwise is how baseline discipline dies of its own weight.

• When the conditions cannot be reproduced. A reading taken during a thunderstorm on a system running in an unusual mode is a data point, not a baseline. Log it as an observation and move on.
• When the quantity has no meaningful drift. A value that either is correct or is broken, with nothing in between, gains nothing from trending.
• When a published limit already settles the question and the operating conditions are easy to match. If a single reading against a specification answers it, the baseline adds work without adding an answer.
• When you will not be back. A one-time call for a customer with no service relationship gets a documented reading, not a baseline program.

The honest way to think about the cost is in hours. On a commissioning or maintenance visit already scheduled, capturing a proper baseline adds a fraction of an hour, mostly writing rather than measuring. Recovering the same information later costs a return trip, because the conditions you needed were only available while the system was in the state you were standing in front of. The ratio is bad enough that it settles the question wherever the site is one you will see again.

Refreshing a baseline

A baseline is superseded by any work that changes the thing it describes. Replace a blower motor, change a filter type, re-pipe a run, adjust a firing rate, and every old reading downstream of that change becomes history rather than a reference. Take a new one before you leave, on the same visit, and mark the old one as pre-change rather than deleting it. The old value still tells you what the machine used to do, which is exactly the question a warranty argument asks.

The trap to avoid: comparing a post-change reading to a pre-change baseline and reporting the difference as deterioration. If you correct one side of a comparison, the other side gets corrected too, or you state plainly in the same sentence that the comparator is from before the change.

Grading a baseline you inherited

Most baselines a tech meets were taken by somebody else, often at a company that no longer services the site. Grade it before you lean on it, because using a weak baseline as if it were a strong one is worse than having none: it produces a confident difference that means nothing, and you will defend it in front of a customer.

Three tiers, and the grading takes about a minute.

• Comparable. The value, the point, the operating state and the conditions that move the number are all recorded. Retake it under the same conditions and the difference is real.
• An anchor only. The value and the point are recorded, the conditions are not. It can tell you the rough neighbourhood the machine used to live in, which occasionally rules out a wild hypothesis, and it cannot support a statement about change. Say so out loud when you use it, in the same sentence as the number.
• Unusable. A value with no point, or a value whose units or reference cannot be established. Do not compare against it at all. Take a new baseline on this visit and note that the old one could not be interpreted, so the next tech does not spend the same minute grading it again.

How to verify a baseline is usable

Hand the record to someone in the shop who was not on the job and ask them to tell you, out loud, how they would go set the system up to retake it. If they can walk through it without asking you a question, it is a baseline. If they ask even one question you have to answer from memory, that answer is the missing field, and it is missing from every other baseline that tech takes. Add it to the form rather than to the one record.

Do this once a quarter on a sampled record rather than reviewing every one. The defects come in patterns, not in isolated misses, and a single sampled record surfaces the pattern.

References

• Manufacturer commissioning and start-up documentation, which states the conditions each rated value was established under
• Trade-standard practice for commissioning records and start-up reports
• See related: Commissioning Readings to Record as a Baseline; What to Write Down So a Reading Survives
• See related: Reading a Trend Against Reading a Value