What Gets Lost When You Record Only the Number

Why this lesson matters

Consider a single line from a monitoring log:

Gas Meter 01    12,437

It looks complete. It is not. At the moment of observation the technician held far more information than survived the trip to the page. This lesson defines exactly what a reading is, so you can measure how much of it your current process throws away.

The anatomy of a reading

A reading that can be relied on downstream carries six properties.

1. Identity. Not a description, a key. GM-01 must resolve to one physical instrument with a known location, a known measured quantity and a known service history. Descriptions drift; keys do not. If the meter is replaced, the new instrument gets its own record and the asset carries both, with the changeover date and both closing and opening indices.

2. Value. The number as read, at the resolution the instrument actually offers. A gauge with 0.2 bar graduations supports reading to perhaps 0.1 bar by interpolation, not 0.01 bar. A mechanical register with a red trailing digit has a known least significant digit. Recording more precision than the instrument provides manufactures noise and generates false exceptions later.

3. Unit, with its measurement type. m3 is not enough on its own. You need to know:

• whether the value is cumulative (a register index that only increases) or instantaneous (a gauge reading at a point in time);
• the index multiplier if the register is geared (x1, x10, x100);
• whether gas volume is uncorrected or temperature and pressure corrected, because these differ materially and are frequently conflated;
• the conversion basis if the value is later expressed in energy terms.

4. Time. The moment of observation, to the minute, not the date the sheet was typed. Anything load related, demand related or shift related is uninterpretable without it. A 5 am and a 2 pm reading on the same meter describe entirely different plant states.

5. Source evidence. Proof that this value came from this instrument. In practice that is a photograph of the dial or register, captured at the time of reading. It is the difference between an assertion and a verifiable observation, and it is the single highest value addition most sites can make to their process.

6. Context. Who read it, from where, and under what plant condition. Plus anything the technician judged worth noting: "read after burner cycled off", "glass fogged, best estimate", "isolation valve part closed".

What the photograph preserves that the number cannot

A source image is not a nicety. It carries information that is genuinely unrecoverable otherwise:

• Full register resolution. The logbook records 12,437 but the register shows 12,437.86. The trailing digits matter when you are differencing two readings a week apart on a low flow asset.
• Needle position within the graduation. Whether a gauge sits just under the mark or just over is the difference between a stable and a drifting asset.
• Unit and scale markings on the instrument face, which settle the "kPa or bar" argument permanently.
• Tag plate and serial number, which confirm identity independently of what the reader typed.
• Physical condition. Corrosion, fogging, condensation inside the glass, a bent pointer, a broken seal, a leak stain under the fitting, frost on a line. These are the observations a good technician makes anyway and that no numeric field captures.
• Plausibility. When a value is later challenged, the image resolves it in seconds instead of requiring a return visit.

Worked example: the cost of a missing property

Take a rotary gas meter on a boiler house header, read weekly.

Missing property	Direct consequence
Identity	Readings split between two asset names; annual consumption understated by the portion filed under the alias.
Resolution	Weekly delta rounded to whole m3; on a low use asset the rounding error exceeds the signal.
Cumulative vs instantaneous	Someone averages the register index and reports a meaningless figure as consumption.
Corrected vs uncorrected volume	Energy conversion applied to the wrong basis; reported gas energy out by a few percent, consistently, in one direction.
Time	A Monday 6 am reading compared with a Friday 4 pm reading; the seven day period is actually 6.4 days, and the calculated daily rate is wrong by nine percent.
Source image	A transposition from 12,437 to 12,347 is undetectable, and the following week shows an impossible 90 m3 spike.

Every one of these is routine. None of them require anyone to be careless.

Resolution, uncertainty and the discipline of not over reporting

Two ideas worth internalising:

Instrument uncertainty. An EN 837-1 class 1.6 pressure gauge with a 0 to 16 bar span has a permitted error of +/- 0.256 bar anywhere on the scale. If your expected operating range is set to a tolerance tighter than that, you are alarming on instrument error rather than on plant behaviour.

Reading uncertainty. Parallax, interpolation between graduations and glass condition add further error on top. The practical rule is to read to the nearest half graduation and to state the range you are working to.

Recording 2.1 bar from that gauge is honest. Recording 2.134 bar is not, and it will eventually cause somebody to chase a fault that exists only in the decimal places.

Turning this into a capture standard

A workable site standard for manual reading:

1. Every instrument has a permanent tag, physically fixed and legible.
2. The asset record states the measured quantity, the unit, the multiplier, whether cumulative or instantaneous, the full scale and the readable resolution.
3. Every reading is photographed at the instrument, framed so the tag and the register are both visible where possible.
4. The value is entered at the instrument, not transcribed later.
5. The capture time is system generated, never typed.
6. Free text notes are encouraged and are treated as part of the record, not as commentary.
7. Unreadable instruments are recorded as unreadable with the reason, never as the previous value.

That last rule is worth enforcing hard. A visible gap is information. A carried forward value is fabricated data that looks identical to real data forever.

Common mistakes at this stage

• Photographing the sheet rather than the instrument. The evidence has to sit between the instrument and the number, not after it.
• Storing images separately from readings. An image in a phone gallery or a shared folder is not evidence; it is an orphan file. Evidence means the image is bound to the reading record.
• Typing the time. Any human entered timestamp will eventually be rounded, guessed or copied.
• Over specifying precision to make the data look better than the instrument allows.
• Discarding the note field in reporting. The note is often where the earliest failure signal on the site is sitting.