A loop can close while one station is still at the wrong height.
A survey crew closes a vertical loop to within its expected range. Later an independent check finds one station height is wrong. Both results can be real. If the same wrong reference height enters all observations touching that station with consistent signs, the loop can cancel the error even while the station coordinate absorbs it.
This guide works a small network in metres, then changes one setup to show which checks reveal a height entry error and which cannot. It builds on reciprocal trigonometric levelling but asks a different question: how does a bad instrument or target height propagate across several legs?
Keep three things visible: the physical mark, the instrument or target centre above it, and the independent observation that can check the connection between them.
Geodesy subject guide · Learning that transfers into life · How X Works hub
Choose your reading route
Route 1 · Chapters 1–3: Build the signed model
Locate all four points in one observed height difference
Trace instrument and target heights and follow one wrong entry through two sights.
Route 2 · Chapters 4–6: Use redundancy with care
Use a loop that crosses the suspected setup differently
Compare loops, detect a cancellation pattern and test a competing explanation.
Route 3 · Chapters 7–9: Check and transfer
Verify the repair from the mark back to the result
Recover the physical reference, document the repair and transfer the method.
Full chapter index · Independent worked case · Questions and practical checks
Chapter index
Build the signed model
Locate all four points in one observed height difference
Treat a height entry as evidence tied to a physical setup
Work two outgoing sights from the same mistaken setup
CHAPTER 1 OF 9 · Build the signed model
1. Locate all four points in one observed height difference
Draw mark A on the ground and an instrument above it. Draw mark B and a target above it. The instrument observes the target centre, not the ground mark B directly. A reduced vertical component describes the height of the target centre relative to the instrument centre in this simplified model. To obtain a height difference between the ground marks, you need the connection at each end.
Define instrument height as the vertical separation from the occupied mark to the instrument reference centre used in the observation. Define target height as the vertical separation from the target mark to the observed target centre. For the simplified one-way geometry, mark B minus mark A equals instrument height plus the observed vertical component minus target height. Always confirm the sign convention and the instrument’s processing before applying it to an actual survey.
As a quick example, the instrument centre is 1.50 metres above A, the target centre is 0.80 metres higher than the instrument centre, and the target is 1.60 metres above B. The ground height difference B minus A is 1.50 + 0.80 − 1.60 = 0.70 metres. A 0.03-metre increase in the entered instrument height increases the computed difference by 0.03. The same increase in the entered target height decreases it by 0.03.
This guide uses a vertical component that has already been reduced consistently for the illustration. Real work needs correct angle conventions, instrument reference offsets and any applicable corrections. The purpose here is to follow the consequences of a height entry, not to provide a complete field procedure or professional tolerance.
CHAPTER 2 OF 9 · Build the signed model
2. Treat a height entry as evidence tied to a physical setup
An instrument height applies to an occupation, not to a point name forever. If you move or re-set the instrument, the new centre may sit at a different height above the same mark. A target height likewise belongs to its physical configuration and its reference mark. An unchanged station label does not guarantee the instrument or target geometry stayed unchanged.
Now consider a setup at B used to observe C and D. If both observations were reduced with the same instrument height entry, an error in that entry can affect both legs. Repeating the directions and distances several times from that same setup does not independently check the stored height. Those repeats can be excellent while their reductions share a common offset.
Keep the raw field measurement separate from the value entered into the reduction. A physical measurement can be wrong because it was taken to the wrong reference point or along a sloping line without the appropriate conversion. The entry can also be wrong because a correct measurement was transcribed, rounded or carried over incorrectly. Comparing the displayed value with the original note helps with some mistakes, but not with a wrong physical measurement made consistently.
A reviewable record names the point, setup, physical instrument and target reference, units, measurement time and original reading. If the instrument has a marked height-measurement point that is not the axis used in the reduction, record how the conversion was handled. A number without its physical endpoints is much harder to check after a crew has moved on.
CHAPTER 3 OF 9 · Build the signed model
3. Work two outgoing sights from the same mistaken setup
Use three fictional points with true elevations B = 10.000 metres, C = 11.200 metres and D = 9.700 metres. Suppose the vertical components and target heights are otherwise correct. From B the true differences are B to C = +1.200 and B to D = −0.300 metres. The instrument-height entry at B is 0.040 metres too large.
The reduced differences become +1.240 and −0.260 metres. The second difference is still negative, but both values moved in the positive direction by the same amount. A sign check prevents a common mistake: an instrument-height error adds to both outgoing reductions even when one line points downhill. It does not make the downhill leg more negative simply because the destination is lower.
Subtract the two outgoing results: (−0.260) − (+1.240) = −1.500 metres. This agrees with D minus C. The shared error has cancelled in the difference between the two sights. That contrast checks a relation between C and D, not whether B’s entered instrument height was correct. Even a perfect independent C-to-D check would not distinguish a common error on both reductions from no error at B in this comparison alone.
Next, hold B at its independently known 10.000 metres and compute C and D using the two bad reductions. The estimates become 11.240 and 9.740 metres. Both are high by 0.040 even though their difference is right. A check can succeed for one quantity and fail to test another. Name the quantity before deciding what a pass means.
CHAPTER 4 OF 9 · Use redundancy with care
4. Use a loop that crosses the suspected setup differently
Suppose there is an independent observation from C to B, with a separately measured target height at B. Its reduction is −1.200 metres under the fictional true geometry. The B-to-C reading made with the faulty instrument-height entry was +1.240. Add the two directed readings: the loop returns +0.040 metres rather than zero. In this case the reverse observation did not share the original error, so the comparison can see it.
A nonzero closure tells you the observations disagree under the model; it does not identify the wrong entry by itself. A wrong target height on the reverse sight, an observation problem or a changed reference could also contribute. Use the pattern over several observations. The two outgoing B sights both show the same signed offset against independent evidence. That is consistent with a common B setup problem, but an independent physical check is needed to verify the cause.
Now use a loop that travels from C to B by reversing the first faulty reduction algebraically, then from B to D using the second faulty reduction. Because the reversal carries the same original evidence, the common error cancels. Closing that calculated route against C to D does not test the B instrument height independently. A second route on a diagram is not always a second measurement.
Check which observations were reused before counting independent confirmations. An exported reverse direction may be a fresh observation or simply a sign change applied to a forward difference. The same distinction applies to report rows derived from one setup. Label the origin of each edge so the review can separate an actual cross-check from another use of the same data.
CHAPTER 5 OF 9 · Use redundancy with care
5. See how a node error can disappear around every loop
Consider a second fictional network. B is used as both instrument station and target, and the same 0.040-metre reference mistake affects the entered height whenever B appears in either role. For each directed leg leaving B the reduction gains 0.040 metres; for each leg arriving at B it loses 0.040. A loop that visits B has one arrival and one departure, so the contributions cancel. The same cancellation can hold across a whole network of loops.
What will the fitted coordinates do? A model that lowers the estimated elevation of B by 0.040 metres predicts a larger difference from B to any other point and a smaller difference from any other point to B. Those are exactly the errors just described. So a network with independent control elsewhere can fit all these observations well while assigning B the wrong elevation. Internal fit has not identified the physical reference error.
This is a useful way to recognise the limit: an edge error pattern that is exactly the difference of point shifts can be absorbed into the fitted point heights. A loop adds differences and returns to where it started, so point shifts cancel around it. In this example, the fault looks to the adjustment like a change in one point coordinate. Merely adding more loops of the same kind may not resolve it.
What would help? Independently measure B against a trusted physical reference without reusing the same height convention that caused the error. A different occupation is useful only if it breaks the shared assumption. Independent measurement of the instrument or target height, a check to the physical mark by a suitable separate method, or another properly documented connection could provide evidence the loop lacked.
CHAPTER 6 OF 9 · Use redundancy with care
6. Compare the signature with other faults
A constant wrong entry can generate a common offset for a set of legs from one setup. A moved target can produce a pattern tied to the time of the move. A distance-dependent error, atmospheric effect or angle-related issue may vary with geometry. These distinctions are clues, not universal fingerprints. In a small network several causes can make nearly the same pattern, and a single residual is rarely enough to choose among them.
Imagine the offsets are +0.040, +0.042 and +0.039 metres on three independent comparisons from the same B setup. The similarity makes a shared height entry worth checking. If instead they grow with line length, or appear only after the target was changed, review those hypotheses too. Check the uncertainty of the comparisons before deciding that small differences between offsets are meaningful.
A wrong target entry can affect several sights that used the same target configuration. Keep both ends in the worksheet rather than blaming the occupied station because it was entered first in the observation record. If all suspicious observations involve both one instrument setup and one target setup, their two possible effects can be confounded. Add a comparison that changes one while holding the other suitably controlled.
Do not cure a shared mistake by giving the whole class of measurements a larger random uncertainty and leaving the physical question unanswered. That can hide a systematic pattern without repairing the mark-to-centre connection. If the cause is unconfirmed, report the competing explanations and the likely effect on the result. Reserve a correction for cases where the evidence identifies what to correct.
CHAPTER 7 OF 9 · Check and transfer
7. Verify the repair from the mark back to the result
A practical review starts with the physical endpoints. Which mark was occupied? Which instrument and target reference points were used? Which recorded height applies to which observations? Check the original notes, instrument setup and any independent height measurement. If a crew reoccupied a point, confirm whether the reduction used the new entry or retained a number from the earlier setup.
If the evidence establishes that one height entry was wrong, correct the affected observations using that verified entry and repeat the relevant reduction. Preserve the original data and document the affected scope. Check both directions and all derived values that depended on the entry. A correct repair can change several height differences at once. That is expected when the original error was shared.
Then compare against observations that did not define the correction. If you chose a height adjustment solely to make one loop close, that same loop is not independent proof of the correction. An external check or independent occupation with a separately measured mark-to-centre height provides stronger evidence. State what the new check verifies and which conditions were shared with the original work.
USGS guidance on trigonometric levelling discusses redundant instrument-height measurements, reciprocal section checks and closed-traverse closure, among other checks. The purpose of citing it here is to show why different checks are valuable together. Use the procedures and acceptance criteria appropriate to the project; the fictional numbers in this guide do not set those criteria.
CHAPTER 8 OF 9 · Check and transfer
8. Leave a readable evidence trail when the cause is uncertain
Start your report with the observed discrepancy and the quantity affected. A clear example is: “Two outgoing differences from one occupation were both about 0.04 metres high relative to independent checks. Their contrast agreed with a C-to-D measurement. This pattern is consistent with a shared occupation error, but the original physical height could not be verified.” This says why one comparison passed without pretending it validated the setup.
If a correction is confirmed, record the measurement used to establish it, the before-and-after differences, and the downstream coordinates or products that were recalculated. Preserve the unadjusted checks as well as the adjusted result. An adjusted network is useful, but it can hide the reason a specific observation was changed if the original reductions are no longer available.
Where independent evidence is missing, avoid assigning a precise cause or quietly moving a point to make the plot look right. Identify what remains supportable, such as a relative difference between independently connected points, and what is uncertain, such as the height of the suspect mark. That distinction is more useful to the next crew than a binary verdict on the whole survey.
For a peer check, give a second person the raw height entries, the reduced differences and the source of each independent observation. Ask them to trace one affected coordinate without seeing your conclusion. If they cannot distinguish a reused observation from a fresh one, the record needs a better link between each value and its source.
CHAPTER 9 OF 9 · Check and transfer
9. Transfer: read a shared reference in a different measurement
Imagine several shelves are measured relative to one floor mark. A team uses the same mistaken reference height for every reading from that mark. Differences between the shelves may agree with a separate check while all their absolute heights are wrong. The numerical instruments are different from those in the survey example, but the same shared offset can cancel when one difference is subtracted from another.
Now change the question: the team only needs clearance between two shelves. An independently measured clearance can be suitable even if the absolute floor reference is uncertain. If the task instead needs the shelf height above a building datum, the clearance cannot replace the datum check. The method transfers because it asks what each comparison measures, not because all measurement problems require the same repair.
Make a final change: the two shelves were measured from different references and their relationship is not known. The error no longer has to be shared, so subtracting the readings need not cancel it. Record each reference and check how they connect before comparing. The changed context shows exactly where the earlier shortcut was justified and where it stopped working.
Write a short answer for a colleague: which value is relative, which value depends on a shared reference, and which independent observation would check it? If you can answer without using the words “the loop passed” as a substitute for evidence, you have transferred the central idea.
Independent worked case and evidence checks
Consider a fictional triangle A, B and C with true directed differences A to B = +0.800 metres, B to C = +0.500 and C to A = −1.300. Suppose a 0.025-metre reference entry error at B subtracts 0.025 from the arriving A-to-B reduction and adds 0.025 to the departing B-to-C reduction. The reported differences are +0.775, +0.525 and −1.300. The loop sum is exactly zero in this simplified example.
Hold A fixed and calculate B from the first reported leg. It is 0.025 metres lower than it should be. C is unchanged, because the two B-related contributions cancel on the way to C. The loop check is real; it just cannot distinguish this pattern from a change in the intermediate point height. An independently reduced observation tied directly to the physical B mark is the kind of check that can reveal the shift.
Now change one fact: the wrong B entry affects only the departing leg, while the arriving target height is independently correct. The loop sum becomes +0.025 metres. The comparison now sees a discrepancy, but it does not prove by itself that B caused it. An independent occupation and verified mark-to-centre measurements help test that explanation. This change shows why the same numerical mistake can be visible in one design and hidden in another.
For a final evidence test, ask whether the supposed independent B observation was computed using the same incorrect physical reference. If so, it may reproduce the mistaken height and make the network appear confirmed. A check’s independence is about its error path, not just which person collected the data. Describe how the proposed observation avoids the particular error under investigation.
A related desk check treats each directed difference as an equation. Write the error on a leg from point i to point j as the positive height error of the instrument setup at i minus the positive height error of the target setup at j. In the special case where the same erroneous offset e belongs to both roles at one point, every leg error is e at its source minus e at its destination. This is exactly the pattern obtained by shifting the fitted point heights by minus e. A loop cannot see that pattern because each point’s contribution occurs once with each sign.
If several occupations at the same mark were genuinely independent, there may not be a single e for that point. Keeping separate setup identifiers allows the review to see which edges share a measurement. For example, a second B occupation with a new verified instrument height might have no outgoing offset, while the first still does. A comparison that crosses both occupations can now contain evidence that the point-shift model alone cannot absorb.
Try one more numerical check: two independent estimates of B differ by 0.025 metres, but one has a substantially weaker tie to the reference than the other. Do not declare the smaller-looking uncertainty correct without checking how it was obtained. If that estimate reused the suspect height entry, its nominal precision may say little about this specific error. Independence and uncertainty both matter, and neither can be inferred from the number of decimal places.
Finally, write which corrections would affect only this setup, which would affect all observations of this mark, and which would change the wider reference. Those three scopes can produce very different downstream recalculations. Check one coordinate that should change and one that should not. The untouched coordinate is a useful guard against applying a local repair to an entire survey by mistake.
Check units at the point where the correction is applied. In the triangle example the verified offset is 0.025 metres, or 25 millimetres. Adding 25 to a field expecting metres would introduce a much larger fault. Write both the entered quantity and its unit, then recompute one leg by hand before accepting the batch adjustment.
Worked comparison
| Comparison | What it checks | What it can miss |
|---|---|---|
| Two sights from one setup | Relative geometry and repeatability under the setup | A shared instrument-height error |
| Independent reciprocal observation | Disagreement between separately reduced directions | The unique cause of disagreement |
| Closed network loops | Consistency of sums of directed differences | A point-height shift absorbed by all touching edges |
| Independent tie to the physical mark | The disputed point relative to the new reference | Errors shared with that reference unless checked |
Questions and practical checks
What sign does an instrument-height error have?
In the stated one-way convention, increasing the entered instrument height increases the reduced destination-minus-source difference.
Does a higher entered target height increase the difference?
No. The target height is subtracted in the stated model.
Why did the same error disappear in two sights?
Subtracting two reductions can cancel a shared additive error. That contrast does not verify the original setup.
Can all loops close with a wrong point height?
Yes, if the edge error pattern is equivalent to shifting a point and is consistently absorbed in the fitted heights.
Does a reciprocal observation always catch it?
Only if the check does not share or cancel the same error. Trace its actual setup and target evidence.
Can I use these figures as field tolerances?
No. They illustrate signs, dependence and network checks. Follow the applicable project procedures and criteria.
Continue with the connected subjects
Build the height observation
Reciprocal trigonometric levelling: build and compare opposite height differences
Traverse angular misclosure: compare a different setup error
Understand what network checks can see
Residual redundancy and leverage: see which errors can be absorbed by a fit
Moving control and false displacement: check the physical reference
Observation weights: distinguish shared error from random scatter
Sources and scope
USGS publication: Procedures and Best Practices for Trigonometric Leveling
USGS Techniques and Methods 11-D3: observation reductions and redundant instrument-height checks
The numerical cases in this guide are original fictional teaching examples, not measurements from a real site. This is an educational explanation, not a field safety instruction, engineering approval or an official certification. For a real project, use qualified practitioners, current product documentation and site-specific evidence. Sources were checked on 5 November 2026.