Why is mathematics important in coordinate-measuring machines? Because a probe does not directly announce, “this bore is round” or “this surface is flat.” It records coordinates. Software then establishes a reference frame, compensates for the probe, fits ideal geometry to imperfect points, calculates residuals and compares a defined result with a tolerance. Mathematics is the bridge from touched points to defensible dimensional evidence.
This article is educational, not an inspection procedure. Real acceptance decisions must follow the drawing, applicable standards, validated software, traceable calibration and an authorised measurement plan. A classroom least-squares fit cannot certify a manufactured part.
Quick Reading Route
- Start with what a CMM measures.
- Use coordinate systems and datums to orient the part.
- Read fitting geometry for the central mathematics.
- Work through a plane example.
- Study tolerance and uncertainty before making decisions.
- Finish with misconceptions and FAQs.
What a CMM Actually Measures
A coordinate-measuring machine moves or observes a sensing system within a three-dimensional coordinate frame. A tactile probe may trigger when its stylus contacts a surface; optical sensors may infer points without contact. The raw evidence is a set of coordinates plus machine, sensor and environmental information.
The NIST discussion of statistical issues in CMM inspection describes a CMM as a computer-controlled device that usually obtains surface points one at a time and highlights model fitting, sampling design and measurement error. That description prevents a common misunderstanding: the measured points and the fitted feature are different objects.
Did You Know? A perfect circle is usually calculated, not touched
If a probe records twelve points around a bore, none is “the centre.” Software fits a circle under a chosen criterion. Different valid criteria can return different centres and diameters from the same points because they optimise different mathematical objectives.
The measurement chain
- locate the part and establish a coordinate system;
- measure points under a defined probing strategy;
- compensate the stylus geometry where appropriate;
- associate points with a feature model;
- fit or evaluate that feature using a stated criterion;
- calculate size, location, orientation or form; and
- include uncertainty before a conformity decision.
Every stage can change the final number. More points do not repair a wrong datum, thermal error or unsuitable fitting rule.
Coordinate Systems, Datums and Transformations
A point is written p=(x,y,z), but its numbers depend on origin and axis direction. The machine has a coordinate system; the workpiece drawing defines another through datums. Alignment finds a transformation between them.
A rigid transformation can be written p_part = R p_machine + t, where R is a rotation matrix and t a translation vector. Translation changes coordinates but not distances. A proper rotation preserves lengths and angles: RᵀR=I and det(R)=+1.
A simple 2D alignment
Suppose a measured point is (102.0, 53.0) mm. The part origin is at machine coordinate (100.0, 50.0) mm, and part x is rotated 30° counter-clockwise from machine x. First subtract the origin to obtain (2,3). Then rotate by −30° into the part frame:
x′ = 2cos30° + 3sin30° ≈ 3.232 mm;
y′ = −2sin30° + 3cos30° ≈ 1.598 mm.
The sign follows the stated frame transformation. Drawing axes before calculating prevents many mistakes.
Datums are functional references
A datum feature is not merely the first surface an operator notices. Drawing rules establish how reference features constrain degrees of freedom. A primary plane can constrain translation normal to it and rotations; secondary and tertiary references complete the frame. Actual standards contain detailed rules, so an article should teach the geometry without inventing drawing interpretation.
Best fit is not automatically datum alignment
Globally shifting a part cloud to minimise all deviations may make a colour map look balanced, yet it can violate the functional datum scheme. Inspection should not optimise away the very location error the drawing intends to control.
Probe Geometry and Contact
A spherical stylus has a non-zero radius. The trigger centre is offset from the contacted surface. If the local outward unit normal is n and stylus radius is r, a simplified surface estimate is p_surface = p_centre − r n, depending on convention and approach direction.
The normal may itself come from an assumed CAD surface or neighbouring points. On a steep curvature or edge, a small normal error changes the compensated location. NIST’s work on small-hole measurement with micro contact probes discusses how probe-tip geometry and mechanical filtering affect small-feature measurement.
Stylus qualification
Measuring a calibrated reference sphere estimates effective stylus size and offsets. The process is not decorative setup: those parameters enter later compensation. Changing stylus configuration, orientation or qualification conditions can change the relevant error model.
Contact force and deformation
Probe contact may elastically deform the stylus, part or fixture. A soft polymer and a thick steel block do not respond identically. The CMM can be repeatable while the contact process introduces a task-specific bias.
Accessibility creates sampling choices
A long stylus may reach a deep feature but increase bending sensitivity. Collision avoidance may leave one region sparsely sampled. Sampling is therefore geometric and practical, not simply “choose many random points.”
From Points to Features: Fitting Geometry
For a line y=ax+b, ordinary least squares minimises the sum of squared vertical residuals. In coordinate metrology, an orthogonal fit often minimises perpendicular distances instead. The residual definition must match the geometry.
For a plane n·p+d=0 with unit normal n, signed perpendicular distance is ri=n·pi+d. A least-squares plane minimises Σri² subject to |n|=1. Its centroid lies on the fitted plane, and the normal corresponds to the direction of smallest point-cloud variance.
Least-squares circle
For centre (a,b) and radius r, geometric residual for point i is √[(xi−a)²+(yi−b)²]−r. Minimising squared geometric residuals differs from fitting an algebraic equation x²+y²+Dx+Ey+F=0 without correction. Fast algebraic fits can initialise an iterative geometric fit, but their objective should not be confused with physical distance.
Other fitting criteria
NIST’s evaluation of CMM software geometry uncertainties discusses reference algorithms for least-squares, Chebyshev, maximum-inscribed and minimum-circumscribed fits. These are not interchangeable labels:
- least squares balances squared residuals;
- minimum-zone or Chebyshev-style form evaluation focuses on limiting separation;
- maximum-inscribed features can be relevant to internal mating conditions; and
- minimum-circumscribed features can be relevant to external envelopes.
The drawing and standard determine which association is appropriate. Software should not silently use whichever returns the most favourable result.
Worked Example: Fitting a Plane and Reading Residuals
Consider five fictional points in millimetres:
| Point | x | y | z |
|---|---|---|---|
| A | 0 | 0 | 10.01 |
| B | 50 | 0 | 10.06 |
| C | 0 | 40 | 9.97 |
| D | 50 | 40 | 10.02 |
| E | 25 | 20 | 10.00 |
A plane z=ax+by+c fits this data exactly closely. The x increase from A to B suggests a≈(10.06−10.01)/50=0.001. The y increase from A to C suggests b≈(9.97−10.01)/40=−0.001. Using A suggests c≈10.01. This approximate plane predicts D at 10.02 and E at 10.015 mm.
Residuals
With that approximate model, residuals z_observed−z_predicted are A 0, B 0, C 0, D 0 and E −0.015 mm. A formal least-squares refit would distribute some of E’s deviation across all parameters rather than leave four zeros. The example shows why an influential central point can tilt or shift the final plane.
Form versus orientation
Residual spread describes departure from the fitted plane. The coefficients a and b describe orientation relative to the coordinate frame. A surface may be very flat but tilted; another may have the correct average orientation but poor flatness. One number cannot represent both.
Sampling blind spots
These points cover corners and centre, yet a bump between them could remain unseen. The fitted plane describes sampled evidence, not every location on the physical surface. Feature size, manufacturing process and risk should guide sampling density and distribution.
Robustness check
If E were recorded as 10.20 rather than 10.00, the fit would change substantially. Before deleting it as an outlier, inspect probe contact, point location and surface condition. An inconvenient point may reveal a real defect.
Size, Form, Orientation and Location
Dimensional inspection questions belong to different categories. Size concerns quantities such as diameter. Form concerns deviations such as flatness, straightness, circularity or cylindricity. Orientation concerns parallelism or perpendicularity to datums. Location concerns position relative to the datum reference frame.
Flatness concept
Flatness is associated with separation between two parallel planes that contain the evaluated surface under the applicable criterion. A least-squares residual range may be informative but is not automatically the standard-defined minimum zone.
Circularity concept
Circularity concerns radial separation of two concentric circles enclosing a cross-section under its definition. A reported least-squares radius does not by itself report circularity.
Position as a zone
A positional tolerance commonly defines a cylindrical or circular zone around true position, with material-condition modifiers possible. Treating it as independent ±x and ±y limits changes the geometry. If a fitted centre error is (0.06,0.08) mm, radial error is √(0.06²+0.08²)=0.10 mm; a diametrical zone value may be twice the radial distance depending on reporting convention.
Pattern relationships
For a hole pattern, relative spacing, rotation and datum location interact. Fitting each hole independently is not the same as fitting a pattern under a collective rule. A measurement plan must preserve the drawing’s relational intent.
Tolerance, Decision and Measurement Uncertainty
A measured value is an estimate with uncertainty. Conformity assessment compares the estimate and its uncertainty with a specification under a decision rule. A value just inside a limit is not infinitely certain.
NIST’s traceability and uncertainty discussion for CMMs emphasises that task-specific CMM measurements may differ greatly from calibration conditions. NIST’s step-gauge work highlights thermal effects, contact deformation, fixturing and coordinate-system generation in an uncertainty budget.
Sources to consider
- machine scale, geometry and probing performance;
- stylus qualification and bending;
- temperature of machine, part and scale;
- part fixturing, cleanliness and deformation;
- point sampling and feature coverage;
- datum construction and fit algorithm;
- software numerical implementation; and
- repeatability across runs and operators.
Some sources are correlated. Adding every maximum directly is overly conservative; treating them all as independent can be over-optimistic. A justified uncertainty model states distributions, sensitivities and correlations.
Guard bands and decision rules
A guard band moves an acceptance boundary inward to manage false-accept risk. It also increases the chance of rejecting conforming parts. The trade-off belongs to an agreed rule, not an inspector’s after-the-fact preference.
Misconceptions Worth Correcting
“More decimal places mean more accuracy”
Display resolution does not equal measurement uncertainty. A result shown to 0.0001 mm may not support that precision.
“More points always improve the answer”
More well-distributed, valid points can help. Many clustered points can overweight one region, while systematic error persists regardless of count.
“The least-squares feature is the true feature”
It is one mathematically associated feature. Other criteria answer other functional or tolerance questions.
“Calibration removes all error”
Calibration supplies evidence under stated conditions. Task geometry, environment, fixturing and analysis still contribute uncertainty.
“A pass result proves the part is perfect”
Conformity means the reported measurand met a specified rule with stated evidence. It does not mean zero deviation or zero uncertainty.
Learning Pathways for Students
Coordinate metrology unites coordinate geometry, vectors, matrices, trigonometry, optimisation, statistics and significant figures. A Secondary student can plot points, calculate distances and examine residuals. An Additional Mathematics student can explore minimisation and derivatives. A computing student can implement circle or plane fitting and visualise a residual map.
The career lesson should remain open: these ideas appear in precision engineering, aerospace, medical devices, optics, robotics and quality assurance. Mathematics helps a student participate in those fields; it does not guarantee admission, certification or employment.
Four-week plan
- Week 1: coordinates, rigid transformations and datums;
- Week 2: line, plane and circle fitting with residuals;
- Week 3: tolerances, sampling and algorithm comparisons;
- Week 4: uncertainty budgets, decision rules and an auditable report.
Parents can ask three powerful questions: “What was directly measured?”, “What was fitted?”, and “Which rule turns that estimate into a decision?” Those questions build industrial data literacy without requiring specialist equipment.
Frequently Asked Questions
Does a CMM measure dimensions directly?
It measures coordinates through a sensing system. Dimensions and geometric characteristics are calculated from those observations and definitions.
Why is least squares popular?
It is mathematically tractable, uses all points and provides residual statistics. It is not automatically the correct association for every tolerance.
What is a residual?
It is the signed or absolute difference between an observed point and the fitted model under a defined distance rule.
Why does point location matter?
Points determine which regions influence the fit. Poor coverage can miss local deviations or make parameters unstable.
Can two software packages disagree?
Yes, if they use different filters, association criteria, datum constructions, convergence settings or standards interpretations. Validated reference datasets help test algorithms.
What does traceability mean here?
It links a measurement result through a documented, unbroken calibration chain with stated uncertainties to appropriate references. It does not mean merely saving a file.
Why measure temperature?
Materials and scales expand with temperature. A small coefficient multiplied by a long length and temperature difference can become significant.
Is repeatability enough?
No. Repeated values can share a systematic bias. Repeatability is one component of measurement performance.
What is a datum shift?
It is a standards-defined possibility under certain material-boundary conditions, not permission to translate results freely. Drawing rules govern its use.
Can a scan replace tactile probing?
Scanning provides dense data but adds sensor, filtering and data-reduction questions. Suitability depends on the task and validated method.
How should outliers be handled?
Investigate their cause using pre-defined rules. Preserve raw data and document exclusions. Never remove a point solely because it causes failure.
What is the most transferable lesson?
An ideal geometric object is often an estimate constructed from imperfect samples. Naming the construction makes the conclusion trustworthy.
Useful Next Reading
Read Why Mathematics? | Gauge Blocks, Calibration Chains and Measurement Uncertainty for traceability foundations. Why Mathematics? | Flat-Pack Furniture, Cut Lists and Tolerance Stacks shows how dimensional variation accumulates. Why Mathematics? | Comparing Percentages Fairly strengthens denominator and boundary reasoning.
The optimistic lesson is that precision is not magic. It is a visible chain of coordinates, definitions, algorithms, checks and honest uncertainty. Students who learn that chain are learning how modern manufacturing turns geometry into evidence.
A Practical Investigation Studio
These activities use synthetic or openly released teaching data. They are designed to make every assumption visible. For each investigation, begin with a labelled diagram or data dictionary, keep unrounded intermediate values, and finish with a short limitation statement. The goal is not to imitate a professional instrument with a spreadsheet; it is to understand which mathematical operation gives the instrument its meaning.
Investigation 1: Coordinate audit
Create ten 3D points for a nominal plane, add small z-errors and state millimetres on every coordinate. Check axis direction, units, duplicated points and whether the chosen points cover the feature. Include one graph or table, one dimensional or limiting-case check, and one paragraph separating the calculated result from the real-world conclusion. Change one input across a sensible range to test sensitivity. If the result changes sharply, report the instability instead of hiding it with extra decimal places. Preserve the original data and record every correction so a classmate can audit the route from measurement to claim. Before calculating, predict the direction of change and explain the physical reason. After calculating, compare the prediction with the result and investigate any disagreement. Repeat the analysis with one deliberately flawed input—wrong unit, reversed sign, missing sample or premature rounding—and describe how the error appears. This makes the activity an error-detection exercise rather than a search for one polished answer. Finish by naming which quantities were observed, which were assumed, which were fitted and which were derived. State the domain over which the model is credible and one condition that would require a richer model. A strong submission includes the raw synthetic data, formulas or code, labelled axes, units, intermediate checks, final result, uncertainty note and a concise explanation suitable for a younger student. Do not claim professional validation from a classroom simulation.
Investigation 2: Centroid and translation
Calculate the centroid, subtract it from every point and verify that the centred coordinates average to zero. Explain why translating the origin should not change fitted shape or residual magnitudes. Include one graph or table, one dimensional or limiting-case check, and one paragraph separating the calculated result from the real-world conclusion. Change one input across a sensible range to test sensitivity. If the result changes sharply, report the instability instead of hiding it with extra decimal places. Preserve the original data and record every correction so a classmate can audit the route from measurement to claim. Before calculating, predict the direction of change and explain the physical reason. After calculating, compare the prediction with the result and investigate any disagreement. Repeat the analysis with one deliberately flawed input—wrong unit, reversed sign, missing sample or premature rounding—and describe how the error appears. This makes the activity an error-detection exercise rather than a search for one polished answer. Finish by naming which quantities were observed, which were assumed, which were fitted and which were derived. State the domain over which the model is credible and one condition that would require a richer model. A strong submission includes the raw synthetic data, formulas or code, labelled axes, units, intermediate checks, final result, uncertainty note and a concise explanation suitable for a younger student. Do not claim professional validation from a classroom simulation.
Investigation 3: Least-squares plane
Fit z=ax+by+c to fictional data and calculate signed residuals and their squared sum. Compare residual pattern with a colour map rather than quoting only one RMS number. Include one graph or table, one dimensional or limiting-case check, and one paragraph separating the calculated result from the real-world conclusion. Change one input across a sensible range to test sensitivity. If the result changes sharply, report the instability instead of hiding it with extra decimal places. Preserve the original data and record every correction so a classmate can audit the route from measurement to claim. Before calculating, predict the direction of change and explain the physical reason. After calculating, compare the prediction with the result and investigate any disagreement. Repeat the analysis with one deliberately flawed input—wrong unit, reversed sign, missing sample or premature rounding—and describe how the error appears. This makes the activity an error-detection exercise rather than a search for one polished answer. Finish by naming which quantities were observed, which were assumed, which were fitted and which were derived. State the domain over which the model is credible and one condition that would require a richer model. A strong submission includes the raw synthetic data, formulas or code, labelled axes, units, intermediate checks, final result, uncertainty note and a concise explanation suitable for a younger student. Do not claim professional validation from a classroom simulation.
Investigation 4: Outlier sensitivity
Move one point by 0.20 mm and refit the feature. Report how the parameters change; do not delete the point without a documented measurement reason. Include one graph or table, one dimensional or limiting-case check, and one paragraph separating the calculated result from the real-world conclusion. Change one input across a sensible range to test sensitivity. If the result changes sharply, report the instability instead of hiding it with extra decimal places. Preserve the original data and record every correction so a classmate can audit the route from measurement to claim. Before calculating, predict the direction of change and explain the physical reason. After calculating, compare the prediction with the result and investigate any disagreement. Repeat the analysis with one deliberately flawed input—wrong unit, reversed sign, missing sample or premature rounding—and describe how the error appears. This makes the activity an error-detection exercise rather than a search for one polished answer. Finish by naming which quantities were observed, which were assumed, which were fitted and which were derived. State the domain over which the model is credible and one condition that would require a richer model. A strong submission includes the raw synthetic data, formulas or code, labelled axes, units, intermediate checks, final result, uncertainty note and a concise explanation suitable for a younger student. Do not claim professional validation from a classroom simulation.
Investigation 5: Circle from points
Fit a circle to points around an arc, then repeat with points around nearly a full circumference. Relate angular coverage to stability of centre and radius estimates. Include one graph or table, one dimensional or limiting-case check, and one paragraph separating the calculated result from the real-world conclusion. Change one input across a sensible range to test sensitivity. If the result changes sharply, report the instability instead of hiding it with extra decimal places. Preserve the original data and record every correction so a classmate can audit the route from measurement to claim. Before calculating, predict the direction of change and explain the physical reason. After calculating, compare the prediction with the result and investigate any disagreement. Repeat the analysis with one deliberately flawed input—wrong unit, reversed sign, missing sample or premature rounding—and describe how the error appears. This makes the activity an error-detection exercise rather than a search for one polished answer. Finish by naming which quantities were observed, which were assumed, which were fitted and which were derived. State the domain over which the model is credible and one condition that would require a richer model. A strong submission includes the raw synthetic data, formulas or code, labelled axes, units, intermediate checks, final result, uncertainty note and a concise explanation suitable for a younger student. Do not claim professional validation from a classroom simulation.
Investigation 6: Probe compensation
Offset a measured point along a known surface normal by a fictional stylus radius. Reverse the normal and show why the wrong sign moves the estimate to the wrong side. Include one graph or table, one dimensional or limiting-case check, and one paragraph separating the calculated result from the real-world conclusion. Change one input across a sensible range to test sensitivity. If the result changes sharply, report the instability instead of hiding it with extra decimal places. Preserve the original data and record every correction so a classmate can audit the route from measurement to claim. Before calculating, predict the direction of change and explain the physical reason. After calculating, compare the prediction with the result and investigate any disagreement. Repeat the analysis with one deliberately flawed input—wrong unit, reversed sign, missing sample or premature rounding—and describe how the error appears. This makes the activity an error-detection exercise rather than a search for one polished answer. Finish by naming which quantities were observed, which were assumed, which were fitted and which were derived. State the domain over which the model is credible and one condition that would require a richer model. A strong submission includes the raw synthetic data, formulas or code, labelled axes, units, intermediate checks, final result, uncertainty note and a concise explanation suitable for a younger student. Do not claim professional validation from a classroom simulation.
Investigation 7: Datums and transforms
Apply a rotation and translation to a point cloud using a matrix. Verify pairwise distances remain unchanged under the rigid transform. Include one graph or table, one dimensional or limiting-case check, and one paragraph separating the calculated result from the real-world conclusion. Change one input across a sensible range to test sensitivity. If the result changes sharply, report the instability instead of hiding it with extra decimal places. Preserve the original data and record every correction so a classmate can audit the route from measurement to claim. Before calculating, predict the direction of change and explain the physical reason. After calculating, compare the prediction with the result and investigate any disagreement. Repeat the analysis with one deliberately flawed input—wrong unit, reversed sign, missing sample or premature rounding—and describe how the error appears. This makes the activity an error-detection exercise rather than a search for one polished answer. Finish by naming which quantities were observed, which were assumed, which were fitted and which were derived. State the domain over which the model is credible and one condition that would require a richer model. A strong submission includes the raw synthetic data, formulas or code, labelled axes, units, intermediate checks, final result, uncertainty note and a concise explanation suitable for a younger student. Do not claim professional validation from a classroom simulation.
Investigation 8: Tolerance zone
Draw a positional tolerance zone and place repeated fitted centres within it. Keep specification decision separate from uncertainty and sampling discussion. Include one graph or table, one dimensional or limiting-case check, and one paragraph separating the calculated result from the real-world conclusion. Change one input across a sensible range to test sensitivity. If the result changes sharply, report the instability instead of hiding it with extra decimal places. Preserve the original data and record every correction so a classmate can audit the route from measurement to claim. Before calculating, predict the direction of change and explain the physical reason. After calculating, compare the prediction with the result and investigate any disagreement. Repeat the analysis with one deliberately flawed input—wrong unit, reversed sign, missing sample or premature rounding—and describe how the error appears. This makes the activity an error-detection exercise rather than a search for one polished answer. Finish by naming which quantities were observed, which were assumed, which were fitted and which were derived. State the domain over which the model is credible and one condition that would require a richer model. A strong submission includes the raw synthetic data, formulas or code, labelled axes, units, intermediate checks, final result, uncertainty note and a concise explanation suitable for a younger student. Do not claim professional validation from a classroom simulation.
Investigation 9: Thermal sensitivity
Apply a linear expansion estimate to a 500 mm steel length over a 3 °C change. State the coefficient used and distinguish model correction from measured temperature uncertainty. Include one graph or table, one dimensional or limiting-case check, and one paragraph separating the calculated result from the real-world conclusion. Change one input across a sensible range to test sensitivity. If the result changes sharply, report the instability instead of hiding it with extra decimal places. Preserve the original data and record every correction so a classmate can audit the route from measurement to claim. Before calculating, predict the direction of change and explain the physical reason. After calculating, compare the prediction with the result and investigate any disagreement. Repeat the analysis with one deliberately flawed input—wrong unit, reversed sign, missing sample or premature rounding—and describe how the error appears. This makes the activity an error-detection exercise rather than a search for one polished answer. Finish by naming which quantities were observed, which were assumed, which were fitted and which were derived. State the domain over which the model is credible and one condition that would require a richer model. A strong submission includes the raw synthetic data, formulas or code, labelled axes, units, intermediate checks, final result, uncertainty note and a concise explanation suitable for a younger student. Do not claim professional validation from a classroom simulation.
Investigation 10: Uncertainty budget
List probe, machine, fixturing, temperature, sampling and fitting contributions. Classify which may be correlated instead of automatically adding every maximum. Include one graph or table, one dimensional or limiting-case check, and one paragraph separating the calculated result from the real-world conclusion. Change one input across a sensible range to test sensitivity. If the result changes sharply, report the instability instead of hiding it with extra decimal places. Preserve the original data and record every correction so a classmate can audit the route from measurement to claim. Before calculating, predict the direction of change and explain the physical reason. After calculating, compare the prediction with the result and investigate any disagreement. Repeat the analysis with one deliberately flawed input—wrong unit, reversed sign, missing sample or premature rounding—and describe how the error appears. This makes the activity an error-detection exercise rather than a search for one polished answer. Finish by naming which quantities were observed, which were assumed, which were fitted and which were derived. State the domain over which the model is credible and one condition that would require a richer model. A strong submission includes the raw synthetic data, formulas or code, labelled axes, units, intermediate checks, final result, uncertainty note and a concise explanation suitable for a younger student. Do not claim professional validation from a classroom simulation.
Investigation 11: Algorithm comparison
Compare least-squares, minimum-zone and inscribed/circumscribed fits on one dataset. Explain why different mathematical objectives answer different inspection questions. Include one graph or table, one dimensional or limiting-case check, and one paragraph separating the calculated result from the real-world conclusion. Change one input across a sensible range to test sensitivity. If the result changes sharply, report the instability instead of hiding it with extra decimal places. Preserve the original data and record every correction so a classmate can audit the route from measurement to claim. Before calculating, predict the direction of change and explain the physical reason. After calculating, compare the prediction with the result and investigate any disagreement. Repeat the analysis with one deliberately flawed input—wrong unit, reversed sign, missing sample or premature rounding—and describe how the error appears. This makes the activity an error-detection exercise rather than a search for one polished answer. Finish by naming which quantities were observed, which were assumed, which were fitted and which were derived. State the domain over which the model is credible and one condition that would require a richer model. A strong submission includes the raw synthetic data, formulas or code, labelled axes, units, intermediate checks, final result, uncertainty note and a concise explanation suitable for a younger student. Do not claim professional validation from a classroom simulation.
Investigation 12: Reproducible report
Record point IDs, coordinate system, fit criterion, exclusions, residuals and rounding. Ask a peer to reproduce the result from the saved inputs and formulas. Include one graph or table, one dimensional or limiting-case check, and one paragraph separating the calculated result from the real-world conclusion. Change one input across a sensible range to test sensitivity. If the result changes sharply, report the instability instead of hiding it with extra decimal places. Preserve the original data and record every correction so a classmate can audit the route from measurement to claim. Before calculating, predict the direction of change and explain the physical reason. After calculating, compare the prediction with the result and investigate any disagreement. Repeat the analysis with one deliberately flawed input—wrong unit, reversed sign, missing sample or premature rounding—and describe how the error appears. This makes the activity an error-detection exercise rather than a search for one polished answer. Finish by naming which quantities were observed, which were assumed, which were fitted and which were derived. State the domain over which the model is credible and one condition that would require a richer model. A strong submission includes the raw synthetic data, formulas or code, labelled axes, units, intermediate checks, final result, uncertainty note and a concise explanation suitable for a younger student. Do not claim professional validation from a classroom simulation.