Chapter map

1. 1. Spatial samples
2. 2. Coordinate system
3. 3. Delaunay mesh
4. 4. Local plane
5. 5. Gradient vector
6. 6. Gradient magnitude
7. 7. Gradient direction
8. 8. Edge-normal flux
9. 9. Candidate threshold
10. 10. Connectivity rule
11. 11. Boundary score
12. 12. Permutation null
13. 13. Simulation null
14. 14. Mesh sensitivity
15. 15. Denoising sensitivity
16. 16. External validation

Visit the How Mathematics Works hub for the wider series and its connections to education, reasoning, science, computing, design and careers.

---

1. Spatial samples

The useful question is not only how to calculate, but what the calculation assumes. Record coordinates and measured values.

A concrete situation

Three vertices carry values 2,5,3.

What the mathematics does not license

Uneven density changes resolution. The same reasoning helps students move from an exercise to a real dataset.

Try it, check it, transfer it

Map sampling effort. Then explain the result to someone who has not seen the formula. If the explanation needs a hidden assumption, add that assumption to the model record.

---

2. Coordinate system

A small example makes the mechanism visible. Use meaningful distance units.

A concrete situation

Projected metres replace latitude degrees.

What the mathematics does not license

Distortion changes slopes. That distinction matters because a correct formula can still answer the wrong operational question.

Try it, check it, transfer it

Choose an appropriate projection. Then explain the result to someone who has not seen the formula. If the explanation needs a hidden assumption, add that assumption to the model record.

---

3. Delaunay mesh

This is where modelling choices begin to affect evidence. Connect points without interior circumcircle points.

A concrete situation

The three samples form one triangle.

What the mathematics does not license

Long skinny triangles are unstable. The habit of checking assumptions is one of the practical benefits of learning mathematics.

Try it, check it, transfer it

Audit mesh quality. Then explain the result to someone who has not seen the formula. If the explanation needs a hidden assumption, add that assumption to the model record.

---

4. Local plane

The central mathematical idea is concrete. Fit f=Gx x+Gy y+C through three values.

A concrete situation

The plane is f=3x+y+2.

What the mathematics does not license

Three points fit exactly, including noise. The same reasoning helps students move from an exercise to a real dataset.

Try it, check it, transfer it

Use smoothing diagnostics. Then explain the result to someone who has not seen the formula. If the explanation needs a hidden assumption, add that assumption to the model record.

---

5. Gradient vector

Here the geometry becomes operational. Read G=(Gx,Gy).

A concrete situation

G=(3,1).

What the mathematics does not license

Direction depends on coordinate orientation. That distinction matters because a correct formula can still answer the wrong operational question.

Try it, check it, transfer it

State axes. Then explain the result to someone who has not seen the formula. If the explanation needs a hidden assumption, add that assumption to the model record.

---

6. Gradient magnitude

This step deserves a careful pause. Compute ||G||.

A concrete situation

√10≈3.162 units per distance.

What the mathematics does not license

Large magnitude can be one outlier. The habit of checking assumptions is one of the practical benefits of learning mathematics.

Try it, check it, transfer it

Inspect vertex influence. Then explain the result to someone who has not seen the formula. If the explanation needs a hidden assumption, add that assumption to the model record.

---

7. Gradient direction

The useful question is not only how to calculate, but what the calculation assumes. Convert vector to an angle.

A concrete situation

The steepest rise points mostly east.

What the mathematics does not license

Circular summaries need care. The same reasoning helps students move from an exercise to a real dataset.

Try it, check it, transfer it

Plot arrows. Then explain the result to someone who has not seen the formula. If the explanation needs a hidden assumption, add that assumption to the model record.

---

8. Edge-normal flux

A small example makes the mechanism visible. Project gradients across a proposed boundary.

A concrete situation

A line normal aligned with G has high flux.

What the mathematics does not license

Tangential gradients do not cross the line. That distinction matters because a correct formula can still answer the wrong operational question.

Try it, check it, transfer it

Separate magnitude and orientation. Then explain the result to someone who has not seen the formula. If the explanation needs a hidden assumption, add that assumption to the model record.

---

9. Candidate threshold

This is where modelling choices begin to affect evidence. Declare a percentile or model-based cutoff.

A concrete situation

Top 5% of magnitudes are candidates.

What the mathematics does not license

A percentile forces candidates under pure noise. The habit of checking assumptions is one of the practical benefits of learning mathematics.

Try it, check it, transfer it

Calibrate globally. Then explain the result to someone who has not seen the formula. If the explanation needs a hidden assumption, add that assumption to the model record.

---

10. Connectivity rule

The central mathematical idea is concrete. Join neighbouring candidate elements.

A concrete situation

Adjacent triangles form a chain.

What the mathematics does not license

A small rule change can bridge gaps. The same reasoning helps students move from an exercise to a real dataset.

Try it, check it, transfer it

Publish tolerance. Then explain the result to someone who has not seen the formula. If the explanation needs a hidden assumption, add that assumption to the model record.

---

11. Boundary score

Here the geometry becomes operational. Sum or average flux along the chain.

A concrete situation

A long aligned chain scores highly.

What the mathematics does not license

Longer paths have more chances. That distinction matters because a correct formula can still answer the wrong operational question.

Try it, check it, transfer it

Normalize and simulate. Then explain the result to someone who has not seen the formula. If the explanation needs a hidden assumption, add that assumption to the model record.

---

12. Permutation null

This step deserves a careful pause. Relabel values over fixed coordinates when exchangeable.

A concrete situation

Every replicate rebuilds gradients and boundaries.

What the mathematics does not license

Spatial trend can violate exchangeability. The habit of checking assumptions is one of the practical benefits of learning mathematics.

Try it, check it, transfer it

Model trend first. Then explain the result to someone who has not seen the formula. If the explanation needs a hidden assumption, add that assumption to the model record.

---

13. Simulation null

The useful question is not only how to calculate, but what the calculation assumes. Generate surfaces under a fitted smooth model.

A concrete situation

Replicates preserve noise and sampling density.

What the mathematics does not license

Misspecified covariance alters tails. The same reasoning helps students move from an exercise to a real dataset.

Try it, check it, transfer it

Check diagnostics. Then explain the result to someone who has not seen the formula. If the explanation needs a hidden assumption, add that assumption to the model record.

---

14. Mesh sensitivity

A small example makes the mechanism visible. Change triangulation or raster resolution within a plan.

A concrete situation

A skinny triangle disappears after resampling.

What the mathematics does not license

Choosing the sharpest mesh is biased. That distinction matters because a correct formula can still answer the wrong operational question.

Try it, check it, transfer it

Report stability. Then explain the result to someone who has not seen the formula. If the explanation needs a hidden assumption, add that assumption to the model record.

---

15. Denoising sensitivity

This is where modelling choices begin to affect evidence. Compare raw, local-average and regularized gradients.

A concrete situation

A boundary persists across mild smoothing.

What the mathematics does not license

Heavy smoothing moves edges. The habit of checking assumptions is one of the practical benefits of learning mathematics.

Try it, check it, transfer it

Show displacement. Then explain the result to someone who has not seen the formula. If the explanation needs a hidden assumption, add that assumption to the model record.

---

16. External validation

The central mathematical idea is concrete. Test the frozen boundary on new observations.

A concrete situation

Independent samples show the same gradient jump.

What the mathematics does not license

Persistence remains descriptive. The same reasoning helps students move from an exercise to a real dataset.

Try it, check it, transfer it

Investigate mechanisms. Then explain the result to someone who has not seen the formula. If the explanation needs a hidden assumption, add that assumption to the model record.

---

A seven-day practice lab

1. Day 1: define the decision, observation unit, X variables and Y or reference condition.
2. Day 2: reproduce the smallest worked example and check every dimension, unit and sign.
3. Day 3: build a transparent baseline and write what it assumes.
4. Day 4: fit the main method inside a training-only pipeline; record every parameter.
5. Day 5: challenge one assumption with a deliberately difficult dataset or operating regime.
6. Day 6: compare validation metrics with the practical workload created by errors.
7. Day 7: present one supported conclusion, one unresolved uncertainty and one safe next experiment.

Questions for a student, parent or mentor

• Which part of the model is a mathematical definition, and which part is a choice?
• What small calculation would catch a sign, scale or dimension error?
• What evidence comes from training, validation and the untouched test?
• What benign condition could look like a fault or prediction failure?
• How would the answer change for another response, asset, school project or time period?
• What should a human review before acting on the score?

A method-specific audit before deployment

Wombling looks for places where a spatial surface changes rapidly. Instead of searching for a high-value cluster, it estimates local gradients and identifies connected elements with unusually large gradient magnitude or flux. The idea began with Womble’s differential systematics and now appears in ecology, geography, image analysis and other fields where boundaries matter more than smooth centres.

For point observations, a Delaunay triangulation creates local triangles. The values at a triangle’s three vertices define a plane f(x,y)=Gx x+Gy y+C. Its slope vector G=(Gx,Gy) estimates the local direction and rate of change. Large magnitudes suggest candidate boundary elements. Voronoi cells, neighbour differences and raster gradients provide related constructions.

Take triangle vertices (0,0),(1,0),(0,1) with values 2,5,3. The fitted plane is f=3x+y+2, so the gradient is (3,1) and its magnitude is √10≈3.162. The steepest increase points mostly in the positive x direction. Whether this triangle belongs to a meaningful boundary depends on the reference distribution, surrounding gradient alignment, sampling density and a rule for joining candidates.

A large local gradient can come from measurement noise, uneven sampling or one outlier. Selecting and connecting only visually attractive triangles creates a severe search problem. A responsible analysis estimates gradients with declared geometry, calibrates maximum or boundary-wide flux through permutation or simulation, checks denoising and mesh sensitivity, reports broken as well as connected candidates and validates the boundary in independent data.

---

Frequently asked questions

What does wombling detect?

It detects zones where a spatial surface changes rapidly, rather than simply locating high values.

What is a spatial gradient?

It is a vector whose direction shows fastest increase and whose magnitude measures the local rate of change.

Why use Delaunay triangles?

Three nearby sampled vertices define a local plane and therefore a gradient estimate.

What makes a boundary candidate?

A declared rule selects elements with large gradient magnitude, aligned direction or high flux relative to a reference.

Why is significance difficult?

Many local elements and many possible connected boundaries are searched, and gradients share observations.

Can a boundary be causal?

No. It is a descriptive discontinuity until independent data and domain evidence explain it.

---

Primary source and fact-check note

William H. Womble, “Differential systematics,” Science 114(2961), 315–322 (1951), DOI 10.1126/science.114.2961.315. Definitions and core claims in this article were checked against this primary publication. Calculations shown here were independently recomputed. Examples are educational illustrations; they are not reported industrial performance results.

Useful next reading

• Geary’s C local spatial differences
• Oden’s Ipop statistics
• How Mathematics Works

Keep the options open. Mathematics education supports clearer questions, transferable problem-solving skills and informed choices across science, engineering, computing, AI, statistics, design and many other careers. It does not guarantee intelligence, income, admission or employment; those outcomes depend on far more than one subject or model.

Extended laboratory: topic-specific validation and responsible use

Sampling audit

Map location density and measurement error.

Geometry audit

Check projection, neighbours and mesh quality.

Gradient audit

Reproduce one local plane and vector by hand.

Selection audit

Archive every candidate element and joining rule.

Null audit

Simulate or permute the complete boundary search.

Sensitivity audit

Vary mesh and denoising within a plan.

Transfer audit

Collect independent samples across the frozen boundary.

Final claim

Report a discontinuity without inventing its cause.

Mechanism workshop: calculate, challenge and transfer

Workshop 1: Spatial samples

Begin with this mechanism: Record coordinates and measured values. Recreate the concrete case: Three vertices carry values 2,5,3. Write coordinates, values, triangle vertices, plane coefficients, gradient components, magnitudes, directions, normal fluxes, candidate flags, connected paths, boundary scores, permutation maxima and validation measurements with units, denominators and assumptions. Calculate the smallest non-trivial example by hand and reproduce it in a spreadsheet or short program. Challenge this limitation: Uneven density changes resolution. Complete the transfer step: Map sampling effort. Change exactly one observed count, expected count, population standard, gamma parameter, reliability weight, neighbour edge, rho value, precision, constraint, time period, interaction type, coordinate, triangle, gradient threshold, connection rule, null model, replicate count, random seed, observation period or validation region. Explain which mathematical object changes, which conclusion survives and which claim must weaken. Keep observation, preprocessing, geometry, likelihood, prior or null model, computation, uncertainty, interpretation and independent validation on separate notebook lines. Reserve complete regions, periods or samples outside tuning when testing transfer.

Workshop 2: Coordinate system

Begin with this mechanism: Use meaningful distance units. Recreate the concrete case: Projected metres replace latitude degrees. Write coordinates, values, triangle vertices, plane coefficients, gradient components, magnitudes, directions, normal fluxes, candidate flags, connected paths, boundary scores, permutation maxima and validation measurements with units, denominators and assumptions. Calculate the smallest non-trivial example by hand and reproduce it in a spreadsheet or short program. Challenge this limitation: Distortion changes slopes. Complete the transfer step: Choose an appropriate projection. Change exactly one observed count, expected count, population standard, gamma parameter, reliability weight, neighbour edge, rho value, precision, constraint, time period, interaction type, coordinate, triangle, gradient threshold, connection rule, null model, replicate count, random seed, observation period or validation region. Explain which mathematical object changes, which conclusion survives and which claim must weaken. Keep observation, preprocessing, geometry, likelihood, prior or null model, computation, uncertainty, interpretation and independent validation on separate notebook lines. Reserve complete regions, periods or samples outside tuning when testing transfer.

Workshop 3: Delaunay mesh

Begin with this mechanism: Connect points without interior circumcircle points. Recreate the concrete case: The three samples form one triangle. Write coordinates, values, triangle vertices, plane coefficients, gradient components, magnitudes, directions, normal fluxes, candidate flags, connected paths, boundary scores, permutation maxima and validation measurements with units, denominators and assumptions. Calculate the smallest non-trivial example by hand and reproduce it in a spreadsheet or short program. Challenge this limitation: Long skinny triangles are unstable. Complete the transfer step: Audit mesh quality. Change exactly one observed count, expected count, population standard, gamma parameter, reliability weight, neighbour edge, rho value, precision, constraint, time period, interaction type, coordinate, triangle, gradient threshold, connection rule, null model, replicate count, random seed, observation period or validation region. Explain which mathematical object changes, which conclusion survives and which claim must weaken. Keep observation, preprocessing, geometry, likelihood, prior or null model, computation, uncertainty, interpretation and independent validation on separate notebook lines. Reserve complete regions, periods or samples outside tuning when testing transfer.

Workshop 4: Local plane

Begin with this mechanism: Fit f=Gx x+Gy y+C through three values. Recreate the concrete case: The plane is f=3x+y+2. Write coordinates, values, triangle vertices, plane coefficients, gradient components, magnitudes, directions, normal fluxes, candidate flags, connected paths, boundary scores, permutation maxima and validation measurements with units, denominators and assumptions. Calculate the smallest non-trivial example by hand and reproduce it in a spreadsheet or short program. Challenge this limitation: Three points fit exactly, including noise. Complete the transfer step: Use smoothing diagnostics. Change exactly one observed count, expected count, population standard, gamma parameter, reliability weight, neighbour edge, rho value, precision, constraint, time period, interaction type, coordinate, triangle, gradient threshold, connection rule, null model, replicate count, random seed, observation period or validation region. Explain which mathematical object changes, which conclusion survives and which claim must weaken. Keep observation, preprocessing, geometry, likelihood, prior or null model, computation, uncertainty, interpretation and independent validation on separate notebook lines. Reserve complete regions, periods or samples outside tuning when testing transfer.

Workshop 5: Gradient vector

Begin with this mechanism: Read G=(Gx,Gy). Recreate the concrete case: G=(3,1). Write coordinates, values, triangle vertices, plane coefficients, gradient components, magnitudes, directions, normal fluxes, candidate flags, connected paths, boundary scores, permutation maxima and validation measurements with units, denominators and assumptions. Calculate the smallest non-trivial example by hand and reproduce it in a spreadsheet or short program. Challenge this limitation: Direction depends on coordinate orientation. Complete the transfer step: State axes. Change exactly one observed count, expected count, population standard, gamma parameter, reliability weight, neighbour edge, rho value, precision, constraint, time period, interaction type, coordinate, triangle, gradient threshold, connection rule, null model, replicate count, random seed, observation period or validation region. Explain which mathematical object changes, which conclusion survives and which claim must weaken. Keep observation, preprocessing, geometry, likelihood, prior or null model, computation, uncertainty, interpretation and independent validation on separate notebook lines. Reserve complete regions, periods or samples outside tuning when testing transfer.

Workshop 6: Gradient magnitude

Begin with this mechanism: Compute ||G||. Recreate the concrete case: √10≈3.162 units per distance. Write coordinates, values, triangle vertices, plane coefficients, gradient components, magnitudes, directions, normal fluxes, candidate flags, connected paths, boundary scores, permutation maxima and validation measurements with units, denominators and assumptions. Calculate the smallest non-trivial example by hand and reproduce it in a spreadsheet or short program. Challenge this limitation: Large magnitude can be one outlier. Complete the transfer step: Inspect vertex influence. Change exactly one observed count, expected count, population standard, gamma parameter, reliability weight, neighbour edge, rho value, precision, constraint, time period, interaction type, coordinate, triangle, gradient threshold, connection rule, null model, replicate count, random seed, observation period or validation region. Explain which mathematical object changes, which conclusion survives and which claim must weaken. Keep observation, preprocessing, geometry, likelihood, prior or null model, computation, uncertainty, interpretation and independent validation on separate notebook lines. Reserve complete regions, periods or samples outside tuning when testing transfer.

Workshop 7: Gradient direction

Begin with this mechanism: Convert vector to an angle. Recreate the concrete case: The steepest rise points mostly east. Write coordinates, values, triangle vertices, plane coefficients, gradient components, magnitudes, directions, normal fluxes, candidate flags, connected paths, boundary scores, permutation maxima and validation measurements with units, denominators and assumptions. Calculate the smallest non-trivial example by hand and reproduce it in a spreadsheet or short program. Challenge this limitation: Circular summaries need care. Complete the transfer step: Plot arrows. Change exactly one observed count, expected count, population standard, gamma parameter, reliability weight, neighbour edge, rho value, precision, constraint, time period, interaction type, coordinate, triangle, gradient threshold, connection rule, null model, replicate count, random seed, observation period or validation region. Explain which mathematical object changes, which conclusion survives and which claim must weaken. Keep observation, preprocessing, geometry, likelihood, prior or null model, computation, uncertainty, interpretation and independent validation on separate notebook lines. Reserve complete regions, periods or samples outside tuning when testing transfer.

Workshop 8: Edge-normal flux

Begin with this mechanism: Project gradients across a proposed boundary. Recreate the concrete case: A line normal aligned with G has high flux. Write coordinates, values, triangle vertices, plane coefficients, gradient components, magnitudes, directions, normal fluxes, candidate flags, connected paths, boundary scores, permutation maxima and validation measurements with units, denominators and assumptions. Calculate the smallest non-trivial example by hand and reproduce it in a spreadsheet or short program. Challenge this limitation: Tangential gradients do not cross the line. Complete the transfer step: Separate magnitude and orientation. Change exactly one observed count, expected count, population standard, gamma parameter, reliability weight, neighbour edge, rho value, precision, constraint, time period, interaction type, coordinate, triangle, gradient threshold, connection rule, null model, replicate count, random seed, observation period or validation region. Explain which mathematical object changes, which conclusion survives and which claim must weaken. Keep observation, preprocessing, geometry, likelihood, prior or null model, computation, uncertainty, interpretation and independent validation on separate notebook lines. Reserve complete regions, periods or samples outside tuning when testing transfer.

Workshop 9: Candidate threshold

Begin with this mechanism: Declare a percentile or model-based cutoff. Recreate the concrete case: Top 5% of magnitudes are candidates. Write coordinates, values, triangle vertices, plane coefficients, gradient components, magnitudes, directions, normal fluxes, candidate flags, connected paths, boundary scores, permutation maxima and validation measurements with units, denominators and assumptions. Calculate the smallest non-trivial example by hand and reproduce it in a spreadsheet or short program. Challenge this limitation: A percentile forces candidates under pure noise. Complete the transfer step: Calibrate globally. Change exactly one observed count, expected count, population standard, gamma parameter, reliability weight, neighbour edge, rho value, precision, constraint, time period, interaction type, coordinate, triangle, gradient threshold, connection rule, null model, replicate count, random seed, observation period or validation region. Explain which mathematical object changes, which conclusion survives and which claim must weaken. Keep observation, preprocessing, geometry, likelihood, prior or null model, computation, uncertainty, interpretation and independent validation on separate notebook lines. Reserve complete regions, periods or samples outside tuning when testing transfer.

Workshop 10: Connectivity rule

Begin with this mechanism: Join neighbouring candidate elements. Recreate the concrete case: Adjacent triangles form a chain. Write coordinates, values, triangle vertices, plane coefficients, gradient components, magnitudes, directions, normal fluxes, candidate flags, connected paths, boundary scores, permutation maxima and validation measurements with units, denominators and assumptions. Calculate the smallest non-trivial example by hand and reproduce it in a spreadsheet or short program. Challenge this limitation: A small rule change can bridge gaps. Complete the transfer step: Publish tolerance. Change exactly one observed count, expected count, population standard, gamma parameter, reliability weight, neighbour edge, rho value, precision, constraint, time period, interaction type, coordinate, triangle, gradient threshold, connection rule, null model, replicate count, random seed, observation period or validation region. Explain which mathematical object changes, which conclusion survives and which claim must weaken. Keep observation, preprocessing, geometry, likelihood, prior or null model, computation, uncertainty, interpretation and independent validation on separate notebook lines. Reserve complete regions, periods or samples outside tuning when testing transfer.

Workshop 11: Boundary score

Begin with this mechanism: Sum or average flux along the chain. Recreate the concrete case: A long aligned chain scores highly. Write coordinates, values, triangle vertices, plane coefficients, gradient components, magnitudes, directions, normal fluxes, candidate flags, connected paths, boundary scores, permutation maxima and validation measurements with units, denominators and assumptions. Calculate the smallest non-trivial example by hand and reproduce it in a spreadsheet or short program. Challenge this limitation: Longer paths have more chances. Complete the transfer step: Normalize and simulate. Change exactly one observed count, expected count, population standard, gamma parameter, reliability weight, neighbour edge, rho value, precision, constraint, time period, interaction type, coordinate, triangle, gradient threshold, connection rule, null model, replicate count, random seed, observation period or validation region. Explain which mathematical object changes, which conclusion survives and which claim must weaken. Keep observation, preprocessing, geometry, likelihood, prior or null model, computation, uncertainty, interpretation and independent validation on separate notebook lines. Reserve complete regions, periods or samples outside tuning when testing transfer.

Workshop 12: Permutation null

Begin with this mechanism: Relabel values over fixed coordinates when exchangeable. Recreate the concrete case: Every replicate rebuilds gradients and boundaries. Write coordinates, values, triangle vertices, plane coefficients, gradient components, magnitudes, directions, normal fluxes, candidate flags, connected paths, boundary scores, permutation maxima and validation measurements with units, denominators and assumptions. Calculate the smallest non-trivial example by hand and reproduce it in a spreadsheet or short program. Challenge this limitation: Spatial trend can violate exchangeability. Complete the transfer step: Model trend first. Change exactly one observed count, expected count, population standard, gamma parameter, reliability weight, neighbour edge, rho value, precision, constraint, time period, interaction type, coordinate, triangle, gradient threshold, connection rule, null model, replicate count, random seed, observation period or validation region. Explain which mathematical object changes, which conclusion survives and which claim must weaken. Keep observation, preprocessing, geometry, likelihood, prior or null model, computation, uncertainty, interpretation and independent validation on separate notebook lines. Reserve complete regions, periods or samples outside tuning when testing transfer.

Workshop 13: Simulation null

Begin with this mechanism: Generate surfaces under a fitted smooth model. Recreate the concrete case: Replicates preserve noise and sampling density. Write coordinates, values, triangle vertices, plane coefficients, gradient components, magnitudes, directions, normal fluxes, candidate flags, connected paths, boundary scores, permutation maxima and validation measurements with units, denominators and assumptions. Calculate the smallest non-trivial example by hand and reproduce it in a spreadsheet or short program. Challenge this limitation: Misspecified covariance alters tails. Complete the transfer step: Check diagnostics. Change exactly one observed count, expected count, population standard, gamma parameter, reliability weight, neighbour edge, rho value, precision, constraint, time period, interaction type, coordinate, triangle, gradient threshold, connection rule, null model, replicate count, random seed, observation period or validation region. Explain which mathematical object changes, which conclusion survives and which claim must weaken. Keep observation, preprocessing, geometry, likelihood, prior or null model, computation, uncertainty, interpretation and independent validation on separate notebook lines. Reserve complete regions, periods or samples outside tuning when testing transfer.

Workshop 14: Mesh sensitivity

Begin with this mechanism: Change triangulation or raster resolution within a plan. Recreate the concrete case: A skinny triangle disappears after resampling. Write coordinates, values, triangle vertices, plane coefficients, gradient components, magnitudes, directions, normal fluxes, candidate flags, connected paths, boundary scores, permutation maxima and validation measurements with units, denominators and assumptions. Calculate the smallest non-trivial example by hand and reproduce it in a spreadsheet or short program. Challenge this limitation: Choosing the sharpest mesh is biased. Complete the transfer step: Report stability. Change exactly one observed count, expected count, population standard, gamma parameter, reliability weight, neighbour edge, rho value, precision, constraint, time period, interaction type, coordinate, triangle, gradient threshold, connection rule, null model, replicate count, random seed, observation period or validation region. Explain which mathematical object changes, which conclusion survives and which claim must weaken. Keep observation, preprocessing, geometry, likelihood, prior or null model, computation, uncertainty, interpretation and independent validation on separate notebook lines. Reserve complete regions, periods or samples outside tuning when testing transfer.

Workshop 15: Denoising sensitivity

Begin with this mechanism: Compare raw, local-average and regularized gradients. Recreate the concrete case: A boundary persists across mild smoothing. Write coordinates, values, triangle vertices, plane coefficients, gradient components, magnitudes, directions, normal fluxes, candidate flags, connected paths, boundary scores, permutation maxima and validation measurements with units, denominators and assumptions. Calculate the smallest non-trivial example by hand and reproduce it in a spreadsheet or short program. Challenge this limitation: Heavy smoothing moves edges. Complete the transfer step: Show displacement. Change exactly one observed count, expected count, population standard, gamma parameter, reliability weight, neighbour edge, rho value, precision, constraint, time period, interaction type, coordinate, triangle, gradient threshold, connection rule, null model, replicate count, random seed, observation period or validation region. Explain which mathematical object changes, which conclusion survives and which claim must weaken. Keep observation, preprocessing, geometry, likelihood, prior or null model, computation, uncertainty, interpretation and independent validation on separate notebook lines. Reserve complete regions, periods or samples outside tuning when testing transfer.

Workshop 16: External validation

Begin with this mechanism: Test the frozen boundary on new observations. Recreate the concrete case: Independent samples show the same gradient jump. Write coordinates, values, triangle vertices, plane coefficients, gradient components, magnitudes, directions, normal fluxes, candidate flags, connected paths, boundary scores, permutation maxima and validation measurements with units, denominators and assumptions. Calculate the smallest non-trivial example by hand and reproduce it in a spreadsheet or short program. Challenge this limitation: Persistence remains descriptive. Complete the transfer step: Investigate mechanisms. Change exactly one observed count, expected count, population standard, gamma parameter, reliability weight, neighbour edge, rho value, precision, constraint, time period, interaction type, coordinate, triangle, gradient threshold, connection rule, null model, replicate count, random seed, observation period or validation region. Explain which mathematical object changes, which conclusion survives and which claim must weaken. Keep observation, preprocessing, geometry, likelihood, prior or null model, computation, uncertainty, interpretation and independent validation on separate notebook lines. Reserve complete regions, periods or samples outside tuning when testing transfer.