Studies in Roughness: A Fractal Primer

Measurements disagree when teams change resolution without modeling how the object changes with scale. Fractal roughness turns that disagreement into evidence instead of dismissing it as instrument noise.

Enterprise problem: organizations need metrics that remain interpretable when observation moves from seconds to days, functions to systems, or individual tasks to portfolios.

Whole-course context: after previewing turbulent bursts, the course now develops the geometry and scaling language needed to describe irregular paths across resolutions.

Today’s slice: this chapter builds intuition from coastline measurement, self-similarity, self-affinity, fractal dimension, and the Hurst exponent.

End-of-day evidence: you will calculate a ruler-length scaling estimate, operate roughness labs, and state the finite range over which a pattern holds.

Still unsolved: one roughness exponent cannot capture every market feature; later chapters add tails, long memory, regimes, and multifractal spectra.

Fractal language is often used decoratively. Precise definitions keep “fractal” tied to measurements across scale.

An irregular boundary has no single observed length independent of ruler size. A long ruler skips bays and headlands; a short ruler follows more detail and reports a larger total.

Lewis Fry Richardson collected such scale-dependent measurements, and Mandelbrot connected them to fractal dimension. The lesson extends beyond geography: the measurement protocol is part of the result. Asking “How long?” without asking “At what resolution?” is incomplete.

Markets raise the analogous question: how does price variation change from minute to hour to day? Software metrics change between request, minute, and month. A scaling law is valuable when it predicts this transformation and exposes where it stops.

Cover a curve with rulers of length ε. If N(ε) rulers are needed, measured length is approximately L(ε) = N(ε)ε.

For a smooth curve, halving ε roughly doubles N, so total length stabilizes. For a fractal curve, count may scale as N(ε) ∝ ε^-D, where D is dimension. Then L(ε) ∝ ε^(1-D). When D > 1, measured length grows as the ruler shrinks.

Taking logarithms gives log N(ε) ≈ -D log ε + constant. Estimate D from the negative slope of a log-count versus log-ruler plot over a justified range. A straight-looking segment with three points is weak evidence; report uncertainty and endpoints.

Time series use different units on horizontal and vertical axes, so self-affinity matters more than geometric self-similarity. If typical increments obey |X(t+h)-X(t)| ∝ h^H, time scales by a while amplitude scales by a^H.

Brownian motion has H = 0.5. Values above 0.5 indicate persistence in the modeled process; values below 0.5 indicate anti-persistence. Interpretation depends on what series is measured and whether trends or breaks have been removed.

Scale estimates become clearer with exact counts. Suppose rulers of length 1, 1/2, 1/4, and 1/8 require 8, 20, 49, and 120 pieces.

Using the first and last points:

D ≈ log(120/8) / log(8/1) = log(15) / log(8) ≈ 1.30

Measured lengths are 8, 10, 12.25, and 15 units. The reported length grows as resolution becomes finer, consistent with a curve rougher than a smooth line but not filling a plane.

This estimate is illustrative. A real analysis fits all points, checks residuals, compares nearby scale ranges, and accounts for finite measurement. Dimension 1.30 does not reveal the boundary’s causal history.

Visual roughness can be confused with the smooth trend underneath it. The 2D lab changes roughness while preserving a labeled smooth reference.

Use the deterministic teaching model as follows:

1. Press Reset and note the Roughness percentage.
2. Identify the Smooth trend and Fractal trace from labels and distinct encodings.
3. Increase Roughness one step and compare the added fine-scale deviations across the time axis.
4. Record the Roughness gap at low, default, and high settings.
5. Reset and confirm that the default curves and metric return exactly.

The output is a teaching simulation, not a forecast or proof that a real series is fractal. Read the labeled curves and Roughness gap rather than selecting the most convincing-looking segment.

Roughness can vary across position and scale, which a single line suppresses. The 3D surface shows how fine structure sits inside coarse terrain while keeping the axes and concentration readouts explicit.

Operate the surface methodically:

1. Read the legend: x is position, y is scale, and z is roughness.
2. Move the Roughness slider and compare the reported surface range.
3. Select a position with the x focus slider and a scale with the y focus slider; record selected x, y, and z.
4. Hold one focus value fixed while moving the other to compare coarse and fine structure.
5. Orbit by pointer or keyboard, then Reset the parameter, focus, and camera.

The mesh is a finite generated terrain. Interpolation and camera perspective can make patterns appear smoother or steeper, so numeric readouts and axis labels carry the interpretation.

A rough boundary appears simpler when observation skips its fine bends, so comparisons made at one convenient resolution can conceal real operational detail. This fractal lens refines a deterministic coastline and reports mean absolute vertical change across dyadic index lags.

Parameters. Iteration depth d ranges from 2 to 8 and adds recursive coastline refinements. Roughness r ranges from 0.10 to 1.00 and increases midpoint displacement, creating a more irregular teaching boundary.

Protocol. Reset the Fractal dimension lens and record Iteration depth, Roughness, the Estimated dimension, and normalized mean absolute change across displayed lags. Raise depth while holding Roughness fixed and note whether the estimate stabilizes; restore it, raise Roughness, and compare the response over the same lag range. The readout estimates D = 2 - H from a log–log regression computed internally over five dyadic lags; Reset before combining parameters.

Assumptions. The coastline is a finite deterministic construction with an exact replacement rule and noise-free coordinates. The ruler method, displayed range, and finite depth determine the estimate; no geographic, market, software, or workload boundary is calibrated by these controls.

Interpretation. Greater mean change at smaller lags demonstrates scale-dependent measurement, while Estimated dimension summarizes only the finite constructed path and five fitted lags. It does not establish universal self-similarity or causation; transfer the method only after defining the real object, ruler, valid range, uncertainty, and decision consequence.

Power laws are easy to see in short log–log plots and hard to establish. Trends, seasonality, structural breaks, and mixtures can imitate scaling.

The estimated exponent depends on horizon range, estimator, overlap, and finite sample. Very short financial horizons contain market microstructure; very long horizons contain few observations and regime change. A scaling range should be chosen for substantive and statistical reasons, then tested elsewhere.

One exponent also compresses heterogeneity. If small and large fluctuations scale differently, a monofractal H is incomplete and a multifractal spectrum may help. Extra flexibility increases estimation risk, so validation must use withheld moments or periods.

Engineering metrics mislead when repository structure, service traffic, model conversations, and agent tool traces are measured at incompatible resolutions. A scale-aware review names the observation unit before it compares complexity, cost, or failure.

Measure one system at three levels: coarse ownership or task boundaries, intermediate services or agent steps, and fine traces or model tokens. Record where edge count, error concentration, or recovery cost changes fastest, then place budgets and observability at that transition. The software, LLM, and agent sections below keep their distinct mechanisms while sharing the same discipline: no scaling claim without an object, ruler, range, and falsifier.

Software architecture can look simple at repository level and rough at dependency-call level. Metrics must state whether they observe functions, services, deployments, incidents, or months.

A service graph’s edge count grows when dynamic calls, retries, and shared infrastructure are measured. A coarse diagram can therefore understate operational surface area. Conversely, counting every trace edge can obscure stable module boundaries.

Use multiple resolutions deliberately: code ownership for change planning, service dependencies for blast radius, and trace spans for incident diagnosis. Look for scale transitions where one abstraction stops predicting effort or failure.

LLM quality changes with token, answer, conversation, task family, and deployment scale. A single average can mix these resolutions into an uninterpretable number.

Token loss may improve smoothly with compute while rare long-context tasks remain rough. An answer-level judge can miss a conversation-level contradiction. Aggregating users can hide a subgroup tail.

Report metrics at aligned scales and test how error moments change with context length or interaction count. Scaling curves describe measured ranges; they do not guarantee indefinite improvement beyond available data.

Agent tasks often split recursively into plans, subtasks, tool calls, retries, and validations. The repeated branching resembles a scale hierarchy, but resemblance alone does not establish a fractal law.

Measure node count, depth, cost, and failure concentration as task resolution changes. A coarse “one task” metric can hide a hundred tool actions; a fine metric can miss that several actions share one mistaken premise.

Use scale-aware budgets: total run cap, branch cap, per-tool limit, and checkpoint frequency. Determine where cost grows faster than useful evidence and stop recursion before it becomes an uncontrolled workload surface.

Startup growth can appear smooth monthly while daily acquisition is dominated by campaigns and launches. Decisions fail when a scale convenient for reporting is mistaken for the scale of operations.

Measure signups daily for capacity, cohorts monthly for retention, and runway across financing horizons. A power-law claim about customer size should specify threshold and range; one large customer is not enough.

Scale analysis can reveal concentration and transition points. It cannot predict which channel or customer creates the next burst, so commitments should remain robust to range uncertainty.

Business reporting fails when daily service constraints are governed by monthly averages or long-term strategy is rewritten from one volatile week. Each decision needs a ruler that preserves the variation capable of changing it.

Report demand, backlog, margin, and customer concentration at one operational, one planning, and one strategic horizon. Identify where aggregation changes the conclusion, test whether the relationship persists in another period, and assign a separate response at each scale. A repeated visual pattern is only a hypothesis until the range and measurement rule are stated.

Personal progress depends on observation scale. Daily noise can hide a monthly trend, while monthly totals can hide exhausting sequences.

Exercise measured by one session, weekly consistency, and yearly health tells different stories. Calendar busyness measured by event count misses preparation and recovery. No one ruler is universally correct.

Choose the ruler that matches the decision, compare at least one finer and coarser view, and avoid moralizing noise. Scale awareness supports kinder, more accurate planning rather than constant self-surveillance.

A scaling statement is weak when it omits the object, ruler, exponent, and range. This exercise produces a claim another analyst can reproduce and reject.

Choose a curve or ordered series with enough resolution. Define at least five ruler sizes or horizons before measuring. For a boundary, count covering segments; for a series, compute a typical increment magnitude or an unrooted absolute moment. Plot both axes logarithmically and fit only the range justified by measurement and mechanism.

Use the 2D lab to compare the Smooth trend and Fractal trace as the Roughness slider changes, recording the Roughness gap. Estimate slope and residual patterns from your chosen external series rather than from the widget. On the 3D surface, compare coarse and fine structure with the position and scale focus sliders and selected x, y, and z values rather than camera impression.

Now challenge the claim. Shift the fitting endpoints, use a later window, remove a trend, and compare overlapping with non-overlapping increments. Record how much the exponent moves. If small choices produce large movement, report instability rather than averaging it away.

Use a sensitivity table:

Fractal roughness is not a visual style; it is a measured relationship between detail and scale. The exponent summarizes how observation changes when the ruler changes.

choose object + ruler
        |
measure across resolutions
        |
fit exponent over finite range
        |
test alternate ranges and windows
        |
state what the exponent does not explain

This discipline prepares the later market argument. Before claiming universal scaling, preserve units, distinguish self-similarity from self-affinity, and report the range where evidence supports the rule.

Fractal concepts are safest when traced to primary definitions and measurements. These sources establish coastline scaling, self-affine stochastic paths, and long-range dependence.

• Benoit Mandelbrot and Richard L. Hudson, The Misbehavior of Markets, Basic Books.
• Benoit Mandelbrot, “How Long Is the Coast of Britain? Statistical Self-Similarity and Fractional Dimension”, Science 156(3775), 1967.
• Benoit Mandelbrot and John W. Van Ness, “Fractional Brownian Motions, Fractional Noises and Applications”, SIAM Review 10(4), 1968.
• H. E. Hurst, “Long-Term Storage Capacity of Reservoirs”, Transactions of the American Society of Civil Engineers 116, 1951.

Roughness becomes scientific only when tied to scale, exponent, range, and uncertainty. Visual resemblance is an invitation to measure, not a conclusion.

• Shorter rulers reveal more detail on an irregular boundary.
• Fractal dimension describes how count grows as ruler size shrinks.
• Time series are usually self-affine because time and amplitude scale differently.
• Brownian motion provides the H = 0.5 reference.
• Finite samples, trends, and breaks can create spurious scaling.
• A single exponent may fail when fluctuation sizes scale differently.

The primer is complete when every roughness claim includes a ruler and a range. Apply these checks to one curve or time series.

• [ ] I can distinguish self-similarity from self-affinity.
• [ ] I can calculate a two-point fractal-dimension estimate.
• [ ] I can explain the Hurst exponent in plain language.
• [ ] I can operate both labs using numeric and non-color readouts.
• [ ] I can state the fitted scale range and sample limitations.
• [ ] I can name three mechanisms that imitate scaling.
• [ ] I can identify a scale transition in a non-financial system.
• [ ] I can explain why simulated fractal terrain is not a forecast.

Source trace. Chapter VII, printed p. 134, supplied PDF pp. 298–300; three construction stages lead to the completed Sierpinski gasket.

Adaptation. This exact mathematical reimplementation retains the three corner triangles after every midpoint split; depth d therefore contains 3^d triangles. Layout and color are teaching adaptations.

Controls. Detail, Panel separation, and View scale alter depth, spacing, and magnification; linked/solo mode selects the comparison scope.

Protocol. Increase Detail one step at a time, compare linked stages, then isolate the completed panel and vary scale.

Readout/evidence. Use the displayed triangle count and panel gap; evidence is the exact 3^d count derived from the same geometry on screen.

Assumptions. The rule is deterministic, midpoint-based, finite-depth self-similarity.

Falsifier. A retained center, non-midpoint split, or count other than 3^d invalidates the construction.

Source trace. Chapter VII, printed p. 135, supplied PDF pp. 301–302; one recursive tetrahedral skewed web.

Adaptation. Four half-scale tetrahedra replace each parent in a true orbitable 3D scene; skew changes geometry without changing recursive connectivity.

Controls. Detail, Skew, and View scale rebuild or magnify the web; pointer and keyboard orbit change perspective, and any interaction stops ambient motion.

Protocol. Hold Detail fixed while changing Skew, orbit to inspect occluded edges, then raise Detail and compare edge density.

Readout/evidence. The displayed detail, skew, scale, and derived geometry readout share the rendered model; orbit changes viewpoint only.

Assumptions. Projection may hide edges but cannot alter connectivity; children remain half-scale tetrahedral copies.

Falsifier. Disconnected children, wrong scaling, or topology that changes merely when the camera rotates falsifies the adaptation.

Source trace. Chapter VII, printed p. 136, supplied PDF pp. 303–304; successive middle-third deletion stages form Cantor dust.

Adaptation. Every open middle third is removed exactly, so depth d leaves 2^d segments, each of length 3^-d.

Controls. Detail, Stage spacing, and View scale expose deletion depth and legibility; linked/solo mode controls comparison.

Protocol. Advance Detail, inspect each deletion stage, and verify gaps persist when zoomed rather than being rendering artifacts.

Readout/evidence. The displayed segment count and gap spacing are derived from the rendered stage; compare them with 2^d.

Assumptions. The mathematical limit is represented only to finite depth and uses exact thirds.

Falsifier. Any retained open middle third, incorrect endpoint, or non-power-of-two segment count falsifies it.

Source trace. Chapter VII, printed p. 137, supplied PDF pp. 305–306; three construction stages and a completed Koch curve.

Adaptation. Every segment becomes four equal one-third segments with ±60-degree middle turns, preserving endpoints while length grows by 4/3 per depth.

Controls. Detail, Panel separation, and View scale change construction depth, layout, and magnification.

Protocol. Compare linked stages, increase Detail, then isolate the completed curve and inspect whether new detail preserves endpoints.

Readout/evidence. The displayed segment count must equal 4^d; length growth follows the same rendered curve rather than a decorative estimate.

Assumptions. Finite depth approximates an ideal deterministic limit with exact thirds and turns.

Falsifier. Unequal thirds, wrong turn angles, lost endpoints, or a count other than 4^d invalidates the mechanism.

Source trace. Chapter VII, printed p. 138, supplied PDF pp. 307–308; coarse and fine ruler overlays motivate fractal dimension.

Adaptation. Both panels place actual chord brackets over the same displayed Koch polyline. Coarse and fine covers report their true ε and N(ε), and dimension is computed from their log ratio.

Controls. Detail, Ruler ratio, and View scale choose the curve depth, cover gap, and magnification.

Protocol. Hold Detail fixed, compare coarse and fine brackets, change Ruler ratio, and record both counts before reading dimension.

Readout/evidence. The live readout reports actual cover counts, ruler lengths, and displayed dimension from the same bracket geometry.

Assumptions. A two-scale slope is evidence only over the displayed finite scale range.

Falsifier. Brackets that do not cover the curve, inconsistent N(ε), or dimension instability across adjacent covers falsifies the claim.

Source trace. Chapter VII, printed p. 139, supplied PDF pp. 309–310; a random fractal curve is paired with coastline context.

Adaptation. A randomized recursive Koch construction chooses orientation and bounded height variation at each replacement. It preserves recursive thirds and endpoints; the coastline panel is conceptual, not observed geography.

Controls. Detail, Randomness, and View scale alter recursive depth, perturbation strength, and view. The deterministic teaching seed is fixed and is not a control.

Protocol. Compare both panels in linked mode, raise Randomness at fixed Detail, then raise Detail and inspect whether structure persists across visible scales.

Readout/evidence. The readout reports recursive segment count and vertical range derived from the same displayed randomized Koch path.

Assumptions. One fixed realization illustrates a mechanism but cannot establish an ensemble law or geographic fit.

Falsifier. Lost endpoints, nonrecursive perturbations, or a readout calculated from data other than the displayed path invalidates the adaptation.

Source trace. Chapter VII, printed pp. 140–141, supplied PDF pp. 311–313; five curdling stages lead to completed random dust.

Adaptation. Every retained parent proposes a genuine 5 × 5 × 5 = 125-way curdling set before seeded retention and 2D projection. The hierarchy is synthetic and not astronomical data.

Controls. Detail, Retention, and View scale change the retained-cell budget, retention probability, and magnification; linked/solo mode selects stages.

Protocol. Trace stages in linked mode, raise Retention at fixed Detail, then inspect a solo stage without changing the model.

Readout/evidence. The live readout gives actually retained projected cells and explicitly states 125 candidates per parent from the same curdling model.

Assumptions. Retention is stationary conditional on a retained parent, projection overlaps are possible, and cell count is bounded for interaction.

Falsifier. Any parent with a non-125 candidate mechanism, retained children without a parent, or indistinguishability from a matched independent-dot control falsifies hierarchy.

Source trace. Chapter VII, printed p. 142, supplied PDF pp. 314–315; a cloud field and branching cluster illustrate physical fractal forms.

Adaptation. The panels use distinct bounded conceptual processes—a seeded cloud proxy and a DLA-inspired branching schematic—so resemblance is not presented as shared causation.

Controls. Detail, Branching bias, and View scale alter particle count, branch geometry, and view.

Protocol. Compare linked panels at matched Detail, vary Branching bias, then inspect each alone and identify which evidence is mechanism-specific.

Readout/evidence. Displayed particle count and maximum radius quantify the cloud proxy; visible branch geometry supplies separate evidence.

Assumptions. The diagrams are qualitative teaching processes without atmospheric or aggregation calibration.

Falsifier. Failure on held-out scale, branch-density, or void statistics—or claiming resemblance proves one cause—falsifies the interpretation.

Source trace. Chapter VII, printed p. 143, supplied PDF pp. 316–317; a branching generator appears beside lung bronchia.

Adaptation. Two recursive branching schematics compare generator and anatomical context; neither is patient data, synthetic anatomy, or a diagnostic model.

Controls. Detail, Branch angle, and View scale change recursion, spread, and magnification.

Protocol. Hold Detail fixed while varying Branch angle, compare both panels, then raise Detail and record branch growth.

Readout/evidence. Branch count and angle come from the same displayed geometry; the context label keeps biological claims bounded.

Assumptions. Symmetry, physiology, diameter, flow, and measured anatomical variability are omitted.

Falsifier. Measured branching that requires incompatible level-specific rules falsifies a single recursive generator as an adequate model.

Source trace. Chapter VII, printed p. 144, supplied PDF pp. 318–319; Ba-ili settlement context is paired with enclosure, dwelling, and altar hierarchy.

Adaptation. Four schematic panels preserve the reported nesting levels without fabricating residents, observations, or a generative ethnography.

Controls. Detail, Nesting, and View scale change schematic density, stroke nesting, and magnification; linked/solo mode changes scope.

Protocol. Compare all four levels, vary Nesting without changing labels, then isolate each panel and check source-level mapping.

Readout/evidence. Settlement cell count and nesting stroke are displayed-model metrics, not counts of observed people or structures.

Assumptions. Visual nesting is descriptive and does not establish cultural cause, universality, or normative hierarchy.

Falsifier. Incorrect source-level mapping, dominant counterexamples, or claims of observed counts from schematic values falsifies the adaptation.

Source trace. Chapter VII, printed pp. 144–145, supplied PDF pp. 320–321; panels A–D move from overview to linked Mandelbrot-set views.

Adaptation. Every displayed cell iterates z ← z² + c; Detail changes resolution and iteration cap, while Viewport and View scale recompute the sampled complex plane.

Controls. Detail, Viewport, and View scale control sampling, location, and zoom; linked/solo mode selects panels.

Protocol. Compare A–D, raise Detail at fixed viewport, then zoom a boundary and check whether classifications stabilize or escape later.

Readout/evidence. Each panel reports non-escaping cells over displayed cells from the same iteration grid used to draw it.

Assumptions. Finite iterations show “not yet escaped,” not proven mathematical membership.

Falsifier. Any recurrence other than z² + c, inconsistent escape classification, or unchanged pixels after a genuine viewport change falsifies it.