WRWEB ENGINEERING
REFORGING45 · 1/7
ALL LESSONS→WEEK 45→MONDAY

UNDERSTAND · PERFORMANCE, SECURITY + ACCESSIBILITY

Latency and throughput

Performance without mythology30–60 MINUTESCORE + PRACTICAL
01

GROUND

Problem

What becomes confusing, fragile, or impossible without understanding latency and throughput? This lesson answers that through explanation, a worked example, two runnable exercises, and a reference solution. No teacher-supplied worksheet is required.

Performance, security, and accessibility became disciplines because functional software can still exclude, leak, stall, or harm.
02

LEARN

Concept explanation

Latency and throughput belongs to “Performance without mythology”. Performance requires distributions and traces; accessibility/security require semantic and trust invariants.

For latency and throughput, trace concrete input, state transition, output, and failure through a system invariant made testable through budgets, trust boundaries, semantic interfaces, and controlled failure.

Quality emerges from system invariants: bounded work, controlled authority, explicit trust, semantic interfaces, and observable failure. Apply that model to supplied normal, boundary, and failure cases; each case below names its input and expected evidence.

Observable

Evidence produced by the latency and throughput experiment: output, state, trace, bytes, timing, or diagnostics.

Invariant

Condition that must remain true while inputs or implementation of latency and throughput change.

Boundary

Point where latency and throughput crosses ownership, representation, time, process, network, or trust.

Example bank

Compare normal, boundary, failure, and cross-layer cases. Predict each observation before revealing the explanation.

Baseline · one variable

SETUPProfile handbook load and one interaction; record Web Vitals and request waterfall.

OBSERVEBudget cites trace evidence and separates CPU, network, rendering, and measurement noise.

WHY IT MATTERSThis isolates the normal contract of latency and throughput; preserve its raw evidence as the control for every later comparison.

Boundary · same contract, harder input

SETUPRepeat under CPU/network throttling and 200% zoom.

OBSERVERecord what remains invariant and the first representation, owner, size, or timing value that changes in Chromium Performance · Memory · Accessibility · Security.

WHY IT MATTERSA boundary example is useful only when one named dimension changes and everything else stays comparable.

Failure · evidence before repair

SETUPAdd long task, locate it in trace, remove cause, compare before/after.

OBSERVECapture the first divergence from the baseline, including exact input, diagnostic, state, and recovery result. Expected recovery: Budget cites trace evidence and separates CPU, network, rendering, and measurement noise.

WHY IT MATTERSThe diagnostic is part of the interface. Repair the proven cause, not the most visible symptom.

Cross-layer · follow ownership

SETUPTrace latency and throughput one layer below its usual abstraction through a system invariant made testable through budgets, trust boundaries, semantic interfaces, and controlled failure.

OBSERVECapture profiles, accessibility tree, keyboard order, security headers, trust boundaries, budgets, and recovery evidence.

WHY IT MATTERSThe lower layer is earned when it explains evidence the current layer cannot. Otherwise keep latency and throughput at the simpler boundary.

03

SEE

Worked example

Start from supplied index.html. Focus: Profile handbook load and one interaction; record Web Vitals and request waterfall.

  1. Run: Serve and open index.html; keyboard-activate the probe; inspect Performance, Accessibility, and Security panels.
  2. Save baseline evidence. Capture profiles, accessibility tree, keyboard order, security headers, trust boundaries, budgets, and recovery evidence.
  3. Boundary case: Repeat under CPU/network throttling and 200% zoom.
  4. Failure case: Add long task, locate it in trace, remove cause, compare before/after.
RESULT
Budget cites trace evidence and separates CPU, network, rendering, and measurement noise. Starter-level baseline: Heading and button are keyboard reachable; activation updates announced output with a non-negative duration.
04

START HERE

Starter material

PREREQUISITESModern Chromium with Performance, Memory, Accessibility, and Security panels. Run security payloads only in supplied local fixtures.

ONE-TIME SETUPSave fixture files, run npx --yes serve ., then open the printed localhost URL.

Create index.html, paste this exact content, then run the command below.

<!doctype html>
<meta charset="utf-8">
<title>latency and throughput</title>
<main><h1>latency and throughput</h1><button id="probe">Run controlled probe</button><output id="result" aria-live="polite">Not run</output></main>
<script>
probe.addEventListener('click', () => {
  const start = performance.now();
  result.value = 'completed in ' + (performance.now() - start).toFixed(2) + 'ms';
});
</script>
RUNServe and open index.html; keyboard-activate the probe; inspect Performance, Accessibility, and Security panels.

STOP / CLEANUPClose recordings and press Ctrl+C in server terminal. Never aim controlled payloads at third-party systems.

05

DO WITH GUIDANCE

Guided exercise

Observe one rule: latency and throughput

  1. Normal case: Profile handbook load and one interaction; record Web Vitals and request waterfall.
  2. Write predicted evidence from this named case before running starter.
  3. Change one input while holding environment constant.
  4. Run exact normal case. Save commands, inputs, outputs, and diagnostics in notebook.
  5. Explain changed evidence using lesson mental model in no more than five sentences.
Concrete guided solution
  1. Copy the supplied index.html unchanged and run: Serve and open index.html; keyboard-activate the probe; inspect Performance, Accessibility, and Security panels.
  2. Write this prediction before inspecting output: Budget cites trace evidence and separates CPU, network, rendering, and measurement noise.
  3. Perform only the named normal case: Profile handbook load and one interaction; record Web Vitals and request waterfall.
  4. Save the raw output, then annotate input → transition → evidence. Use Chromium Performance · Memory · Accessibility · Security to confirm the transition rather than inferring it.
  5. Compare prediction with evidence; if they differ, keep both and write the rule that explains the difference. Reference baseline: Heading and button are keyboard reachable; activation updates announced output with a non-negative duration.
06

DO ALONE

Independent exercise

Find the boundary: latency and throughput

  1. Create second case from blank file: Repeat under CPU/network throttling and 200% zoom.
  2. Then create controlled failure: Add long task, locate it in trace, remove cause, compare before/after.
  3. Use Chromium Performance · Memory · Accessibility · Security to prove behavior, then repair controlled failure.
  4. Compare result against supplied acceptance checks and reference approach before marking complete.
Concrete independent solution
  1. Duplicate the starter into a clean comparison case; change only this boundary: Repeat under CPU/network throttling and 200% zoom.
  2. Save its evidence beside the baseline and identify the first changed value. Capture profiles, accessibility tree, keyboard order, security headers, trust boundaries, budgets, and recovery evidence.
  3. Create the exact controlled failure: Add long task, locate it in trace, remove cause, compare before/after.
  4. Reproduce baseline exactly: Profile handbook load and one interaction; record Web Vitals and request waterfall.
  5. Write observation table with columns input, state transition, output, and failure for latency and throughput.
  6. Run boundary case unchanged: Repeat under CPU/network throttling and 200% zoom.
  7. Trigger controlled failure: Add long task, locate it in trace, remove cause, compare before/after.
  8. Compare evidence with reference outcome: Budget cites trace evidence and separates CPU, network, rendering, and measurement noise.
  9. Rerun baseline, boundary, and repaired failure together. Accept only if all reproduce: Budget cites trace evidence and separates CPU, network, rendering, and measurement noise.
07

COMPARE

Expected result

  • Budget cites trace evidence and separates CPU, network, rendering, and measurement noise.
  • Heading and button are keyboard reachable; activation updates announced output with a non-negative duration.
  • Controlled latency and throughput failure produces captured evidence; repair restores stated invariant without hiding error.
08

PROVE

Acceptance checks

Lesson is complete only when every check is true. Each check is stored locally and travels with your JSON backup.

0/5 complete · saved on this device

09

UNSTICK

Hints

Reveal hints
  1. Start with supplied normal case exactly as written: Profile handbook load and one interaction; record Web Vitals and request waterfall.
  2. For boundary case, change only named dimension: Repeat under CPU/network throttling and 200% zoom.
  3. If result is confusing, diff raw inputs and evidence before editing implementation.
  4. If tool shows nothing useful, move observation one boundary lower: representation, runtime, OS, or network.
10

VERIFY

Solution

Attempt both exercises before opening reference approach.

Reveal reference solution
  1. Run unmodified starter and preserve baseline evidence: Heading and button are keyboard reachable; activation updates announced output with a non-negative duration.
  2. Reproduce baseline exactly: Profile handbook load and one interaction; record Web Vitals and request waterfall.
  3. Write observation table with columns input, state transition, output, and failure for latency and throughput.
  4. Run boundary case unchanged: Repeat under CPU/network throttling and 200% zoom.
  5. Trigger controlled failure: Add long task, locate it in trace, remove cause, compare before/after.
  6. Compare evidence with reference outcome: Budget cites trace evidence and separates CPU, network, rendering, and measurement noise.
11

PREDICT · INSPECT · BREAK · DEBUG · MEASURE

Interrogate reality

Prediction: write expected output, state transition, ordering, and failure evidence before running either exercise.

Inspection: Use Performance and Memory profiles, Lighthouse carefully, accessibility tree, keyboard traversal, security headers, and controlled attack labs.

Measurement: Use budgets and distributions: Web Vitals, memory growth, attack surface, keyboard steps, contrast, error rates, and recovery time.

INSPECT

Capture raw evidence before explaining.

BREAK

Change one assumption and force controlled failure.

DEBUG

Find cause with Chromium Performance · Memory · Accessibility · Security before editing fix.

TOOL DRILL · keyboard only · record one retrievable command or shortcut
12

MASTERY + FRONTIER + BOUNDARY

Own the knowledge

TEACH

Explain latency and throughput at beginner, intermediate, and senior depth.

REBUILD

Recreate smallest useful example from blank file without notes or AI.

RETRIEVE

Schedule recall for day 1, 7, 30, and 90.

Creative frontier lab

Try first without opening the solutions. The constraints invite invention; the reference gives one concrete direction, never the only valid answer.

Constraint inversion

Re-solve latency and throughput by removing the most convenient abstraction. remove one risky or expensive capability, then measure what becomes simpler.

CONSTRAINTKeep the same inputs, observable result, and failure evidence; change the means, not the contract.

ORIGINAL IDEATurn subtraction into a design tool: the missing abstraction should reveal which responsibility it used to hide.

Reveal frontier solution
  1. Freeze the contract as three fixtures: Profile handbook load and one interaction; record Web Vitals and request waterfall. / Repeat under CPU/network throttling and 200% zoom. / Add long task, locate it in trace, remove cause, compare before/after.
  2. List every convenience used by the starter; remove the highest-level one while preserving Serve and open index.html; keyboard-activate the probe; inspect Performance, Accessibility, and Security panels..
  3. Implement the smallest replacement using remove one risky or expensive capability, then measure what becomes simpler.
  4. Run all fixtures and compare raw evidence. Keep the simpler version unless the removed abstraction has a demonstrated benefit.
Representation x-ray

Build an explanation artifact for latency and throughput: make the budget, trust boundary, ownership path, and semantic interface directly inspectable.

CONSTRAINTA peer must be able to locate the first divergence without reading implementation code.

ORIGINAL IDEATreat the explanation itself as a product: make invisible transitions visible, replayable, and diffable.

Reveal frontier solution
  1. Create one row or timestamped event for each transition in: Profile handbook load and one interaction; record Web Vitals and request waterfall.
  2. For every row record input, representation, owner, operation, output, and tool evidence from Chromium Performance · Memory · Accessibility · Security.
  3. Replay Repeat under CPU/network throttling and 200% zoom.; highlight only changed rows.
  4. Replay Add long task, locate it in trace, remove cause, compare before/after.; stop at the first divergent row and attach its recovery action.
Adversarial remix

Combine the boundary and failure into a new user-visible scenario for latency and throughput. build an adversarial fixture that tests slow, hostile, zoomed, keyboard-only, and failure states together.

CONSTRAINTDo not merely add more input. Invent a recovery interaction, alternate representation, or self-checking behavior.

ORIGINAL IDEAMake the system teach its own limits: the artifact should expose the invariant and offer a safe next action when it breaks.

Reveal frontier solution
  1. Combine these two pressures without changing them: Repeat under CPU/network throttling and 200% zoom. AND Add long task, locate it in trace, remove cause, compare before/after.
  2. Name the invariant that must survive and the user-visible evidence when it cannot: Budget cites trace evidence and separates CPU, network, rendering, and measurement noise.
  3. Implement this original direction: build an adversarial fixture that tests slow, hostile, zoomed, keyboard-only, and failure states together.
  4. Demonstrate baseline, combined failure, recovery, then baseline again; save the sequence as a regression fixture.

Capability frontier

Push latency and throughput until another layer becomes justified. Record one robust technique, one contextual trade-off, and one labeled hack or historical curiosity.

CORE · PRACTICAL · CONTEXTUAL · HACK · FRAGILE · HISTORICAL · GOLF

Boundary

Tools identify signals, not intent. Passing audits cannot replace threat modeling, user testing, or architecture.

If this vanished tomorrow…

Use manual inspection, timing, logs, semantic documents, least privilege, and adversarial reasoning.

Why next layer is earned

System design is earned when local correctness must survive scale, failure, shared state, and evolving requirements.