Page 6 of 8~104 min topic

Machine learning in plain English

Practise reliable habits: delivery times

Keep a holdout set, write the label guide, and test on ugly real photos first.

~13 min this pagePractise reliable habits — small checks before action

1Learn the idea

Read

Priya’s envelope of holdout leaves

See it

Machine learning loop
  1. 01Examples inLabeled data
  2. 02Pattern huntAdjust to fit
  3. 03Guess newUnseen input
  4. 04Mistakes teachMore signal

Teach with examples — not hand-written rules for every case

Habits beat annual policy PDFs. Priya Nair installs a small repeatable check for delivery times inside Riverside Garden Club: name the task, name a falsifier, name an owner, name a stop. Rehearse first on a lower-stakes cousin such as customer churn, then on pressure from the standing case (build a photo sorter for plant health without pretending the model understands gardens).

During a real interruption at Riverside Garden Club, Priya Nair stress-tests “Priya’s envelope of holdout leaves” on delivery times: one queued question, one hurried call, one hallway challenge. If the idea only works in a quiet workshop, it will not survive the standing case (build a photo sorter for plant health without pretending the model understands gardens).

Read

Label guide the volunteers can argue with

Time-box trust around delivery times. New uses at Riverside Garden Club get a probation window with hedged language and explicit metrics Priya Nair can chart. During probation, Priya Nair resists superlatives about delivery times even when customer churn looks safer by comparison. Hedging is operational, not shy.

Count something crude about delivery times—misses last week, minutes lost, or people affected—and write the number beside customer churn. Priya Nair needs that comparison before anyone at Riverside Garden Club declares victory on the standing case (build a photo sorter for plant health without pretending the model understands gardens).

Read

Habit: change one thing per experiment

Keep checks for delivery times tiny enough to survive a busy day at Riverside Garden Club: open a source, glance at a second opinion, confirm an irreversible action. Priya Nair logs overrides about delivery times in two lines; patterns become the vendor agenda. If overrides never occur while customer churn keeps producing tickets, audit a sample—rubber stamps hide in silence.

On “Habit: change one thing per experiment”, Priya Nair edits language about delivery times the way an editor would: strike “sentient,” “infallible,” and “just a tool” wherever they hide responsibility inside Riverside Garden Club. customer churn stays nearby as a plain-language control.

Read

When to stop and call an extension officer

Teach the delivery times habit to one other person this week while the standing case (build a photo sorter for plant health without pretending the model understands gardens) is still live. A habit that only lives in Priya Nair’s head is not yet part of Riverside Garden Club’s practice, and it should still mention when customer churn is the better rehearsal tool.

For “When to stop and call an extension officer”, a second person at Riverside Garden Club challenges Priya Nair’s note on delivery times and asks whether customer churn already solves most of the need with less mystery. That challenge is part of finishing the standing case (build a photo sorter for plant health without pretending the model understands gardens), not a delay tactic.

Go deeper

Before you start

Why this matters

Apply the pocket check—task, falsifier, owner, stop—to delivery times in under three minutes. Priya Nair then rehearses the same four fields on customer churn. Notice which field felt artificial; that friction often reveals a labelling problem at Riverside Garden Club.

Check your understanding

Page assessment

Answer from memory. Completion is saved from this evidence, not from opening the next page.

1. In Priya Nair’s scene, what bounded task does delivery times perform at Riverside Garden Club?
2. Which observation would most change your judgment about delivery times, and why?
3. How should customer churn alter the quality bar or the language you use?
4. Who can correct a miss before harm spreads, and what authority do they need?
5. How does this page advance the case: build a photo sorter for plant health without pretending the model understands gardens?

All responses are required.