Skip to lesson
中文
The collection/AN OPEN STUDY ROUTE

WEEK 23 / 24 · Keep what matters

What is worth keeping

Audit value: calculate net time saved and let unused work go

10 core hours + 2 optionalREAD · BUILD · PRACTISE
Begin the lesson ↓
What is worth keepingTHE PERSONAL FDE FIELD LIBRARY

The question: Does the tool make life easier, or replace the original work with supervising models and maintaining software?

Before you begin: Have actual use and maintenance records and distinguish measured, estimated, unknown and synthetic results.

FIG. 23.1THIS WEEK’S REASONING PATH
  1. 01Comparable baseline
  2. 02Count all effort
  3. 03Calculate net value
  4. 04Keep or simplify
Teaching diagram · Derived from this week’s lab; a suggested process, not observed results.

Concepts and learning objectives

Compare equivalent tasks at similar quality. Include your operation, supervision, review, rework and maintenance, including failed attempts. Background model runtime during which you can do other things is not automatically human labour. Record one-time building, cash expense and ongoing operation separately.

Time is not the only value: fewer omissions, traceability and independent skill matter too, but report them separately. An experiment may have negative financial or time returns while teaching something useful, provided it is labelled learning investment. Stopping unused work is an engineering decision.

Reading and selected sections

Priority Source and selected sections Budget
Required review R17 · Harvard CS50 SQL — Aggregate comparable tasks and inspect denominators and missing data. 30 min
Optional review R67 · Google SRE — Monitoring Distributed Systems — Assess user outcomes rather than busy infrastructure. 30 min
Optional review R74 · GOV.UK — Plan user research — Return to the original problem and check the need still exists. 30 min

Studio lab

  1. For P1, P2 and P3, list actual uses, outcomes, active time, review and maintenance. Say “no record” when absent; do not infer human savings from model runtime.
  2. Calculate weekly net time saved: comparable successful tasks' old active time minus all new operation, review, failed rework and maintenance. List unfinished demand separately.
  3. Separate build time, incremental subscription costs, API and hosting. State allocation assumptions for subscriptions already used elsewhere and avoid double-counting.
  4. Keep, simplify, pause or delete. Remove unused features from the main flow. A negative-saving project may continue as bounded learning, with a stated reason.
  5. Give saved time a concrete purpose: focused study, exercise or rest. Avoid pouring every spare hour into unlimited model comparisons.

Deliverables and acceptance

Deliverable Evidence of completion
Honest value report Measurements and assumptions separated; failures and maintenance included.
Retention decision At least one feature reviewed for necessity, with reasons to keep or remove it.
Next-month use plan Natural triggers and maintenance limits, not only a feature roadmap.

Independent exercise: Calculate net time on synthetic records containing failed runs and maintenance. Explain why two minutes of generation is not a measure of saved labour.

Privacy and safety: Timestamps and task categories reveal routines. Publish reviewed aggregates or labelled synthetic examples with methods and sample scope, not a private raw ledger.

Everyday use: Revisit week one's friction. Stop valueless reminders, jobs, connectors and cloud resources and revoke permissions no longer needed.

Pass or revisit: Make a real trade-off supported by an honest report. Negative savings are not a result to conceal.

A reusable collaboration prompt

Audit actual records, including failure, review and maintenance. Separate building and cash costs. Do not convert unknowns into precise benefits. Give evidence for keeping, simplifying or pausing each project without assuming more features are needed.
FIG. 23.2SYNTHETIC EXAMPLE

72 − 18 − 14 = 40 minutes

Old active time72
Operation & review18
Maintenance14
Net saved40
Source: the manuscript’s net-time example. Twelve equivalent tasks at six minutes each; excludes cash, one-time building and future faults. This is not a benefit forecast.

Sources and verification scope

Reference IDs and scope follow the supplied manuscript, baseline 2026-09-27. External authenticated labs were not all completed; recheck versions and access before implementation.

  1. R17
    Harvard CS50 SQL ↗

    Querying, relationships, design and writes are the main selections.

  2. R67
    Google SRE — Monitoring Distributed Systems ↗

    Select monitoring practices suitable for small personal services.

  3. R74
    GOV.UK — Plan user research ↗

    Problem-discovery methods; not authorisation to use an organisation's processes.