CCM: Code, Change, and Confidence

A book about the things we can count, the changes we must understand, and the evidence we need before accepting them.

We begin with a source tree and end with a defensible explanation of a change. The running checkout example keeps the algorithms, mathematical claims, and economic assumptions attached to an executable result. We can follow the story from the beginning or use the contents to return to a particular argument.

Opening

• What a number can promise

  • How the argument develops

  • Three readers, one account

Seeing a change

• A first measurement

  • The inventory before the estimate

  • The short estimate

  • Formats belong to their readers

• Reading the source tree

  • Language names and policies

  • What the totals preserve

  • Inventory as an audit trail

• From two snapshots to a task

  • What comparison actually does

  • The model’s edge direction

  • A task is more than a changed set

• A checkout change, calculated

  • Running the complete example

  • Shared and joint work

  • Behavior and evidence

  • Pricing the declared activities

Inside the machine

• The shape of the machine

  • The order of one invocation

  • Few dependencies, explicit tools

• Selecting work without losing control

  • Ignore rules have scope

  • The walk and the queue

  • Results have no scheduling order

• Counting languages and mixed documents

  • A declarative catalog

  • One physical line, several languages

  • Typst is a document language

  • Work tokens are not API tokens

The finite calculus

• The mathematics of affected parts

  • Three closure laws

  • Breadth-first evaluation

  • What the closure does not contain

• When checks become necessary

  • Why positive activation matters

  • Alternatives and conjunctions

  • From necessity to completion

• Evidence that can survive a change

  • Admissibility and conclusion

  • Residual demand

  • A failure counterexample

  • What reuse can justify

• Shared work and additional joint work

  • A proof by contribution

  • The checkout calculation

  • What the CLI compares

• The outcomes evidence leaves possible

  • Complementary evidence

  • An admitted failure can refine cases

  • The relationship to entropy

Prices and experiments

• From quantities to an economic scenario

  • A source-based AI workload

  • Residual checks and hosted tools

  • Human work and an optional repeat scenario

  • Overrides and interpretation

• What the 2026 research profile contains

  • Spending anchors

  • Language effects

  • Prices and time

  • Sensitivity and calibration

• Experiments we can reproduce

  • Startup is part of the question

  • Three workloads with CCM 0.6.0

  • Where the improvement came from

  • Correctness is another experiment

• Learning from real development work

  • A useful resource ledger

  • Separate the activities before fitting

  • Prediction error before a confidence story

  • Testing an AI effect

  • Revision as a scientific operation

Research and publication

• Where the calculus can grow

  • Description length and change description

  • A geometry before a topological interpretation

  • Observers and composition

  • Path observations and distributions

  • Typed graphs before graph spectra

  • The gate back into the runtime

• A report, a book, and a visual identity

  • One logo, several media

  • The publication pipeline

  • Screenshots as observations

  • The last design rule

• Contracts and a reading desk

  • The supported surfaces

  • Useful invocations

  • Limits and evidence

  • A small glossary

  • Research and source references