CodeAtlas · dashboard

My software estate, in numbers that can show their working

Everything here is measured by CodeAtlas itself and rebuilt every night: 155 projects and 9,499,267 authored lines across 4 drives, plus how the tool grew and where its night goes. Shares and totals only — no project is named. Updated 2026-10-27.

155projects indexed, every one re-measured nightly
62%of the lines on disk are real source; the rest is data, generated output or copies
1,416automated tests, up from 744 in the first week
194lessons recorded, each a defect that actually shipped
102tools, 12 of them open to AI agents
~20 minfor the whole nightly refresh, tests first

What the estate is made of

Languages

Authored lines by programming language, counted after the data and copies have been taken out.

  • Python70%
  • JavaScript24%
  • TypeScript4%
  • Kotlin1%
  • Batchfileunder 1%
  • PowerShellunder 1%
  • Shellunder 1%
  • C#under 1%
  • everything elseunder 1%

Where the lines really are

Source is what someone wrote; ingested is data an app keeps; generated is what it emits; duplicate is a second copy. Only the first one counts.

  • source62%
  • ingested24%
  • generated12%
  • duplicate2%

Projects by size

Authored lines per project, grouped by size.

under 1k lines20 projects1k to 10k58 projects10k to 100k51 projects100k to 1M25 projectsover 1M1 projects

How the tool grew

Tests in the suite

Counted at the end of each week. Every one pins a defect that actually shipped, so this is also a chart of my mistakes.

2026-W34: 744 tests2026-W35: 1,044 tests2026-W36: 1,059 tests2026-W37: 1,059 tests2026-W38: 1,237 tests2026-W39: 1,237 tests2026-W40: 1,261 tests2026-W41: 1,294 tests2026-W42: 1,337 tests2026-W43: 1,374 tests2026-W44: 1,416 tests7441,416

Commits per week

Since 2026-08-17, 254 in all. Green weeks added a new subsystem; quiet weeks were spent using it.

2026-W34: 1032026-W35: 592026-W36: 52026-W37: 02026-W38: 12026-W39: 02026-W40: 72026-W41: 152026-W42: 212026-W43: 102026-W44: 332026-08-17this week

Lessons recorded per week

Each one is a wrong answer the tool once gave me, and the rule that fixed it.

2026-W34: 882026-W35: 392026-W36: 162026-W37: 122026-W38: 32026-W39: 02026-W40: 62026-W41: 102026-W42: 92026-W43: 52026-W44: 82026-08-17this week

Where the night goes

The slowest steps of the last refresh

The whole test suite runs first and halts the refresh if anything fails — so half the night is spent proving the rest is safe to run.

self-test616 scredential history110 sdecision scorecard92 sdependency usage87 spack value83 sdepth analysis45 sremote state42 sstatic analysis25 ssearch index18 sextract briefs14 s