TranscriptX SLOC estimate (2026-09-02)

Maintainer assessment under docs/reviews/. Dated snapshot, not a contract. Hosted at /guide/reviews/sloc-estimate-2026-09-02/ after make docs.

Measured with tokei 14.0.0, scc 4.0.0, cloc 2.08, and radon 6.0.1 via scripts/count_sloc.py (also bundled in the personal Cursor skill ~/.cursor/skills/count-sloc/). Git-tracked files only; 2,873 paths after dropping archive/, artifacts/, vendored/cache trees, and binary assets.

This is source lines of code (comments and blanks stripped). Markdown, JSON, YAML, and TOML are not treated as code in the headlines.

Headlines

Estimate = median of tokei / scc / cloc code columns. Range = min–max of those three. radon is Python-only and is a cross-check, not averaged in.

Estimate

SLOC

Range

Substantive product — app source under src/ + packages/ source (including frontend src); no tests, no generated build/

193,818

193,029–210,704

Maintained — product + tests + scripts/tools

360,257

359,047–382,290

Python-only product

192,982

192,196–209,868

radon Python product SLOC is 194,174 (within 1% of the trio median). tokei is the high outlier (it counts more docstring-adjacent lines as code). scc and cloc agree closely; the median follows them.

Use ~194k as the product figure and ~360k as the maintained figure. Quote the range if the number will be compared later.

Role summary (code languages only)

Role

tokei

scc

cloc

radon

estimate

range

product

210,704

193,818

193,029

194,174

193,818

193,029–210,704

tests

165,038

160,211

160,055

160,178

160,211

160,055–165,038

scripts

6,548

6,228

5,963

5,796

6,228

5,963–6,548

generated (frontend/build)

512

512

326

—

512

326–512

other (Dockerfiles, Makefile, CSS at repo root, …)

456

456

474

—

456

456–474

Tests are almost as large as product (~160k vs ~194k). That is a real maintenance surface, not an artifact of the counters.

Tracked JSON elsewhere is another ~150–230k physical/code-ish lines (fixtures, schemas, examples). It is not in the headlines.

Product breakdown

Python unless noted. Estimate = median of tokei / scc / cloc.

Slice

estimate

range

files

src/transcriptx/core

120,615

120,368–131,260

775

src/transcriptx/web

44,550

44,297–48,031

237

src/transcriptx/services

10,877

10,796–11,323

71

src/transcriptx/io

7,423

7,423–8,533

77

src/transcriptx/app

5,744

5,744–6,285

47

src/transcriptx/export

2,270

2,260–2,384

19

src/transcriptx/utils

1,386

1,242–1,894

6

src/transcriptx (package root)

23

23–55

2

packages/… Python

94

43–102

1

frontend src (TypeScript + CSS)

886

886

5

core is ~62% of product SLOC; web is ~23%. Frontend source is small; committed frontend/build (~512 lines) is generated and excluded from the product headline.

Calibration against older counts

Source

What it counts

Figure

This run

SLOC (comments/blanks stripped; no markdown/JSON)

product 193,818; core 120,615; web 44,550

scripts/log_code_size.py (same day)

raw physical lines in src + config + scripts, including markdown/JSON

255,570 lines / 1,294 files

stocktake 2026-07-17

approximate physical LOC

core/ ~94k; web/ ~28k

The July stocktake is low relative to current SLOC (core ~121k, web ~45k): the tree grew, and that note was a coarser physical-line sketch, not a comment-stripped count. log_code_size.py remains useful as a crude time series; do not use it as the substantive-code estimate.

Method (short)

  1. git ls-files, then drop archive/, artifacts/, node_modules/, dist/, *.egg-info, caches, and binary suffixes.

  2. Run tokei, scc, cloc, and radon on that list; bucket by role (product / tests / scripts / docs / generated / other) and by directory slice.

  3. Headline languages exclude markdown, JSON, YAML, TOML, XML, SVG, and similar.

  4. role=product is src/, packages/ source, lib/, app/. *.test.ts and tests/ are tests. frontend/build is generated.

Re-run:

python3 scripts/count_sloc.py --root .
# other repos:
python3 ~/.cursor/skills/count-sloc/scripts/count_sloc.py --root .

--install adds missing Homebrew tokei/scc/cloc and pip radon. --json PATH writes the machine-readable totals. Radon is a host tool, not a project dependency.