lacheisis

Lachesis measures each thread and decides how much is allotted, Atropos cuts it.

A zero-dependency Windows productivity tracker in Go. It samples your foreground window every few seconds, measures where your time actually goes, and scores the day by yield — outcomes per focused hour — never volume. Volume metrics get farmed (Goodhart); ratios and artifacts don't.

Usage

go build -o lacheisis.exe .

lacheisis track                    # run the collector (detach with Start-Process)
lacheisis report [YYYY-MM-DD]      # score, focus blocks, commit yield, per-app thread
lacheisis allot chrome 2h          # give chrome 2h/day; report shows overruns
lacheisis classify title youtube distracting   # add a classification rule

Run it in the background:

Start-Process lacheisis.exe -ArgumentList track -WindowStyle Hidden

Example report:

Lacheisis — 2026-07-12
Score 74 — a well-measured day
  focus 0.72 · depth 0.61 · cohesion 0.88 · yield 0.67
Thread measured: 3h42m active · 25m idle
  [+] productive 2h40m · [·] neutral 45m · [!] distracting 17m
  3 commits · 11 files touched · 1.4 commits per deep hour

  [+] code                     2h10m  ████████████████████
  [·] chrome                   1h05m  ██████████  (54% of 2h allotted)
  [+] windowsterminal            27m  ████

Deep work — 2 block(s), 1h55m total, longest 1h20m, 4 switches/h:
  09:12–10:32  code             1h20m  ● a1b2c3f fix sessionizer (lacheisis)
  11:04–11:39  code               35m

✂ Atropos: slack exceeded its allotment (1h30m / 1h)

Scoring (v0.2): attendance is farmable, yield is not

Foreground hours can be farmed — leave an editor focused and look busy. So the 0–100 score is built only from ratios and outcomes:

Component	Definition	Weight
Focus	productive time / active time	35
Depth	deep-work time (≥25 min productive blocks) / productive time	25
Cohesion	1 − min(1, context switches per active hour / 30)	15
Yield	commit-linked deep-work hours / deep-work hours	25

• Rules classify every sample productive | neutral | distracting; title rules run before app rules (chrome on YouTube is distracting even if chrome is neutral). Manage with lacheisis classify.
• Focus blocks sessionize the sample stream: alt-tab glances (≤2 samples) don't break a block, idle always does.
• Yield scans the configured "repos" (read-only git log) and attributes each commit to the deep-work block containing it, or the nearest one that ended ≤30 min before.
• Anti-gaming: farmed activity loops one window title. Deep blocks with near-zero title entropy are flagged ⚠ looping?, and no metric anywhere rewards raw volume.

How it works

• Win32 GetForegroundWindow → process image name, sampled every 5s
• GetLastInputInfo marks you idle after 60s without input; locked screen counts as away
• Samples land in append-only JSONL at %LOCALAPPDATA%\lacheisis\data\, one file per day
• Config (poll interval, idle threshold, allotments, rules, repos, score weights) at %LOCALAPPDATA%\lacheisis\config.json