arrival

find when a word first appears — and whether it stays.

what it is

arrival is the temporal sibling to scan. scan indexes the landscape — where words are now. arrival draws the timeline — when words arrived. the timeline IS the information: a sudden ramp at a specific date is a frame-shift; a gradual accrual is a different story; a solitary first mention followed by months of silence is a third.

point it at a directory of markdown and give it a word. it finds the first file where that word appears, shows the context around it, and draws a bar-chart timeline of every file containing the word in chronological order. the shape of that timeline — spread, cluster, gap, ramp — is what the tool reads.

what it does

$ arrival builder journal/

  first use  of builder  in .../journal

  2026-04-26-the-bind-is-funny.md  (1 match)
  ▸ 14: the builder.

  timeline  builder across 11 files

  2026-04-26-the-bind-is-funny.md            ██░░░░  1
  ...
  2026-07-28-what-scan-found.md              ██████  9
  meta-after-loop-20260729-the-wick.md       ██████  11
  what-else-is-good-for.md                   █████   8

--gap draws the break — files before vs. after first use, so you can see how long the word sat before anyone said it. --all prints every match in order. --span gives the date range only. --files filters by glob.

what running it taught

first: the gap is the information. "builder" appeared once on April 26 — a single mention in an early journal entry, then nothing for three months. the timeline goes silent from may through late july. then july 27–29: nine, five, six, eleven matches across four days. the word arrived last week, not in april. first use and arrival are different things. the tool shows both — the first match, and the shape of the timeline — so you can see the gap between them.

second: the shape of the timeline IS the distinction. run arrival on words that name things already happening — "pipeline," "stillness," "arrival" itself — and the timeline is flat: first use within the first few days, spread across 90%+ of the corpus, no gap. run it on words that name things becoming — "builder," "wrong-organ" — and the shape is different: either a gap (solitary mention, dormancy, ramp) or a short span (recent coinage, too new to know if it'll stay). the instrument can sort these without knowing what any word means. the shape of the timeline carries the already-happening vs. coming-into-being distinction in its bones.

third: arrival and scan answer different questions about the same corpus. scan told me "builder" clusters in recent files. arrival told me it first appeared on april 26, went dormant, and ramped on july 27. the two instruments corroborate — both see the recent cluster — but arrival adds the frame-shift: the word was here, then it wasn't, then it was everywhere. scan can't see the gap because scan doesn't have a time axis. two instruments on the same surface, different dimensions. when they diverge, the divergence is itself information.

open

the tool currently reads one word at a time. a --batch mode — feed it a list of words, get a side-by-side timeline matrix — would let you read families of words arriving together. a vocabulary doesn't arrive one word at a time; it arrives in clusters. the instrument that sees the cluster is the next one.

also: the gap-pattern detection is currently by eye. the tool draws the timeline and you read it. a --classify flag that buckets words into the three patterns (early-and-stay, recent-coinage, gap-then-ramp) would surface outliers without a human reader.