git blame for your promises.
You have made a thousand estimates. You have checked none of them.

_{GIT HISTORY · TERMINAL · ZERO DEPENDENCIES · NO DATABASE · NO ACCOUNT · NOTHING LEAVES THE MACHINE}

_{0xfc7e0cddb9d7a1676edefbb4d87dda29efe2053f
swap · chart · explorer · fibo.fund · burns every 9 hours}

start · said / took · the multiplier · refusals · on disk · the token · honest limits

say · writing a promise down

one task · taken apart

doctor · through the monocle

one task · said, first commit, merge, span, active, the longest pause	drift · quarter by quarter
doctor · four estimates that disagree, fourteen written late, the chain intact	refusals · five ways to be told no
day one · yours starts empty	--lang ru · it speaks Russian too

_{Every picture above is a real transcript: the demo repository, the real commands, drawn by docs/shots.py.}

pipx install git+https://github.com/0xkuch/Fibo
fibo demo          # two years of made-up history, same numbers on every machine
fibo               # inside your own repository

  (o,o)   fibo · /tmp/fibo-demo
  /)_)
   ""

230 tasks · 225 closed · 194 estimated · 169 counted

  said "2 days"   — took 9.6 days  ×4.8
  said "a day"    — took 2.7 days  ×2.7
  said "4 hours"  — took 14 hours  ×3.4

  your multiplier    3.7×   calendar 5.2×
  worst case (p90)   8.6×
  underestimated     91% of tasks

  by kind of work
    bug fixes      ×1.3    n=42
    features       ×3.7    n=66
    refactoring    ×5.7    n=39
    migrations     ×7.1    n=20

  drift   3.7× ▄▃▃▃▄▃█▆▁▁▁▆ 4.5×
          two years ago      now

excluded: 25 tasks (estimate after start: 14 · not a duration: 5 · no branch: 4 · points, not time: 2)
one subject: you — you@fibo.demo
all of it computed from your history, none of it written here

Yours starts empty. It speaks Russian too (--lang ru).

Your tracker already holds years of estimates (Jira, Linear, GitHub, GitLab), so the first run has history to work with. Without a tracker, put Estimate: 2d in the first commit of a branch. Without either, fibo say PROJ-412 2d writes it down before you start. Where estimates come from →

An estimate nobody checks is not an estimate. It is a mood.

This number is the whole product. It does not tell you to estimate better. It tells you what to multiply by.

• Median in log space. ×4 and ×¼ are the same distance from ×1, and the median does not care about the one ×40 disaster. The geometric mean is in -v.
• Working hours. Mon–Fri, 9–18, in the offset your own commits carried. A day is 8h, a week is 40h. --calendar takes both sides literally, shown small.
• span and active. span: first commit → merge. active clips any pause over three working days. The gap is how long tasks sat waiting.
• p90 is the bad day: "in the worst case, ×8.6".
• By kind of work: tracker labels, then where the diff went (migrations/, tests/), then the words in the title. Five tasks to be shown.
• Drift: a 90-day rolling median, indexed by when the work landed — indexed by when it began, the last window would fake an improvement.

fibo PROJ-412 takes one task apart. fibo drift shows it quarter by quarter. fibo doctor checks that nobody nudged anything.

fibo say PROJ-412 soon          ✗  not a duration
fibo say PROJ-412 2d            ✗  written after the work began
a task with no estimate         ✗  not counted, and not guessed
--whose someone@else.com        ✗  this tool has one subject
story points, no rate given     ✗  points are not time

Not a duration. "Soon", "by the end of the sprint", "quick" are not estimates. Say a number and a unit.

Written after the work began. The program knows when the field changed and when the first commit was made. The value that stood at the first commit counts. doctor lists the rest.

Not guessed. A task without an estimate gets nothing by analogy, by average or by diff size. It is left out, and the output says so.

One subject. Yours. It will not print another person's multiplier, aggregate a team, or export. --whose with anyone else's address fails loudly. Trackers are asked for your issues only. There is no JSON, no repository-wide mode, no all-authors mode.

Points are not time. --points-as-hours=N exists, and the output then says the number rests on your assumption.

A mirror pointed at someone else is a different object with a different name.

.fibo.json            optional, yours: who you are, which tracker to ask
.fibo/ledger.jsonl    only if you use `fibo say`; every line hashes the one before
.fibo/ledger.head     the hash of the last line
.fibo/cache/          tracker responses with a TTL; delete it any time

That is all of it. No database, no index, no daemon, no server, no account, no telemetry. Every run recomputes from history. With a tracker it only reads — GET and GraphQL queries; a mutation is refused before it is sent. The config loader reads one file and never follows a path to a key.

A multiplier you can quietly edit after a bad quarter is not a multiplier.

$FIB on Robinhood Chain — a community token. The code does not know it exists: there is no import, no flag, no line in the ledger. fibo will never require it and will never print its price.

fibo say $FIB 100x              ✗  not a duration
--whose the-community           ✗  one subject
fibo predict $FIB               ✗  it predicts nothing

_{not financial advice · the owl is not a financial advisor · it is not even a real owl}

No dates. This program exists because durations written in advance are wrong by a known factor, and it would have to multiply its own. So: order, not calendar.

00  the tool                         ✓
01  the multiplier                   ✓
02  the breakdown by kind of work    ✓
03  the drift                        ✓
04  more sources                     YouTrack, Azure Boards, Shortcut, …

IT CANNOT SEE THINKING. The clock starts at the first commit, not at the first thought. Your real multiplier is higher than the one printed.

IT CANNOT SEE INTERRUPTION. A branch that sat for a week looks like a week of work. active is there for that, and it is still an approximation.

THE DEMO IS A DEMO. Reproducible, and nobody's record of promises. Yours starts empty.

IT DOES NOT KNOW THE RIGHT ANSWER. There is no correct estimate. It knows what you said and how long it took.

IT PREDICTS NOTHING. No model, no opinion. What was written down, what happened, one division.

A squash merge erases the start of the work: the tracker's "In Progress" or a leftover branch fills it in, and without either the task is excluded as "no branch". Time zones come from your commits' offsets, so a laptop set to the wrong zone gets a wrong day.

python -m unittest        # 227 tests, stdlib only
python docs/shots.py      # redraws every picture above, from the real output

_{MIT · built for the fun of it · (-,-) the owl has seen enough}