export.md

Your export, file by file

Settings → Download gives you a zip: a folder of plain files holding everything LoopUpward knows about you. It is built in your browser and goes nowhere else.

This page documents every file and every column. The same reference ships inside the zip as README.md, so the export explains itself even if you never come back here.

LoopUpward contains no AI and never calls one. What it does instead is record honestly — including the parts that did not go well — so that when you want to ask something of your own life, you have the material to ask with. The export is the deliverable; the rest of the app exists to make it worth reading.

Two rules that shape every file

1. Which day a row belongs to

Every file has a day (or planned_day) column holding a local date, YYYY-MM-DD. It is not always the calendar day its timestamp falls on.

Routines whose window wraps past midnight belong to the evening they started. A night routine set to 21:00 → 02:00 is one thing that happens on one night, so ticking its last step at 1:15am on the 9th writes day = 2026-08-08. The spill ends when that routine's own window ends.

Nothing else moves a day; every other row uses its calendar day.

Group by day for "how many days did I do this". Group by the timestamp for "what time of day do I do this". Both are in every file on purpose.

2. How timestamps are written

Each moment appears three ways, so you never do timezone arithmetic:

Column	Example	What it is
*_at	2026-08-18T06:12:44+05:30	ISO-8601 with the UTC offset in force at that moment
*_local_time	06:12	the wall clock you saw
*_weekday	Tuesday	the weekday it fell on

Durations are always minutes, one decimal, everywhere.

Empty means "not recorded", never zero. A blank mood is a day you did not rate. Nothing in the bundle is backfilled with a guess.

3. Which zone a timestamp was rendered in

focus_sessions.csv and events.ndjson recorded the UTC offset in force at the moment, so they are exact — travel and daylight saving included. Every other file is rendered in your own timezone from Settings, which is right except during a trip. manifest.json lists which files got which, under timezone.

4. What the record cannot know

The behavioural event log and the focus-session log started later than the rest of the app. manifest.json gives the exact date under observation_begins, and the bundle's own README repeats it.

Before that date:

• Zeros in daily.csv for app_opens, events and focus_sessions mean not observed, not "did nothing".
• What a habit's target was, what schedule it was on, and which steps a routine held cannot be reconstructed — so the files that need those say unknown in a *_source column and leave the verdict blank.

A blank verdict is deliberate. Raising a water habit from two glasses to three must not turn every past two-glass day into a failure, and a routine step added last week must not appear as six months of skipping. Where the export cannot establish what was being asked for at the time, it says so instead of using today's answer.

The files

Read first

File	What it is
README.md	this reference, written for someone who has never seen the app
manifest.json	coverage windows, timezone grades, and the vocabulary of every coded column — what the record can and cannot know
context.md	your timezone, when your day rolls over, and what each area of your life is for
summary.md	the computed numbers, formatted to read

Whole life, or one period

The download can be narrowed to the current week, the current month, or any range you pick. A scoped bundle has the same files and the same columns as a whole-life one, so two of them can be compared or concatenated. Two things are added so a slice reads on its own:

• items.csv, areas.csv and labels.csv also carry rows from outside the window whenever something inside it points at them, marked in_window = false. No id in the bundle is left dangling.
• items.csv carries opening_value, what each tracker read the moment the window opened, without which a counter that moved 3 to 7 inside the window cannot be told from one that started at zero.

manifest.json records the window under window.

The three kinds of file

• Record — what happened, at the time it happened: logs.csv, actions.csv, focus_sessions.csv, events.ndjson, habit_days.csv, routine_steps.csv, seeds.csv, list_entries.csv, day_order.csv, and the journal / reflection / note files.
• State as of export — how things look today: items.csv, areas.csv, labels.csv, context.md. A goal reading "done" here says nothing about the journey; that is in the Record files.
• Derived — plain counting over the Record, and recomputable: schedule_expectations.csv, daily.csv, summary.md.

schedule_expectations.csv — what was supposed to happen

Everything else records what you did, and a missed day is not a row anywhere — it is the absence of one. This file supplies the other half: one row per scheduled item per day it was alive, saying whether that day was required (a fixed schedule named it), not_due, eligible (a quota schedule like "4× a week", where the period is owed rather than the day), or unknown.

Adherence becomes a join rather than a reconstruction: filter expectation = required and take the share with met = true. Retiring a habit ends its expectations that day, so the months after you deliberately stopped are not counted against you.

areas.csv

area_id · name · emoji · color · position · description · why_it_matters · target_share (0–1) · created_at / created_local_time / created_weekday.

items.csv

Every node, including finished, archived and trashed ones.

Column	Meaning
item_id, title	identity
kind	note, folder, quote, idea, dream, goal, habit, routine, list, project, book, milestone, principle, promise, lesson, memory
tracker	none, check, counter, percent, money, habit, book
status	active, done, someday, archived
area_id, area_name	resolved up the parent chain, exactly as the app reads it
parent_id, parent_title	what it nests inside
horizon	someday, life, year, quarter, month, week, today, date
horizon_period	which instance: 2026-W28, 2026-08, 2026-Q3, 2026, or the pinned day
date_repeats_yearly	a birthday rather than a one-off
target, current, unit, progress_fraction	the meter
cadence, cadence_days, cadence_count	the schedule. Weekday numbers are 0 = Sunday
window_start, window_end	routines only. End before start = wraps past midnight
pulled_today, pinned, labels, note	the rest of its state
step_count, planned_minutes, entry_count	routine and list contents
created_*, completed_*, deleted_*	three timestamp groups
age_days, days_to_complete	computed for you

actions.csv

Every concrete thing planned for a day. planned_day is what it was meant for; done_at is when it was actually ticked; days_late is the gap, and negative means early.

Moves between days are not visible here — this file only has the final date. Every move is an action.rescheduled event, and the totals are in summary.md.

amount is how much finishing it moves the linked item's counter. 0 is a deliberate choice: a directional step that belongs to a goal without pretending to move its meter.

logs.csv

Every unit of progress ever recorded. Item counters are derived from these, not the other way round.

op is add (a delta — a habit day, +1 chapter) or set (a snapshot — money now reads ₹40,000). Sum add rows; take the last set row. Mixing them gives nonsense.

source says how the value got there:

Value	Means
today_check	ticked on the Today list
item_page	changed on the item's own page
focus_timer	recorded inside a focus session
routine_run	written by a routine's script
parent_cascade	a child finished, so this parent's meter moved. via names the child
manual	typed in somewhere else
import	carried in from another device
unknown	written before provenance existed — not a bug, and not a claim it was manual

focus_sessions.csv

One row per timer attempt, written when it ended for any reason. Abandoned attempts sit here on the same footing as completed ones; that is the point of the file.

kind is focus, routine_step, routine_run, day_run or rest. outcome is completed, expired (countdown ran out and the screen was left), abandoned (closed mid-session), skipped or interrupted (swapped for something else).

planned_minutes blank means it was untimed and counted up. actual_minutes is wall clock with pauses excluded. over_under_ratio is actual ÷ planned — above 1 means it took longer than you thought.

A routine produces one routine_run row plus one routine_step row per step. Do not add both, or you double-count the time.

habit_days.csv

What a habit or routine meant on one specific day. "Clean" is the habit; "clean the side desk" is what it meant on Tuesday. Carries day_plan, whether the day was logged, value_logged against daily_target (2 of 3 glasses), and steps_done / steps_total for routines.

routine_steps.csv

One row per routine, per day, per step. position is where the step sits in the written script; tick_order is the order you actually did them, which is often different. done_at is blank on days recorded before step timestamps existed, even where done is true.

daily.csv

One row per day, everything paired and nothing interpreted: journal presence, mood, energy, sleep hours and quality, stress, focus rating, tasks planned and completed, habit and routine days, focus minutes planned versus actual, abandoned sessions, pauses, app opens, first open, last write, distinct routes, and total events.

The app deliberately does not compute correlations for you. It emits honest columns; the question is yours.

There is no usage table: the app-open columns are derived from app.opened and page.viewed events, so there is one record of what happened rather than a counter to reconcile.

list_entries.csv, seeds.csv, labels.csv, day_order.csv

• list_entries.csv — lines inside list items. picked_at is when a line became "the one I'm on"; days_being_tried counts from there.
• seeds.csv — raw captures. days_resting is how long a thought sat before it grew into something.
• labels.csv — your tags, with items_tagged counted for you.
• day_order.csv — the manual order you dragged a day into, one row per position. entry_kind separates a real action from a scheduled habit row or a longer-horizon item pulled onto today.

events.ndjson

The full behavioural log: one JSON object per line, oldest first. Read it with jq, or pandas.read_json(lines=True).

Every line carries id, type, at, at_ms, local_time, weekday, day, timezone, utc_offset_minutes, item_id, item_title and a payload.

Notable types:

type	What it records
item.created	the birth snapshot: kind, tracker, horizon, area, parent, target, cadence, labels
item.horizon_changed	re-scoping — how a goal drifts from "this quarter" to "someday"
item.completed	with ageDays and openActionsRemaining
action.rescheduled	the procrastination signal — from, to, daysMoved
routine.step_skipped	sentToBack: true = "skip for now"; false = an optional step left out
day_run.abandoned	with completedBefore — how far it got
day.reordered	via: "drag" or via: "sort" — different acts, recorded differently
app.opened, page.viewed	being here at all. Routes are recorded in shape (/item/[id]), never which note you had open

Events are append-only. Nothing is edited or deleted, including the records of things that did not work out.

Human text

• journal.md — every daily entry in order, with ratings and your words.
• reflections.md — every period reflection, and whether its intentions were met. Scoring is plain counting against the named items: days logged for a habit, units added for a meter, done-or-not for a one-off. An intention naming nothing countable is shown as not scored.
• notes.md — note bodies, in the app's own markdown.

raw.json

Every table exactly as stored, plus both streams and your settings. If a CSV lost some nuance to being a table, it is here.

Questions this bundle can answer

• What time of day do I really work, versus when I think I do?
• How much longer do things take than I plan for?
• What do I keep pushing, and by how many days in total?
• Which routine step do I quit on?
• Am I creating things faster than I finish them?
• Which weekday breaks my streaks?
• Does sleep track with what I get done?

Signed in or not

The bundle is identical either way. Signed in, the rows come from your private cloud; signed out, they come from this browser's own storage. The same generator builds the same files, so keeping your data local never costs you a lesser export.