Export gaps are memory gaps: why the missing parts distort the picture

Two incomplete records are being joined every time you open your archive: the
platform's file (sampled, capped, format-limited) and your own recollection
(old, edited, loyal to the present you). Psychology has spent decades
documenting how the second one fails; the interesting problem for anyone
reading an export is that you cannot see either failure from inside.

The memory side: how the reader distorts

• Choice-supportive bias. People preferentially remember the positive

  features of options they chose and the negative features of options they
  rejected - then repeat the asymmetry when recalling whole chapters of
  life.¹ Two years after a miserable job you remember the commute and
  forget the salary; two years after leaving an account you remember the
  friends and forget the argument that closed it. The bias runs in the
  direction of your current self, which is why it is invisible to your
  current self.

• Rosy retrospection and fading affect. On average, the emotional tone of

  past events drifts positive as time separates you from them - the bad
  weeks get their sting edited out first.² Nostalgia research adds a
  second step: recalled periods acquire a warmth that the record of those
  periods often contradicts - mundane posts from a "golden year" are
  frequently mundane.³

• Mood-congruent recall. What you are feeling now biases which material

  comes to mind: a rough week reaches for proof that things were always
  rough; a good one reaches for the highlight reel.⁴ So the same
  archive yields different "true stories" depending on the day you open it.

Each bias is well supported and, crucially, each operates on retrieval -
which parts you reach for - rather than on the file. The file does not change
when your mood does. That asymmetry is the only thing you can really lean on.

The file side: where the record genuinely stops

The archive's holes are structural, and this product documents each one rather
than filling it with guesses:

• Sampled lists. Search history is capped at your most recent 100-500

  entries - a floor, never a lifetime total.⁵

• Undated rows. Rows without parseable timestamps count in totals and

  leave every time-based view, because unknown time is not zero time.⁶

• Format limits. HTML exports collapse structured splits (the four search

  kinds, for instance) at the source - fewer fields, not hidden fields.⁷

• What no export contains. The eight-things post enumerates what is gone

  before the file is even generated - and the questions post does the same
  for answers that were never recordable, feelings and intentions
  chief among them.⁸

The honest response to a gap is to name it in the view: counts show their
windows, comparison views state their caps, and unknown stays visibly unknown
instead of being smoothed to a plausible number.

Reading a partial record through a biased reader

The practical discipline is to keep the layers separate:

1. Let the file win arguments about fact. "I only ever posted twice that

   year" versus the archive's count: the archive is evidence, the recollection
   is testimony, and testimony is the softer of the two - especially when
   choice-supportive revision is working in the background.¹

2. Suspect your favourites. The periods you are sure were wonderful are

   exactly where rosy retrospection and choice-supportive bias concentrate.
   Open the posts from that year before you settle on the story of it.

3. Notice which chapters you reach for. Mood-congruent recall means your

   instinct to re-read 2021 today tells you about today. Read 2019 too - the
   contrast is where the interesting information is.⁴

4. Mark gaps instead of filling them. When the record stops (sample cap,

   undated rows, deleted media), write down that it stopped. The mind
   abhors a vacuum and will supply a plausible invention if you let it; a
   written "unknown" blocks that move.

The complement: what memory does better than the file

Fair is fair: the archive is worse than you are at some things. It cannot tell
you what an afternoon felt like, why a caption mattered, or which of five posts
was actually the joke. The file holds rows; you hold meaning. The failure mode
this post warns against is not using memory - it is letting memory edit the
rows: the moment recalled history contradicts recorded history and you decide
the record is wrong, because the memory feels so vivid. It is supposed to.
Vividness is a property of remembering, not of truth.

Related reading

The eight-things post for the full list of what the export cannot recover; the
questions post for categories of answers no archive can hold; the
missing-half-your-life post for the structural shape of what gets sampled away
before the file exists.

¹: Mather, M. and Shafir, E. - choice-supportive bias: chosen options
accumulate remembered positives, rejected options remembered negatives; cited
as published memory research.
²: Ross, M. and Wilson, A. - time and self-esteem work on rosy
retrospection and shifting attribution of past outcomes; cited as framing.
³: Wildschut et al. and the nostalgia literature - recalled
periods take on warmth that contemporaneous records often lack; cited as
framing.
⁴: Bower, G. - mood-congruent memory research; cited as framing for
retrieval bias.
⁵: packages/shared/src/analytics/compare.ts - the in-product
100-500 search sample caveat.
⁶: packages/shared/src/analytics/activity-taxonomy.ts - raw counts
never timestamp-filtered; dated counts tracked separately.
⁷: packages/shared/src/parsers/html-to-json.ts - HTML exports lose
structured splits at the source.
⁸: content/blogs/2026-10-07-eight-things-no-export-can-bring-back.md
and content/blogs/2026-10-07-questions-your-instagram-export-will-never-answer.md

• the two standing lists of structural absence.