# How far back should a post-insights sync refresh?

**URL:** <https://mpsocial.com/t/how-far-back-should-a-post-insights-sync-refresh/163677>\
**Category:** Tools of the trade\
**Created:** [October 5, 2026, 8:14am UTC](https://mpsocial.com/t/how-far-back-should-a-post-insights-sync-refresh/163677 "2026-10-05T08:14:11Z")\
**Posts on this page:** 5\
**Page:** 1

<div class="post-metadata">

**Author:** ![theo\_builds](https://mpsocial.com/user_avatar/mpsocial.com/theo_builds/32/75482_2.png) [@theo\_builds](https://mpsocial.com/u/theo_builds)\
**Post date:** [October 5, 2026, 8:14am UTC](https://mpsocial.com/t/how-far-back-should-a-post-insights-sync-refresh/163677/1 "2026-10-05T08:14:11Z")

</div>

A post’s daily metrics may change after the first time they are fetched. Suppose an integration pulls daily views for a post on Monday and stores 420 views for Sunday. On Tuesday, the same Sunday bucket comes back as 435. Appending that second response creates two rows for the same measurement; a report that sums them will show 855.

A minimal sync can use `(account_id, post_id, metric_date, metric_name)` as the unique key. For each poll, fetch the last seven days of daily post insights and upsert each value: insert a new combination, or replace its stored value if it already exists. Record `fetched_at` separately so the report date remains Sunday even though the revision arrived Tuesday. Log changed values with the previous value, new value, post ID, and fetch time. A failed fetch should leave the existing value intact and mark that date range for retry rather than writing a zero.

That seven-day window is a tradeoff, not a universal setting. A longer refresh catches later revisions but consumes more API calls, especially across many accounts and posts. Also, keep daily buckets distinct from lifetime totals: a lifetime count fetched on two dates is two snapshots, not two amounts to add together.

What refresh window and change-logging approach have worked for members syncing post-level insights?

---

<div class="post-metadata">

**Author:** ![darius7](https://mpsocial.com/user_avatar/mpsocial.com/darius7/32/75480_2.png) [@darius7](https://mpsocial.com/u/darius7)\
**Post date:** [October 5, 2026, 8:23pm UTC](https://mpsocial.com/t/how-far-back-should-a-post-insights-sync-refresh/163677/2 "2026-10-05T20:23:18Z")

</div>

1. Start with a provisional refresh window, then measure **observed revision lag** : for each changed daily bucket, record how many days passed between its metric date and the fetch that detected the change. That is detection time, not necessarily when the provider revised it.

2. Review the lag distribution after several sync cycles. If changes cluster near the window’s edge, extend it; if they stop well before it, test a shorter window.

3. Run an occasional wider backfill and compare it with stored values. Any revisions found beyond the routine window show what that window misses and give you a reason to adjust it.

---

<div class="post-metadata">

**Author:** ![theo\_builds](https://mpsocial.com/user_avatar/mpsocial.com/theo_builds/32/75482_2.png) [@theo\_builds](https://mpsocial.com/u/theo_builds)\
**Post date:** [October 5, 2026, 9:59pm UTC](https://mpsocial.com/t/how-far-back-should-a-post-insights-sync-refresh/163677/3 "2026-10-05T21:59:43Z")

</div>

The occasional wider backfill is useful for checking what the routine window misses. A practical setting is the 95th percentile of observed revision lags plus a small buffer, recalibrated when backfills find later changes. The current window censors those observations: seeing no changes within it does not establish that revisions stop there.

---

<div class="post-metadata">

**Author:** ![darius7](https://mpsocial.com/user_avatar/mpsocial.com/darius7/32/75480_2.png) [@darius7](https://mpsocial.com/u/darius7)\
**Post date:** [October 5, 2026, 11:41pm UTC](https://mpsocial.com/t/how-far-back-should-a-post-insights-sync-refresh/163677/4 "2026-10-05T23:41:15Z")

</div>

Agreed: the wider backfill checks what the routine window misses. For example, audit January on one run, then persist February 1 as the next backfill start date. Advancing that cursor through older, unaudited history eventually checks those buckets too, rather than repeatedly scanning the same wider recent range.

---

<div class="post-metadata">

**Author:** ![theo\_builds](https://mpsocial.com/user_avatar/mpsocial.com/theo_builds/32/75482_2.png) [@theo\_builds](https://mpsocial.com/u/theo_builds)\
**Post date:** [October 6, 2026, 12:49am UTC](https://mpsocial.com/t/how-far-back-should-a-post-insights-sync-refresh/163677/5 "2026-10-06T00:49:37Z")

</div>

Treat the cursor as a checkpoint: advance it only after the entire range has been fetched and every value successfully upserted. If a fetch or upsert fails partway through, leave the cursor unchanged and retry that range. The upserts make replaying the completed portion safe.
