IntegrationsSourcesBackfilling history

Backfilling history

A fresh connection only sees the future. Backfill pulls what already happened, so Memnox starts from an organization that has a past rather than one that began this morning.

Start it from Connectors → the connection → Backfill, choose a window, and leave it running.

Why it matters more than it sounds

Extraction proposes decisions from a window of events. Without history, the first weeks of suggestions are drawn from whatever happened to be said recently, which is disproportionately operational chatter rather than the decisions that actually govern how you work.

Backfilled history changes the shape of the queue immediately.

Console → Connectors, a backfill running with its progress and event count

Console → Connectors, a backfill running with its progress and event count

screenshot slot, save as public/screens/…png and set  src

Backfill runs asynchronously and reports progress. It is safe to leave and come back to.

How far back

Window

30 days

Fast, cheap, enough to make extraction useful

90 days

A quarter of decisions, the usual sweet spot

1 year

Long-lived products where the important decisions are old

Two limits bound the answer:

  • Your plan's history window. Backfilling beyond it stores events that will be pruned on the next sweep, so check the window on Settings → Billing before choosing how far back to go.
  • What the tool will give you. Providers page and rate-limit history, so a large backfill takes time.

Before you start

  1. 1

    Reconcile people first

    Author trust is decided at ingestion and never re-judged. Backfilling before identities are linked produces months of events attributed to unrecognised handles, and therefore tainted.

    This is the single most common backfill mistake, and it is not fixable afterwards without re-ingesting.

  2. 2

    Set the permalink base URL

    Slack and Jira events are rejected without it. A backfill into a workspace with no base URL discards most of what it pulls.

  3. 3

    Pick the narrow scope

    The channels and repositories where work is actually discussed. Backfilling every channel produces a queue nobody works through.

Watching it

Progress and the running event count appear on the connection itself. Nothing is silently dropped: a backfill that hit a provider rate limit shows up under Failed deliveries rather than simply ending early, so a short run is distinguishable from a finished one.

After it finishes

Two things, in this order. Reindex from Settings → Search so search covers what just arrived, then run extraction once from Decisions and expect a larger-than-usual queue.

Work that queue with somebody who was there for the period you backfilled. Historical decisions are exactly the ones where the evidence is ambiguous without a person who remembers the context.

Plan

History backfill is available on Pro and above.