Instapect
Sign inJoin free

Methodology

Scoring changelog

The Instapect Score is a published model, and published models get things wrong. This page records every change to how it is calculated, why we made it, and what it did to existing scores. The current model is performance-v3.

How the model is reviewed

  1. 1

    Every score records its model

    Each report stores the version of the scoring model that produced it, and shows its own weights and arithmetic. Two scores are only comparable when their model versions match.

  2. 2

    We check scores against our own audits

    We regularly compare the score distribution with the engagement data across every public Instapect audit, broken down by follower tier. When one group of accounts scores high or low for structural reasons rather than performance, that is treated as a model problem, not a finding about those accounts.

  3. 3

    Changes are versioned and published here

    A change to any formula, weight, reference threshold or grading rule ships with a new model version and an entry on this page explaining what changed, why, and what it did to existing scores.

  4. 4

    Old reports are re-scored, with the old score kept visible

    Cached reports are re-scored from the posts and counts already stored with them, so no data is re-fetched and nothing about the account changes. The report then shows the score it had under the previous model next to the new one.

Formulas for every sub-score are in the Instapect Score documentation.

Change history

· Model change · performance-v3

Fairer scores for very small accounts, and provisional grades on thin samples

Accounts under 1,000 followers now have their own engagement reference band, a top Engagement Quality score needs a real comment base, and grades built on fewer than six posts are labelled provisional.

What changed

  • New engagement reference band for accounts under 1K followers: bottom quartile 6%, typical 12%, top decile 25%. The 1K–5K band keeps its previous anchors (2% / 4% / 8%).
  • Engagement Quality is capped at 80 when the typical post has fewer than 10 comments. A handful of comments on a small account can push the comment-to-like ratio past 5% without meaning the audience is unusually conversational.
  • Scores built on fewer than 6 posts with visible likes are labelled provisional on the report, with the post count shown. The number is still published; it is no longer presented as a firm grade.

Why

  • Before this change every account under 5K followers was measured against one band whose top-decile mark was 8%. Across Instapect's own audits, accounts under 1K followers had a median engagement rate of about 15% (bottom quartile about 6%, top quartile about 25%). Almost any small account therefore maxed the Engagement sub-score, which carries 35% of the weight.
  • In the reports scored under performance-v2, accounts under 1K followers averaged 75 points, 7 of 13 scored 80 or higher, and 9 of 13 received the maximum Engagement score. Several of those grades rested on three to five posts.
  • Larger accounts were not affected by this problem, which is why a small account could outscore a large one on the same week's data without performing better against its peers.

Effect on existing scores

  • Accounts under 1K followers move closer to the middle of the scale unless their engagement is high even for their size. Accounts above 1K followers are affected only by the Engagement Quality cap and the provisional label.
  • Every cached report has been re-scored from its stored sample. Each report shows the score it had under the previous model.

Known limitations

  • The under-1K anchors are based on 25 audited accounts plus the previous reference values. They will be recalibrated as the number of audits grows.
  • Our audit sample suggests the reference values for 10K–100K followers may be set high, but that sample is dominated by news and brand accounts. Those anchors are under review and unchanged in this version.
  • The Instapect Score is a published heuristic, not an official Instagram metric, and it only uses publicly visible data.

· Review · performance-v2

Scoring review: model versions recorded and every calculation shown

An internal review of how scores were stored and displayed. No formula changed, but reports became auditable.

What changed

  • New reports store an explicit model version. Older reports without one are labelled as legacy records rather than silently treated as the current model.
  • Each report renders its own weights and the arithmetic behind its total, instead of a generic description of the current model.
  • The content date range and the experimental status of the model are shown next to the score.
  • The public report directory now lists every non-delisted report with search, instead of the first 60.

Why

  • A user compared a report with an earlier reading of the same account and saw a different score. The review found the difference came from a model change on 29 August, not from the account, and that the reports gave no way to tell.

· Model change · performance-v2

The score measures performance, not audience size

Follower count was removed from the headline score and reported beside it instead, so accounts are graded against peers of the same size.

What changed

  • Audience Reach weight changed from 20 to 0. It is still shown as context.
  • Engagement weight changed from 30 to 35, Engagement Quality from 20 to 25, Growth from 15 to 10.
  • The 100K–1M engagement reference was split into 100K–500K and 500K–1M bands.
  • Content Performance now rewards how far the best post rises above the typical one, instead of rewarding uniform likes.

Why

  • Including follower count in the score meant large accounts scored well regardless of how their posts performed, and small accounts could not score well at all.

Effect on existing scores

  • Large accounts with low engagement scored lower than before; reports created before this date are not directly comparable with later ones.