Updating from Siftsy Score V2 to V3
Siftsy Score Version 3 is an evolution, not a replacement. If you've been reading v1 or v2 scores, your instincts still work: the scale, the signals, and the core question are unchanged. What v3 adds is precision — a clearer read on what the audience is reacting to, and guardrails against the ways comment sections can mislead a naive average.
How to switch
Switching is a team setting. In Team Settings, open Analysis Settings and select Version 3 as your team's score version. Every score updates instantly, including posts you've already analyzed. Nothing is re-run and nothing is lost, because all versions are built from the same per-comment AI scores; a version is just a different way of rolling them up.
The switch is fully reversible. You can toggle your team between versions at any time — flip to v3, compare it against v2 on posts you know well, and flip back if you're not ready. Your history stays intact either way.
What stayed the same
The same three signals. Every version blends Sentiment, Relevance, and Consensus, each on the same 1–10 scale with the same definitions. Consensus still measures how commenters treat each other, not whether they agree with the post.
The same weighting as v2. Version 1 weighted all three signals equally. Version 2 doubled Sentiment (66% / 16.5% / 16.5%) because the mood of the section is what most people mean by "how did this land." Version 3 keeps that exact blend — a section that scored 5.9 under v2 typically scores 5.9 under v3.
The same scored comments. All versions run on the same context-aware, per-comment AI scoring. Versions are just different ways of rolling those comments up, which is why switching versions never loses history.
The same job. High still means on-topic, mutually supportive, and not hostile. Low still means negativity, irrelevance, or infighting is running the section.
What changed, and why it helps
Version 3 changes four things about how the score is read and rolled up.
A simpler reading scale
The four bands of v1/v2 (Immaculate / Good / Sub-Par / Off) become three: Immaculate (7.0+), Mixed (4.5–6.9), and Off (below 4.5). Mid-range scores are usually mixed reactions, not mild failures, and the labels now say so.
Sentiment is split into Content and Topic
This is the headline change. Versions 1 and 2 gave each comment one sentiment reading; v3 gives it two — attitude toward the post and creator (Content) and toward the subjects discussed (Topic) — averaged 50/50 into the familiar Sentiment number. The average is compatible with the old versions; the split is new information. "They hate this ad" and "they hate this product category" now produce visibly different breakouts, and each calls for a different response.
A negativity cap keeps hostile minorities visible
Averages hide outrage: 88% praise plus 12% genuine hostility can still average high. Version 3 lowers the score's ceiling as the very-negative share grows (12% very negative caps the score at 8.5). A section with a real hostile contingent can no longer read as flawless.
Engagement weighting uses a new curve and applies to fewer axes
In v1 and v2, weighted mode used a steep power curve: the most-liked comment counted as 1× and every other comment was a fraction of that — a comment with 100 likes out of a 1,000-like top comment carried under 2% of its weight. Version 3 replaces it with a linear-log multiplier: every comment starts at 1× and the most-engaged comment counts up to 5× (at full weighting strength). Likes also now scale only the sentiment readings — Relevance and Consensus describe the comments themselves, not audience mood, so they keep structural weight only.
Reply chains no longer drown the thread
In v1 and v2, every comment counted equally, so a long argument nested under one comment could outweigh dozens of independent top-level reactions. In v3, top-level comments carry full weight and replies share a capped pool, so the score tracks the breadth of the section rather than its loudest sub-thread.
The takeaway
Version 3 keeps the number you already know how to read and makes it harder to fool. Same scale, same signals, same weights as v2 — plus the Content/Topic split that turns a temperature check into a diagnosis, and guardrails that keep reply pile-ons, hostile minorities, and viral outliers from writing the headline. Because the toggle is instant and reversible, the safest way to evaluate v3 is simply to turn it on.