How the score works What controlled tests found

Resonance 1

How well a paid video earned and kept the attention of the people it was shown to, and how often they did something with it, compared with a frozen reference set. 50 is the reference average. Higher means more of each, per person reached, not more people reached.

It is an experimental method. A score built from observed delivery is not a forecast for an unpublished video, a measure of belief change, or proof of what caused the result. Persuasion needs a controlled test; the Reference model page shows what those tests have found.

Engage25%

Did the opening stop the scroll?

3-second plays ÷ impressions

Reference average 23.56%

Interest40%

Did people watch 50% or more?

average watch time ÷ video length

Reference average 26.95%

Act35%

Did viewers do something?

shares + saves + comments ÷ 3-second plays

Reference average 1.60%

Design choice

Three components: Engage, Interest, Act

Attention earned at the open, attention kept given the length, and something done with it. Each is a rate a partner can check against Ads Manager.

Design choice

Weights 25 / 40 / 35

Interest carries most because it is the hardest to buy; Act next because it is the closest platform signal to intent. Tested below against three alternatives.

Design choice

Evidence weight k (2,000 / 300 / 800)

How many impressions, viewers or plays the reference average is worth. A take with far more evidence than k is scored on its own rate; one with less leans toward the average.

Design choice

Minimum 1,000 impressions; limited below 15,000

Below the minimum a take is not scored; between the two it is scored and flagged.

From the data

Reference averages and spreads

Learned from the first round of results, then frozen as ref-1. Later rounds are scored against it.

From the data

The range around each score

From the counts themselves: 400 draws of each rate consistent with what was recorded.

Each component starts as a rate from platform counts.
The rate is weighted by its evidence: (count + k × reference average) ÷ (total + k).
It is placed on the reference scale: how many reference spreads above or below the reference average, on the log-odds scale.
That distance becomes 0–100 as the share of a normal curve below it, so 50 is average and 84 is one spread above.
Resonance is the weighted sum. A missing component's weight is shared among the others and the take is flagged.

Impressions×1.00 = 49,3293-second plays×1.00 = 9,149Average watch time (s)×1.00 = 15.1Shares×1.00 = 48Spend ($)×1.00 = 392.7

Spend moves nothing: it is a delivery condition. Doubling impressions and plays together leaves the rates alone and only narrows the range.

As recorded43.8

Engage 25

Interest 86

Act 9

With your changes43.8

Engage 25

Interest 86

Act 9

Formula resonance-1, reference ref-1.

Take	Resonance 1 (25 / 40 / 35)	Equal thirds	Attention-heavy (40 / 50 / 10)	Action-heavy (15 / 25 / 60)	Evidence weight × 0.5	Evidence weight × 2
calm · Danielle	44 #9	40 #10	54 #6	31 #10	43 #9	44 #9
calm · Ruth	50 #5	57 #5	46 #8	60 #4	51 #5	50 #5
calm · Stephen	63 #3	63 #3	72 #2	54 #6	65 #3	59 #3
coin · Danielle	39 #10	44 #7	55 #5	26 #11	39 #10	39 #10
coin · Ruth	58 #4	57 #4	58 #4	58 #5	58 #4	58 #4
coin · Ruth · no captions	49 #6	43 #9	52 #7	42 #9	48 #6	50 #6
coin · Stephen	48 #8	49 #6	30 #10	68 #3	48 #8	48 #8
listen · Danielle	35 #11	33 #11	22 #11	47 #8	35 #11	35 #11
listen · Ruth	65 #2	64 #2	61 #3	69 #2	66 #2	64 #2
listen · Stephen	48 #7	44 #8	45 #9	49 #7	48 #7	48 #7
listen · Stephen · short cut	91 #1	92 #1	90 #1	93 #1	93 #1	89 #1
Rank agreement with Resonance 1	—	0.89	0.75	0.75	1.00	1.00

Rank agreement is 1 when every alternative orders the takes the same way. Ranks that move three or more places are marked.

Resonance vs. spend

-0.37

Near zero: budget does not buy a score.

3-second plays vs. spend

0.95

A raw count mostly measures budget.

Resonance vs. 3-second plays

-0.36

Ranking by views alone tells a different story.

Built 10/3/2026, 4:16:07 PM from 9 flights in round-1. Frozen: later rounds are scored against these numbers, and a revised formula would be published as Resonance 2 beside this one, never in its place.

Engage · pooled 23.56% · centre -1.24 · spread 0.33

Interest · pooled 26.95% · centre -0.91 · spread 0.26

Act · pooled 1.60% · centre -4.15 · spread 0.20

Stand-in records until the partner's export arrives: the takes are synthetic readings of the library scripts and the paid results are generated in Meta's export format. Upload the real export on Reviews and every screen recalculates from it.

Score method