all tiers · start here · why Sadness produced nothing

Why Sadness produced nothing — and what fixes it

A plain-language walkthrough, then listenable examples from each alternative rule.

1. The short version

Ask the miner for a chain where Sadness rises steadily, and it returns nothing at all. Not "a few", not "low quality" — zero. That looks like a bug. It is not. It is the scoring scale doing exactly what it was told, and the fix has to change the rule, not the code.

2. How a score becomes a number between 0 and 1

Every emotion is scored by a model, and those raw scores are not comparable between emotions — one model head may run hot, another cold. So before mining, each emotion is rank-normalised: a clip's value is replaced by the fraction of the corpus it scores above. The lowest clip becomes 0, the highest becomes 1, the median becomes 0.5. After that, "moved 0.25" means the same thing for every emotion — you crossed a quarter of the corpus.

One detail matters enormously here: when many clips have exactly the same score, they cannot be ranked against each other, so they are all given the midpoint of the range they jointly occupy. A tie of a million clips becomes a million copies of one number.

3. What that does to Sadness

90.0 % of clips score at or below zero on Sadness. Zero does not mean "slightly sad"; it means the model found no sadness. Those clips are all tied, so they all collapse onto a single point near 0.45. The clips that do carry sadness land up at 0.85–1.00. And in between there is nothing.

So the rule fails for a reason nobody guesses first. It is not that the chain cannot move far enough — one hop from the spike to the sad clips covers 0.45, far more than the 0.25 required. It is that the rule also says every single step may be at most 0.25, and there is nothing in the gap to step onto. With k=2 the one step is too long. With k=3 or more there is no middle clip to stand on. Both fail, so the miner returns zero — correctly.

4. The same picture for three emotions

Same axis, same bins. The only thing that differs is whether there is a gap wider than the step cap.

Sadness gap 0.449

Anger gap 0.121

Concentration gap 0.013

Read it as one sentence: if the widest empty stretch on an emotion's axis is wider than the step cap, the strict rule can never build a chain on that emotion. If it is narrower, the rule works fine.

5. Which emotions this predicts — and it does

emotion	clips ≤ 0	largest gap	wider than the 0.25 cap?	strict rule, k=3
Sadness	90.0 %	0.449	yes — blocked	0
Awe	95.1 %	0.474	yes — blocked	0
Distress	90.5 %	0.453	yes — blocked	0
Disappointment	84.3 %	0.422	yes — blocked	0
Helplessness	88.8 %	0.105	no	19,693
Anger	49.9 %	0.121	no	18,029
Concentration	2.7 %	0.013	no	14,763

The prediction holds exactly. Every emotion whose gap exceeds the cap returns 0 chains under the strict rule; every emotion whose gap is under the cap returns tens of thousands. Measured over 40,000 tracks.

One correction worth flagging: Helplessness was expected to be blocked too, but its gap is only 0.105 — comfortably under the cap — and the strict rule already yields 19,693 chains for it. It was never broken. A rescue tier for it is included purely as a control.

6. The alternative rules, and what each one costs

All of these are looser than the strict rule. None is a drop-in replacement; each buys examples by giving something up.

rule	what it changes	what it costs	Sadness k=2	k=3
`BASE`	The strict rule, unchanged, as a control. Rank-normalise the emotion over the whole corpus, then require the chain to rise by at least 0.25 end-to-end with every consecutive step at most 0.25.	Nothing -- this is the strict rule everything else is measured against.	0	0
`S1`	Positives-only renormalisation. Throw away the huge block of clips that score zero -- 'no sadness' is not a degree of sadness -- and rank-normalise only the clips that actually carry the emotion. The gap disappears, and the ordinary 0.25 / 0.25 test then works normally.	The number stops being comparable with other emotions: 'moved 0.25' now means a quarter of the sadness-BEARING subpopulation, not a quarter of the corpus. The chains also come from a small, self-selected slice of the data.	0	3,162
`S2`	Raw-delta. Ignore the normalised scale altogether and require the chain to rise by at least 1.0 in RAW score units.	Raw scores are uncalibrated and not comparable between emotions, so the same 1.0 means different things on different axes.	7,924	5,920
`S3`	Absent to present. The chain must start at or below 0.05 (the emotion is absent) and end at or above 1.0 (it is clearly present). The most literal reading of 'from not sad to sad'.	Says nothing about the SHAPE of the path -- only that it begins absent and ends present. Any intermediate clip is unconstrained.	7,826	5,806
`S4`	Same total move, bigger allowed step. Keep the 0.25 end-to-end requirement on the normalised scale but lift the per-step cap so a chain may cross the gap in one hop.	The chain is no longer a smooth ramp. It is allowed to jump the whole distance in a single step, which is exactly what the step cap existed to forbid. Use it for k=2 pairs, where there is only one step anyway, and be careful reading it as a gradual progression.	19,060	0

Two readings to be careful about. S4 shows 0 at k=3 not because it fails, but because a rule with no step cap has no way to prefer a middle clip — the search takes the biggest jump available first and has nothing left to reach for. Treat S4 as a k=2 rule. And BASE shows 0 at k=2 for every emotion, including healthy ones: when the required total and the per-step cap are the same number, a two-clip chain would have to move exactly the cap and no more. That is why every manifest tier reports zero chains at k=2.

7. Does the fix generalise?

emotion	gap	S1 k=3	S2 k=2	S3 k=2	S3 k=3	S4 k=2
Sadness	0.449	3,162	7,924	7,826	5,806	19,060
Awe	0.474	1,901	5,786	5,762	4,071	16,092
Distress	0.453	2,730	7,973	7,875	5,644	18,257
Disappointment	0.422	5,035	13,551	13,495	9,951	23,089
Helplessness	0.105	3,218	5,722	5,595	5,081	29,373

Yes. The mechanism is not specific to Sadness: it is a property of any emotion whose scorer outputs a large block of exact zeros. The same rules recover comparable numbers for Awe, Distress and Disappointment, which have the same shape of axis. S3 is the one to reach for first — it is the most interpretable ("started not sad, ended clearly sad") and it needs no renormalisation. S1 is the one to reach for if you want a genuine multi-step ramp rather than a pair.

8. Listen to them

Every example is labelled with the rule that produced it. The question to answer is simply: does it sound like the emotion is growing?

tier	emotion	rule	k	why it is here	shown
`sad-Sadness-S3-k2`	Sadness	`S3`	2	the literal 'not sad -> clearly sad' pair	20	listen →
`sad-Sadness-S3-k3`	Sadness	`S3`	3	the same, with one unconstrained clip in between	20	listen →
`sad-Sadness-S1-k3`	Sadness	`S1`	3	a genuine three-step ramp, once the zero block is removed	20	listen →
`sad-Sadness-S2-k2`	Sadness	`S2`	2	a pair defined purely by raw score movement	20	listen →
`sad-Sadness-S4-k2`	Sadness	`S4`	2	one hop straight across the gap, step cap lifted	20	listen →
`sad-Awe-S3-k2`	Awe	`S3`	2	does the winning rule generalise? Awe has the widest gap (0.474)	20	listen →
`sad-Distress-S3-k2`	Distress	`S3`	2	generalisation check, gap 0.453	20	listen →
`sad-Disappointment-S3-k2`	Disappointment	`S3`	2	generalisation check, gap 0.422	20	listen →
`sad-Helplessness-BASE-k3`	Helplessness	`BASE`	3	CONTROL: Helplessness has a gap of only 0.105, below the 0.25 step cap, so the STRICT rule already works here -- it was never blocked	20	listen →