YouTube Data API returning inconsistent results with duplicates

Question

There have been numerous questions about inconsistent results from the YouTube Data API: 1, 2, 3, 4, 5, 6. Most of them have accepted answers that seem to indicate there was a problem with the API request that was fixed by the instructions in the answers. But none of those situations apply to the API request discussed here.

There have also been two questions about duplicates in the API results: 1, 2. Both of them have the same answer, which says to use the next-page token. But both questions say the token was used, so that answer is not helpful.

Yesterday, I submitted a series of API requests to get the list of most-viewed videos about 3D printing. The first request in the series was:

https://www.googleapis.com/youtube/v3/search?q=3D print&type=video&maxResults=50&part=id,snippet&order=viewCount&key=<my key>

I ran that in a VBA sub, which took the next-page token from each result and resubmitted the URL with &pageToken=<nextPageToken> inserted.

The result was a list of 649 unique video IDs. So far so good.

After making some changes in the VBA code and seeing some duplicates in the result set, I went back today and ran the original VBA sub again. The result was again a list of 649 video IDs, but this time the list included duplicates and it also included IDs that were not in yesterday's list and was missing IDs that were there yesterday. Here is a comparison from the first two pages and the last two pages of the two result sets:

Page	# on page	# overall	Run 1	Run 2	Same as	Seq	Dup
1	1	1	f2mdMcf-fJs	f2mdMcf-fJs	1
1	2	2	WSauz5KVKTU	WSauz5KVKTU	2	Seq
1	3	3	zsSCUWs7k9Q	XYIUM5TkhMo	None
1	4	4	B5Q1J5c8oNc	zsSCUWs7k9Q	3	Seq
1	5	5	cUxIb3Pt-hQ	B5Q1J5c8oNc	4	Seq
1	6	6	4yyOOn7pWnA	LDjE28szwr8	None
1	7	7	3N46jQ0Xi3c	cUxIb3Pt-hQ	5	Seq
1	8	8	08dBVz8_VzU	4yyOOn7pWnA	6	Seq
...
1	13	13	oeKIe1ik2O8	e1rQ8YwNSDs	11	Seq
1	14	14	FrG_eSECfps	RVB2JreIcoc	12	Seq
1	15	15	pPQCwz2q96o	oeKIe1ik2O8	13	Seq
1	16	16	uo3KuoEiu3I	pPQCwz2q96o	15	NOT
1	17	17	0U6aIwd5h9s	uo3KuoEiu3I	16	Seq
...
1	47	47	ShGsW68zbIo	iu9rhqsvrPs	46	Seq
1	48	48	0q0xS7W78KQ	ShGsW68zbIo	47	Seq
1	49	49	UheJQsXOAnY	0q0xS7W78KQ	48	Seq	Dup
1	50	50	H8AcqOh0wis	H8AcqOh0wis	50	NOT	Dup
2	1	51	EWq3-2VuqbQ	0q0xS7W78KQ	48	NOT	Dup
2	2	52	scuTZza4f_o	H8AcqOh0wis	50	NOT	Dup
2	3	53	bJWJW-mz4_U	UheJQsXOAnY	49	NOT
2	4	54	Ii4VYsh9OlM	EWq3-2VuqbQ	51	NOT
2	5	55	r2-OGUu57pU	scuTZza4f_o	52	Seq
2	6	56	8KTnu18Mi9Q	bJWJW-mz4_U	53	Seq
2	7	57	DconsfGsXyA	Ii4VYsh9OlM	54	Seq
2	8	58	UttEvLJP3l8	8KTnu18Mi9Q	56	NOT
2	9	59	GJOOLH9ZP2I	DconsfGsXyA	57	Seq
2	10	60	ewgmg9Q5Ab8	UttEvLJP3l8	58	Seq
...
13	35	635	qHpR_p8lA4I	FFVOzo7tSV8	639	Seq
13	36	636	DplwDDZNTRc	76IBjdM9s6g	640	Seq
13	37	637	3AObqGsimr8	qEh0uZuu7_U	None
13	38	638	88keQ4PWH18	RhfGJduOlrw	641	Seq
13	39	639	FFVOzo7tSV8	QxzH9QkirCU	643	NOT
13	40	640	76IBjdM9s6g	Qsgz4GbL8O4	None
13	41	641	RhfGJduOlrw	BSgg7mEzfqY	644	Seq
13	42	642	lVEqwV0Nlzg	VcmjbJ2q8-w	645	Seq
13	43	643	QxzH9QkirCU	gOU0BCL-TXs	None
13	44	644	BSgg7mEzfqY	IoOXQUcW24s	646	Seq
13	45	645	VcmjbJ2q8-w	o4_2_a6LzFU	647	Seq	Dup
14	1	646	IoOXQUcW24s	o4_2_a6LzFU	647	NOT	Dup
14	2	647	o4_2_a6LzFU	ijVPcGaqVjc	648	Seq
14	3	648	ijVPcGaqVjc	nk3FlgEuG-s	649	Seq
14	4	649	nk3FlgEuG-s	27ZLFn8Dejg	None

The last three columns have the following meanings:

• Same as: If an ID from Run 2 is the same as an ID from Run 1, then this column has the # overall for Run 1.
• Seq: Indicates whether the number in column "Same as" is one more than the previous number in that column.
• Dup: Indicates whether an ID from Run 2 occurred more than once in that run.

Problems:

• The videos XYIUM5TkhMo, LDjE28szwr8, qEh0uZuu7_U, Qsgz4GbL8O4, gOU0BCL-TXs, and 27ZLFn8Dejg were returned as #3, 6, 637, 640, 643, and 649 in Run 2, but were not returned at all in Run 1.
• The videos FrG_eSECfps, r2-OGUu57pU, lVEqwV0Nlzg were returned as #14, 55, 642, in Run 1, but were not in Run 2.
• The videos 0q0xS7W78KQ, H8AcqOh0wis, and o4_2_a6LzFU were returned as #49, 50, and 645 in Run 2, but then each appears a second time in that run (as well as appearing in Run 1 as #48, 50, and 647).

These results are troubling. They mean that no single search will return a reliable list of videos for a given value of q.

I mentioned at the beginning that previous questions about inconsistent results from the YouTube Data API had answers that seemed to resolve those inconsistencies. Is there a way to do that for this search? Is there something wrong with the way I'm composing the search that is causing the problem?

If there isn't a way to fix the search, then I suppose the only way to get a list of videos on the topic with high confidence of it being complete is to run the search multiple times and merge the results until no new IDs appear that were not in a previous result set. But even then, one would not know if there are other videos lurking undetected.