So I spent some more time poking around for info on IRLA, and found that the fluency measure is supposed to be scored only on word reading accuracy, not on objectively measured rate. Rate is subjective: "appropriate rate and expression". And passing to the next level (color) is based almost entirely on the other measures of vocabulary, comprehension, and range of reading.
This is the progression:
Determine entry level for initial test: out of a list of 15 vocabulary words, correctly read and define at least 5. If 12 or more are correct, proceed to the next level of vocabulary words, and continue testing vocabulary until at least 5 but no more than 11 are correct. This is the level for the initial cold read assessment text.
Initial cold read qualifier: read text aloud with 99-100% accuracy, and "appropriate" rate and expression. If accuracy is below 99%, drop down a level for cold read, until accuracy rises to 99%. Conduct comprehension assessment at that level.
Comprehension assessment: if student obtains 9/12 comprehension points at this level, then this is their reading level.
Now where things get interesting is, as I mentioned above, once you've determined an entry point, the oral reading fluency measure is no longer important. (It literally constitutes 1% of the points required. Also, accuracy for exiting is 98%, rather than the 99% for initial placement.) All the remaining criteria for advancing up through the levels are based on vocabulary, comprehension, and range of reading. So it is hypothetically possible for, say, a careless reader with high comprehension to be placed a few grade levels below their actual level by the initial cold read, but then to progress rather rapidly up the levels by passing all the comprehension benchmarks.
I would say that, if one suspects a child is misplaced for fluency reasons, one should request that the comprehension assessments for exiting the level be administered, rather than re-assessing with the placement test. That should make it rather obvious that there is only a very tiny, and not very significant, reason for considering keeping them at that level. I suspect that is why, in your situation, they started backpedaling. Even though he may or may not meet the cold read qualifier for purple, he probably meets all the orange-to-purple transition criteria.
You could also suggest to your DS to slow down very slightly while reading aloud, which may increase his accuracy (if that was the issue) to above 99%. As long as he is still reading at what sounds like a natural pace, and not struggling to decode words (a few immediate, fluent self-corrections are allowed) he should get credit for appropriate rate and expression.
