← Back to Unit 0 · Block 3 · Unit 0 · Block 4 · Unit 0 · Block 5 · Source Dossier assessment
This is NOT a topic you can choose. The exemplar uses microplastics in drinking water, which is deliberately not on the Step 1 list. Use it to learn the moves. Don't reuse the content.
Read the shaded boxes. Green boxes explain why a passage earns its band. The dossier itself shows what a strong (not perfect) submission looks like inside the 1500-word limit.
One flaw is planted and NOT annotated. Somewhere in the dossier below is a real OPVL error we left in on purpose. Finding it is a class activity. Bring your candidate to Block 4.
About how this was made: the dossier below was written to model strong student work. It was drafted with AI and revised by Ms. Jayanthi/Mr. Ignash. Its AI Documentation Template is on the last page, filled in the same way you'll fill in yours.
Prefer paper? Printed copies available in class.
The Exemplar Dossier
Topic: Are microplastics in drinking water harming human health?
Why this is contested: Nobody is really arguing about whether microplastics are in our water anymore. They're in bottled water, in tap water, and according to a few recent studies, in human blood. The argument is about whether the amounts we swallow actually hurt us. The science on that question is new and kind of messy, and I found sources looking at the same studies and saying opposite things. That gap is why I picked it.
Source 1 (Traditional): World Health Organization report
Citation: World Health Organization. “Microplastics in drinking-water.” WHO report, Geneva, 2019. who.int.
Origin: A technical report from the WHO, which is the UN's public health agency. Their water quality team wrote it in 2019 and the report lists the outside experts who reviewed it. One thing I noticed: it doesn't do any new experiments. It collects and grades studies that other scientists already did, so it's a review, not primary research.
Purpose: It was written to help governments decide whether microplastics in drinking water need regulation. The audience is policymakers, not people like me, and I think that explains the careful, hedged language. The WHO also has its own reason to stay calm here. If people panic and stop trusting tap water, that causes real harm in places where tap water is what people have.
Value: Of my three sources, this is the only one that rates the quality of its own evidence. It says openly which studies were well designed and which weren't. For my question, its main finding matters most: the evidence so far doesn't show harm at normal exposure levels, but there also isn't much good evidence yet in either direction.
Limitations: It's from 2019. The lab methods that detected nanoplastics in 2024 didn't exist yet, so its particle counts are probably too low. It also only covers swallowing microplastics, not breathing them in. And “no evidence of risk” is not the same thing as “evidence of no risk.” The report admits this itself, but you could easily miss it if you skim.
Source 2 (Weaker): viral Instagram reel
Citation: @wellnesswarrior_fit (handle anonymized). “Your bottled water has 240,000 pieces of plastic in it. Stop poisoning your family.” Instagram reel, January 2024, about 2.1M views.
Origin: A wellness influencer with no scientific credentials listed anywhere on the account. The reel went up a few days after a real study from Columbia University (published in PNAS in January 2024) that found around 240,000 plastic particles per liter of bottled water. So the number is real and I could trace it. The reel never links the study, though.
Purpose: The account's purpose is to spread misinformation and scare its followers.
Value: I almost wrote that this source has no value, but that's not right. It shows how science actually travels online. The number 240,000 made it from the paper into the reel, but the careful parts got dropped on the way. And reach matters: two million people watched this. Nobody in my family is going to read a WHO technical report, but this reel was in our family group chat by February, and it's part of why my aunt switched to a steel filter bottle. If my question is what the public believes about bottled water and why, this source answers it better than PNAS does.
Limitations: The reel turns a detection finding into a harm claim. The Columbia study counted particles (and about 90% of them were nanoplastics, which are far smaller than microplastics) but it made no claims about health effects at all. “240,000 pieces of plastic” also hides that size difference, and the health question is different for different sizes. There's no link, no author name, and no way to check anything without leaving the reel.
Source 3 (AI): Claude, asked cold
Citation: Claude (Anthropic), conversation of 14 August 2026. Prompt: “Are microplastics in drinking water dangerous to human health? Explain the state of the evidence.” Full transcript saved. AI Documentation Template attached as the last page of this dossier.
Origin: This one felt strange to OPVL because it isn't a fixed document. If my friend asks the same question tomorrow she'll get different text. So what I'm evaluating is one conversation that my exact wording produced, not “what Claude thinks about microplastics.” I also can't see what it was trained on, or when its information stops.
Purpose: Part of an AI's purpose comes from the prompt, which means part of it came from me. When I asked “is it dangerous,” the whole answer got organized around danger. Later I asked “what's the strongest evidence they're harmless” and the same tool sounded noticeably different. I didn't expect that, and it changed how I read the first answer.
Value: It gave the clearest overview of my three sources. It separated “what's been detected” from “what's been shown to cause harm” without me asking it to, and it brought up breathing in microplastics as an exposure route, which neither of my other sources mentions. Ten minutes with it saved me at least an hour of searching.
Limitations: The biggest problem: it told me the WHO “considers microplastics in drinking water safe.” I had the actual report open, and it doesn't say that. It says the evidence is insufficient to conclude there's a risk, and that more research is needed. Those are different claims. I only caught the difference because I was checking. It also cited no sources unless I asked, and it can't know anything published after its training data ends, which in this fast-moving topic could matter a lot.
Cross-source synthesis (~300 words)
All three sources agree on the detection facts. Microplastics are in drinking water, bottled water carries more particles than tap, and the 2024 nanoplastics study multiplied the counts. They split on what it means. The WHO says the evidence isn't strong enough to show harm and treats the whole thing as a research gap. The reel treats detection as proof of poisoning. The AI landed closest to the WHO, except it rounded the WHO's careful wording up to “considered safe.”
Each source turned out to be good for a different question. For what the evidence actually shows, the WHO report, because it's the only source that grades its own evidence. For what the public believes and why, the reel, because the reel is an example of the problem, not a report about it. For a first overview of the whole debate, the AI, though of the three it was the hardest to check.
Someone who only read my AI source would miss two things the WHO report gave me. First, the evidence grading, meaning which studies are solid and which are one-off results. Second, the worry about what happens if water panic spreads in places where tap water is the only option. No chatbot summary brought that up. They would also inherit the AI's rounding error about the WHO's conclusion with no way to notice it.
The strange part only appeared when I put all three next to each other. The reel and the AI made the same kind of mistake in opposite directions. The reel turned “we found particles” into “poison.” The AI turned “not enough evidence” into “safe.” Both took an uncertain statement and made it sound settled, and the professional-sounding one was much harder to catch. If I had three more sources, I'd want a peer-reviewed toxicology study from after 2024, something from the bottled water industry, and a second AI conversation asking the opposite question.
AI Documentation Template (for Source 3, as required)
Tool(s) used: Claude (Anthropic), August 2026, via the class platform.
Prompts (representative): “Are microplastics in drinking water dangerous to human health? Explain the state of the evidence.” / “What's the strongest evidence that they're harmless?” / “What did the WHO 2019 report actually conclude? Quote its language if you can.”
What the AI contributed: the split between “detected” and “shown to cause harm”; pointing out inhalation as an exposure route; a first overview of the debate.
What I contributed: picking the topic; finding and actually reading the WHO report and the PNAS abstract; catching the “considered safe” error by checking the report; all three OPVLs; the synthesis; every word of the final text.
OPVL of the AI's contribution: see Source 3 above. The OPVL is the dossier entry.
One thing the AI got wrong: it said the WHO “considers microplastics in drinking water safe.” The 2019 report actually concludes the evidence is insufficient to determine risk and calls for more research. Verified against the report itself and corrected in my dossier.
Same Source, Different Band: Three Pairs
In each pair, version A and version B describe the same source, but one earns the upper band and one doesn't. The difference is never length. Read the pair, name the difference yourself, then check the note.
Pair 1: The Limitation (this one moves your grade)
Version B: Exemplary (7–8)
“The AI told me the WHO's conclusion was 'considered safe,' but the 2019 report actually says the evidence is insufficient to conclude risk. I caught it because I had the report open while we talked.”
Pair 2: The Origin
Version B: Exemplary (7–8)
“A 2019 technical report from the WHO's water quality team, with named outside reviewers. It doesn't run new experiments; it collects and grades other people's studies. That matters because its conclusions can only be as current as the studies it had in 2019.”
Pair 3: The Synthesis Sentence
Version B: Exemplary (7–8)
“The reel and the AI made the same kind of mistake in opposite directions. The reel turned 'we found particles' into 'poison,' and the AI turned 'not enough evidence' into 'safe.' So the real skill probably isn't avoiding bad sources. It's noticing when any source makes uncertainty sound settled.”
AI Documentation: for this exemplar itself
This document models the protocol it teaches.
Tool(s) used: Claude (Anthropic, Fable 5), July 2026, working from the course's assessment page, OPVL framework, and rubric.
Prompts (representative): “Create an annotated Source Dossier exemplar on a contested topic NOT on the student list, matching the course's OPVL framework and rubric bands, with strong/developing discrimination pairs.” / “Review what you wrote and analyze for what reads as AI-consistent language.”
What the AI contributed: the full first draft, and a second draft after the voice problem was flagged.
What Ms. Jayanthi/Mr. Ignash contributed: the decision to build exemplars; the requirements (off-list topic, annotated range, planted flaw); catching that the first draft read as AI-written and sending it back; review and final judgment on every band claim.
One thing the AI got wrong: the first draft didn't sound like a student. It had AI habits all through it: an em-dash in nearly every paragraph, a tidy summary line closing every section, perfectly balanced sentences, no lived detail. Ms. Jayanthi/Mr. Ignash flagged it and the draft was rewritten. Keep that list. You'll want to check your own AI drafts for the same habits.