← Prism overview
Prism · Worked example

Reading one contested question.

What it looks like when Prism turns a real public consultation into a structured, auditable account of what was actually said — taken here from a single position-group: whether and how children's ages should be verified online.

The setup.

This is drawn from one position-group in a real consultation — the US Federal Trade Commission's 2024 review of its Children's Online Privacy Protection Rule, used as a public-data demonstration on a corpus comparable in scale and character to a ministerial consultation. The full docket was reduced to the discrete positions each comment advances, and positions expressing the same idea were grouped together.

The intent is not to argue a conclusion about age verification. It is to show that a free-text consultation can be turned into a structured, auditable account of what was said — one in which every figure resolves to named comments, and every claim resolves to the verbatim text it was drawn from.

What the group contains.

Seventeen distinct commenters, twenty-four position-statements, one question. The group resolves into three coherent positions plus a clearly-marked outlier — and Prism keeps them distinct rather than averaging them into a single number. Every quote below is real, ordered by how representative it is, and traces to a named comment in the public docket.

Cluster: “Age verification mechanisms and limits”
Raised by 17 distinct submitters across 24 extracted positions · distributed broadly across submitters.
Oppose 8 Support 6 Modify 5 Other 1 Non-substantive 4
cohesion 0.46 · isolation 0.36 · not flagged as a minority cluster
Oppose · 8

Scepticism that verification works

The largest position: age checks are trivially circumvented. A feasibility objection, not opposition to protecting children — and Prism keeps it distinct from the calls for stronger enforcement rather than merging the two.

“It's not very hard to fake your age if you really need something. I've faked my age when I want to watch a movie or get to some pdf that isn't available normally.”

Isabel Yang · FTC-2024-0003-0079 · centrality 0.83

“The dialog on websites to verify a user's age is too easy to simply ignore or provide a false age.”

Dennis Bi · FTC-2024-0003-0057 · centrality 0.80

…and six more oppose voices in this group.

Support · 6

Calls for stronger enforcement

A second bloc holds that platforms and regulators should strengthen verification rather than abandon it — the opposite practical conclusion from the same concern for children.

“social media platforms must enhance age verification processes to ensure the efficacy of these protective measures”

Justin Leach · FTC-2024-0003-0157 · centrality 0.77

Modify · 5 · most decision-relevant

A concrete reform proposal

Three different commenters, in three different wordings, independently land on the same amendment: raise the covered age above 13. A precise, actionable change that emerged from unprompted convergence — not one source repeated.

“the rule should be made more stringent by increasing the age to 18”

Parent & Licensed Counselor · FTC-2024-0003-0145 · centrality 0.71

“I suggest raising the age to 18 rather than 13”

Chris Marchioni · FTC-2024-0003-0121 · centrality 0.70

“13 is far too young, the age should be 16 or 18”

Matthew Hilbert · FTC-2024-0003-0092 · centrality 0.66

Other · 1 · outlier

An outlier, kept visible

One commenter goes further than the rest. It is recorded under a separate posture rather than forced into the three above — the taxonomy is open, and genuinely novel positions stay visible rather than being discarded.

“maybe we have to ban until 18 years of age”

Melanie Hauck · FTC-2024-0003-0091 · centrality 0.61

Four further phrases in this group were tagged non-substantive — affect, boilerplate, framing — counted but never quoted. Quarantining them keeps the substantive map clean while leaving them in the audit trail.

Read together, the group is a compact map of a real debate: a feasibility critique, a demand for stronger measures, and a specific amendment, with one position not reducible to the others. A decision-maker sees not only the balance of opinion but a concrete proposal worth a response — and can verify each element against the source.

Source: PRISM cluster 20260607-165457-cc20 · clustering run 20260607-165457 (k=25) · docket FTC-2024-0003. Figures reproducible from the same run.

How to read the figures.

Each group carries a small set of descriptive measures. They are diagnostics to orient a reader, not automatic decision rules — the analyst decides what matters; the figures only say where to look.

Distinct submitters / positions

Here, 24 position-statements from 17 different people. The two numbers differing is normal — one person may advance several positions — and the gap between them is itself informative.

Posture breakdown

Support · Oppose · Modify · Clarify · Other · Non-substantive. Each position is tagged by the stance it takes, not just the topic it mentions — so a group can hold genuine disagreement and the reading shows its internal structure. Non-substantive marks sentiment or boilerplate that carries no position: counted, never quoted.

Cohesion (0–1)

How closely the positions resemble one another. High means a tightly unified group; lower means a broader thematic group spanning related sub-positions. This one sits mid-range — best read as "the age question" rather than a single narrow claim.

Isolation (0–1)

How cleanly the group separates from its neighbours. A sharper boundary scores higher — how confidently it can be treated as a distinct topic versus one shading into adjacent ones.

Minority flag

Marks a position smaller in volume but internally coherent, so a genuine minority view is preserved when totals are summarised rather than washed out by louder aggregates.

Centrality (per quote)

How representative a given quote is of its group. Evidence is ordered most to least typical, so the first quote under each posture is the clearest exemplar — not a cherry-pick.

Distributed broadly

A concentration check: the positions come from many different people rather than one or a few. This is the first line of defence against mistaking amplified repetition for breadth of opinion.

Breadth, not amplification.

A consultation's value rests on separating genuine breadth of opinion from the same input repeated under many names. This docket contains coordinated submissions — form-letter campaigns in which one drafted text is filed many times. That is common and not in itself improper; the relevant question is whether a given group reflects many independent people or one source amplified.

The age-verification group is independent in this sense: its positions are distributed across many separate commenters, and its strongest evidence is a set of distinct individual voices rather than a recurring template. Other parts of the same docket are not — and Prism surfaces that distinction rather than letting repetition inflate an apparent consensus. (Elsewhere in this same docket, a group that read as broad consensus turned out to be one advocacy form-letter filed a dozen times — a separate worked example.)

Two cautions, stated plainly. These measures describe structure, not intent: near-identical positions establish coordination, not motive. And convergence on shared language can reflect a genuine groundswell as readily as an organised campaign — the structural signal flags a candidate for review; timing, filing patterns and provenance are what confirm it. The judgment remains the analyst's.

The audit trail runs both ways.

The quotes above run one direction — from the group to its evidence. The trail also runs back: from a position to the full comment it came from. And because a single comment usually speaks to several matters at once, each of its positions is filed where it belongs while the rest goes to the topics it actually addresses. Two of this group's commenters, shown whole:

Source comment · the Oppose voice above · FTC-2024-0003-0079
Isabel Yang's comment, across three topics.
Safeguarding Children Against Online Harms Age verification mechanisms and limits Persistent identifiers and internal operations exception rules
I'm a teenager that, like many of my friends, often go on social media. Many times when I download a game or enter a sketchy website, it'll have a popup asking for my age. It's not very hard to fake your age if you really need something. I've faked my age when I want to watch a movie or get to some pdf that isn't available normally. This is a big issue because websites often will track my information, especially if it's something like shopping.

The age-verification scepticism quoted above (green) is one of three positions in this short comment — it also speaks to online-harm safeguarding and to tracking via persistent identifiers. 99% of the text carried a position.

Source comment · the Modify proposal above · FTC-2024-0003-0092
Three sentences, three topics.
Age verification mechanisms and limits Safeguarding Children Against Online Harms Corporate misuse and exploitation of minors' data
I agree with other commenters, 13 is far too young, the age should be 16 or 18. Also, FTC need to compel the tech companies to build more robust (not easily worked around) tools for parents to manage digital privacy, usage and content for minor children. It is clear to me that what they have built are designed to fail.

“13 is far too young…” — the raise-the-age proposal quoted above (purple) — sits beside a call for better parental tools and a verdict on the current ones. Each routed to its own topic; the connective tissue greyed out.

Doubt any of it? Start from a position and trace it back to the words on the page, or from a comment and trace it forward. The Prism overview shows the same span-level highlighting on longer comments.

Scope and limits.

This is a public-data demonstration; the figures are reproducible from the same consultation and run. The corpus is a US docket chosen for availability and for its resemblance to a ministerial consultation in scale and form — not for its subject.

Prism is built for meaningful human involvement: it organises, measures, and makes traceable, but defers contested and borderline cases to a human reviewer and records its uncertainty rather than resolving it silently. Nothing here is a final determination — it is a structured, checkable starting point for one.

See the rest of the method.

This is one reading from one group. The Prism overview shows how the whole pipeline works, with live colour-highlighted examples. If you're working with large volumes of free-text responses, we'd be glad to walk you through it.

info@szlamkaconsulting.com