Evidence appendix to the Harbor brief · 9 September
Three real buying surfaces, read end to end: Progressive, GEICO and The Zebra. Everything below is quoted from pages any shopper can reach. No personal information was entered and no quote was submitted — this is the surface up to the point where a carrier asks who you are.
Progressive’s coverage page states exact dollar limits for the coverages that cannot hurt you, and no number at all for the ones that decide whether you are ruined.
| Coverage | What the page tells you |
|---|---|
| Pet injury | $1,000 |
| Trip interruption | $500 total · $100 lodging · $50 transport · $50 food |
| Custom parts & equipment | $5,000 |
| Deductible Savings Bank | $50 per period |
| Liability | “up to your policy’s limits” |
| Comprehensive | “events out of your control” |
| Collision | “pays to repair or replace” |
observed Progressive, Car Insurance Coverages. Not one exclusion appears anywhere on the page.
“What isn’t covered varies by policy and insurer, but car insurance policies typically don’t cover: routine maintenance · wear and tear · personal items inside your vehicle · ride-sharing and commercial usage · general mechanical breakdowns · injuries to pets.”
inferred Every item on that list is something a reasonable person already assumes. Nobody believes car insurance pays for an oil change. The exclusions that actually decide claims — what counts as an event out of your control, what “actual cash value” means at a total loss, whether the damage and the debris are covered — appear nowhere on any buying surface I read. And the answer opens by disclaiming itself: “varies by policy and insurer.”
“To best compare car insurance quotes from different providers, ensure you select the same coverage, policy limit, and deductible amount.”
inferred That is the correct instruction and it describes a task the surface does not support. There is no way to export your current coverage, no way to import it into a competitor’s quote, and no shared vocabulary between carriers. The carrier names the precondition for comparison and ships nothing that meets it.
Progressive’s coverage calculator asks nine questions before recommending anything:
observed Completed for Arkansas, it returns a real coverage table — bodily injury $50k/$100k, property damage $50k, UM/UIM $25k/$50k, med pay $5k, rental $40/day. GEICO has an equivalent tool and also links to it from a content page, not from the quote.
inferred That question set is a financial-planning interview: net asset exposure, loan-to-value position, liquid savings, substitute-vehicle access, risk tolerance. The carriers know exactly what determines correct coverage. The purchase path asks for a ZIP code and returns a price.
observed And the calculator undercuts itself in its own FAQ: “our car insurance calculator recommends a minimum set of coverages for your situation.”
“Insurance companies pay us a commission when you purchase a policy or get a quote through us.”
“It depends! … Your final rate can still change if the carrier uses updated or additional information when finalizing your policy.”
inferred Price is the only comparable number on the whole surface — and it is provisional. Meanwhile the comparison layer earns on the quote itself, so the volume of shopping is partly a product of how the channel is paid rather than of consumer intent.
Everything specific enough to compare is trivial. Everything that decides an outcome is vague.
If that holds, then shopping does not produce a decidable comparison. Coverage cannot be normalized, exclusions are not published, the only comparable number is provisional, and the tool that would tell you what you actually need sits outside the path. A person who shops and stays may not be hesitating. They may be responding correctly to a comparison that cannot be made — and a real switching cost on the other side of it.
inferred This is a reading of three websites, not a finding about people. Written before the account reads, and kept here unedited. It has since been tested twice: the live account read corrected part of it — inside a policy, coverages are itemised and individually priced — and a fourth interview independently produced the buyer-side version of it: “there are gaps in that knowledge, and that’s why I think you need an agent.” It still has not been tested on home insurance.
Inside a live account, the whole policy-management surface contains zero occurrences of “cancel”, “non-renew” or “terminate”. There is no way to leave from the place you manage the thing you are leaving.
| Action | How it works |
|---|---|
| Add a vehicle | Self-service, and a price preview is explicitly offered — “Just checking the price? Quote a vehicle.” |
| Add a driver | Self-service, same offer |
| Reduce or change coverage | Self-service. No price preview offered. |
| Leave | Not available in the app at all. Filed under Contact Us — phone or chat |
observed And the carrier’s own guidance names a deadline the account never does: “If you’ve set up an automatic payment plan, be sure to contact your insurer and cancel before the first payment of the renewal term is automatically paid.”
Read that against the account. The policyholder is on automatic payments. The renewal term begins in eighty-five days. So the renewal executes by default, the only way to stop it is to reach a human, and the deadline is the moment the money moves — a date the account never states, on a forward calendar that says “nothing scheduled.”
inferred You are invited to preview what it costs to add. You are not offered a preview of what it saves to reduce. And you cannot leave without speaking to someone. Each of those is defensible on its own; together they describe which decisions the surface is built to support.
A public forum for insurance questions reports 324,000 weekly visitors and 4,000 weekly contributions. It functions as a helpdesk for a product people have already bought.
Searching it for “am I covered” over the past year returns policyholders who do not know what they own:
inferred These are questions, not complaints — behavior rather than opinion, which is why I am willing to use them. Several are asked after the event, when the answer can no longer change the outcome.
And one is the tree again, in different clothes. A homeowner’s basement flooded in a rainstorm; the carrier called it groundwater and denied the claim. Rain versus groundwater is the same structure as wind versus lightning — coverage turning on a peril distinction the buyer learns at claim time. Unlike the version in my third interview, I found this one rather than introducing it.
I searched the same forum for the opposite — people saying a renewal notice had explained an increase, or that comparing coverage was easy. A year of results for “renewal notice” returned seven posts, every one a confusion or a dispute, and not one person saying a document made sense.
That is not evidence, and I am not counting it as any. Nobody posts to say a document made sense. The search could not have falsified the finding whatever the truth is — which makes it a badly designed test, not a passed one. A real falsification would have to sample the people who never posted.
Recorded because a brief that only reports the tests that worked is not a record.
Seven claims in the brief were wrong and were fixed. One correction was itself wrong and was reversed. Kept here, out of the argument, because it is a working record rather than a finding — and because a study that publishes its own errors should not make the reader wade through them to reach the evidence.
| # | What was wrong | Found by |
|---|---|---|
| 1 | A coverage example from the first interview was carried into the third and raised by me. The agreement that followed proves nothing, because I supplied it | Me, on review |
| 2 | I described my own findings to the fourth interviewee mid-call. Everything they said afterwards about transparency is discarded | Transcript |
| 3 | A fourth instance of a pattern, assembled from two adjacent remarks the subject never connected | Challenged in review |
| 4 | Telematics “interpreted in detail” — traced to no observation on any page | Later review |
| 5 | “Price is almost never what makes them move” — contradicted by the brief’s own §04 | Later review |
| 6 | Four carrier-staffing claims, named and checkable, reading as market data when they were one informant’s untested testimony. Cut rather than caveated | Later review |
| 7 | “Both carriers” written as though it meant the industry. It means the two this household used | Later review |
AND ONE CORRECTION THAT WAS ITSELF WRONG
I removed the ~16% home shopping rate as unsourced. It was sourced — to Harbor, on page one of the assignment. Restored. It is the only one that ran the other way: hunting for over-claims and deleting something true. Worth more than the other seven put together, because it is the failure the discipline causes rather than the one it catches.
Six of the seven ran toward the conclusion — the direction you would expect from someone who wants their thesis to hold. Every one was caught by a later reading, not by the first. That is what the evidence tags and the pre-registered kill conditions are for, and it is the only honest basis for believing the parts nobody has checked yet.