The report: what it found, and what it threw away · The workings: every figure behind it · Outreach: what an SDR sends, per tier
Nobody picked these companies.
646 went in. Two came out worth calling. Press play and watch every single drop, with the reason attached. Nothing here is pre-calculated — every number is read from the run’s own output, so this page can’t show you a step that didn’t happen.
The whole machine
Code collects and scores at no cost; Claude only ever judges what the code hands it. The step in colour deletes the analyst’s own list of weaknesses before the reviewer sees the claim — hand someone a list of known flaws and they check that list and stop looking.
Now watch it run
646companies
Ready
Press play, or it starts on its own.
Read the public complaint database
109,004 complaints. 46 months. Every company named. No AI touches the data - this part is plain code.
This part costs nothing.
Drop companies too small to measure
A percentage built on three complaints is not a percentage. 74 are big enough to measure.
572 removed
Ask who is taking a GROWING SHARE
Not who has the most. The whole market doubled, so everyone went up. Share strips that out.
53 had no rising share
Remove companies that just appeared
Rising from zero just means they started existing. 9 set aside.
9 removed
Check nothing merged or vanished
One company in 646 reports steadily, then stops dead. Whoever jumped that same month just inherited their complaints.
MERGER CAUGHT
An analyst writes the case for each
The best honest argument for each one - weaknesses attached.
12 claims
A DIFFERENT model tries to destroy each one
It never sees the first one’s doubts. Hand someone a list of known flaws and they check that list and stop looking. 181 tests.
verdicts land
One rewrite, then a final reviewer
Repairable claims go back once. Second attempt is final - hold or drop. 24 corrections demanded, all applied.
5 held, 0 dropped
Sort EVERY company into a tier
Fail a check and you drop a tier. You never get deleted. That is why this list is 74 names and not 2.
nothing discarded
The two bits worth watching twice
The merger check. No statistical test could do this one — the numbers it corrupts look perfectly healthy. It took something actually going looking.
The attack. This is where 5 of the 12 accounts died, including the second-best one on the list.
16 more are out because their share is falling — several while their raw complaint count climbs. Counting complaints finds big companies. Measuring share finds change.
Built from the pipeline output: trend3.json, tiers.json and both sets of review verdicts. Re-running the pipeline changes this page; nothing here is written by hand.