How we score dependency

The two questions

Answered separately, on the evidence, before any rating is written down.

Each answer counts on its own, and the higher one sets the rating

Why the higher answer wins. An owner voting alongside an organization they trust acts on both levers at once, with one vote. Where either lever alone decides the outcome, that vote reaches the thing that decides the outcome. We never average the two answers, and we never require both to be high.

The codes carry no digits, on purpose. A reader never has to learn them: each answer is written as a sentence and the code sits beside it as a small tag for anyone checking our work. They were renamed so their characters cannot be read as a rating, since a political answer of the second step produces a rating of the third and the old digits invited exactly that mistake.

The five passes

Every item on every map, and every care report, goes through all five. Only the last one changes the entry.

Each is a different agent with a different disposition, and the disposition is the point. An agent told to verify an entry behaves nothing like one that believes an entry surviving it is worth more than an entry it agreed with.

1. Pass 1 · Researching

   The Analyst

   “The answer is in the sources, not in the thesis.”

   One pass researches your topic, answers the two questions our maps ask, and writes a first draft with every claim linked to a source.

   May not assert what it did not read, report a number without its source, or quietly rate a different outcome from the one the item states.

   This is where a preliminary report is sent to whoever asked for it. They can read it from here on, while the rest of the work continues.

2. Pass 2 · Checking

   The Skeptic

   “An entry that survives me is worth more than an entry I agreed with.”

   A second pass, which did none of the research, opens every source and recomputes every figure to find what the first pass got wrong.

   May not change anything or propose a rating, and may not argue from what past shareholder votes achieved. How a vote went is the state of a campaign, not the shape of the dependency.

3. Pass 3 · Searching wider

   The Bloodhound

   “The first pass found what it went looking for, and no more.”

   A third pass assumes the draft is incomplete rather than wrong, and goes looking for the evidence nobody searched for the first time.

   May not change anything, weaken an existing finding, bring in what it did not open and read, or say what the rating should be.

4. Pass 4 · Checking again

   The Gatekeeper

   “Evidence that raises a rating clears the same bar as evidence that lowers one.”

   Anything the third pass found is checked the same way the original claims were, so new evidence clears the same bar as old evidence.

   May not reopen what the previous pass cleared, and may not change anything.

5. Pass 5 · Deciding

   The Editor

   “The reader is owed a decision, and owed the disagreement where there is one.”

   A final pass reads every finding, decides what stands, and publishes the report. Where two passes disagree and neither is clearly right, it says so.

   May not introduce evidence of its own, average two positions, or pick silently where two passes genuinely conflict.

   This is where the finished care report replaces the preliminary one, at the same link.

Passes two, three and four never see each other's work in progress, and none may be the agent that produced what it is checking. Otherwise the second reader is agreeing with itself. The handoffs carry as much of the design as the roles do: the second pass's “I could not verify this either way” is the third pass's work order, the third pass's finds are not evidence until the fourth has opened them, and the fifth is the only one that ever sees all four sets of findings at once, which is why it is the only one allowed to decide.

Two arguments no pass may use, because neither is a test this method contains: that the votes cited were minority results and the company did not change, which describes the state of a campaign rather than the shape of the dependency; and “others must also act” standing alone, which is the conduct answer and says nothing about the second question.

A preliminary report and a care report

Two names for two different things, and the difference is how much checking has happened.

When you name something you care about, the first pass researches it and writes it, and we send you that within a few minutes. That is a preliminary report. It is one pass of work. Every claim in it is meant to trace to a source you can open, and none of those sources has been opened by anyone but the pass that cited them.

The other four passes then run on it, and when the last one has decided what stands, the finished version replaces the preliminary one at the same address. That is a care report. It has had every claim checked against its source, a wider search run for what the first pass missed, that new evidence checked to the same standard, and a final pass adjudicating the lot. It is the same treatment the published dependency maps get.

The link never changes. Anything you share while the preliminary report is up will show whoever opens it the care report once it lands, and we email you when that happens. Each version we replace is kept rather than deleted, so a sentence quoted from a preliminary report can still be found.

What every rating carries

A rating you cannot check is not worth much.

• Both answers. The conduct answer and the political answer, in our own words, with their codes.
• The rationale. Written for someone who knows none of this, naming the companies or industry whose conduct decides the outcome, and naming the lobbying where the lobbying is doing the work.
• Every source. Each factual claim carries a citation to a page we opened and read. No source, no claim, and no number without the source it came from.
• Which passes ran, when, and what each found.
• Which revision of this page produced it, and when.
• Which model ran each pass, separately, because the passes do not have to run on the same one. A checking pass on a different model family from the pass it checks does not inherit that family's blind spots, which is a stronger reading of the rule that no pass checks its own work. Where a model was asked for by a name that moves between builds, we record the exact build that answered, because otherwise we cannot tell a change in our method from a change in the instrument.

What earns each answer

• C-decides needs a mechanism, not an association. A source has to show that changed corporate conduct moves the outcome, not merely that companies are involved in it.
• P-blocks is the hardest answer to earn and the easiest to overclaim. It needs named companies or trade associations, a named bill, rule or decision, and a source showing the activity was the primary obstacle rather than one pressure among several. Where the companies deny it, we record the denial next to the claim. Where the measure passed anyway, that is P-slows.
• Only spending a vote can reach counts. A publicly traded company, or a trade association funded by public-company members. Where the influence traces to privately held firms, to an industry with no listed members, or to actors with no corporate owner at all, no vote reaches it and the political answer cannot carry the item.
• Scope is the whole item, not the convenient part. Where a large share of the outcome sits with actors no shareholder vote reaches, that lowers the conduct answer. Where an item names several things and only one is argued, the rating still covers all of them.
• We rate the outcome as written. Where an item asks for a change in law, we do not quietly rate the behavior the law would produce instead.
• No category of harm is non-corporate by name. Disease, accidents, crime and war are not outside corporate conduct by definition. Saying most of an outcome sits with actors a vote cannot reach is a claim about magnitudes and it carries a source like any other.
• Absences are findings. "We found no documented corporate lobbying on this" is an answer we publish, and it takes at least three searches before we will write it.
• Zeros and ones are scored honestly. Where the lever does not reach, we say so and name who the outcome belongs to instead, rather than reaching for the nearest corporate connection. That is a discipline about each item on its own, not a target for how the totals should look.

Three rules that sit on top

Applied after the two answers, never instead of them. Both answers stay on the record exactly as the evidence gave them; these rules change only the published rating.

Where this is hardest to apply

The method puts a great deal of weight on one judgment: whether corporate political activity is a primary obstacle or merely a contributing one. That single step separates Pivotal from Helpful, with nothing in between, and it rests on the least settled evidence we handle. Lobbying is usually reported rather than admitted, often denied, and rarely the only force in the room.

We hold that judgment in check three ways. The standard above is deliberately strict. Both answers are published on every entry, so you can disagree with our answer rather than only with our label. And every Pivotal rating says which lever carries it.

What we do not do is check the rule against its own output. How many items come out Pivotal is a finding about the world, not a target. Deciding in advance what share of the answers is allowed to be high, and then trimming to reach it, would replace the evidence with a quota.

What a rating is not

It is not a claim about how important the subject is. Several of the gravest items on these maps are rated Independent, and that rating says the harm is real and the lever is elsewhere.

It is not a finding of any organization whose list we scored. The UN, the Program for Public Consultation, JUST Capital and the Doughnut Economics Action Lab supply the items; the dependency ratings are ours, and none of those organizations is affiliated with us.

It is not a prediction. It says where the leverage sits, not what will happen.

Tell us where we are wrong

Every rating carries its sources so you can check it, and both answers so you can see the reasoning rather than only the conclusion. Where a source does not say what we say it says, where a lobbying claim is stronger than the evidence behind it, or where we have missed the actor who actually decides an outcome, we would rather hear it than not. Corrections change the published rating.

Name something you care aboutSee the dependency maps

Revision 1.2, 13 September 2026. Claude was used to conduct the analysis. The work is human led, and human review is not yet applied to every item.