Open methodology · v1.7 · June 2026

How The Rake scores companies

Updated June 3, 2026. v1.7 introduces flag clamping: flags still adjust the spectrum position (±1 or ±2), but the final score is now clamped to the band determined by the structural (pre-flag) base score. This means flags move a company within its label band but can no longer push it into a different label. The change ensures the label reflects what the dimension scores measured, while flags provide nuance within that range.

View methodology.md on GitHub →

The Rake scores tech companies on how their revenue models relate to the interests of the people who use them. Every score is derived from public information, applied through a consistent framework, with sources cited and confidence levels shown. We are not publishing legal findings or making accusations. We are applying a transparent system to public facts.

If you disagree with a score, this methodology is your starting point.

The five profile labels

Every company receives a spectrum position from 0–100 and a corresponding profile label. The spectrum position is calculated mechanically from the dimension scores and flags — it is not an editorial judgment. The label is determined automatically by where the spectrum position lands.

0–20

Adversarial

This product is designed to work against you — it makes more money when you fail, stay confused, or can't leave.

21–40

Extractive

This product takes more value from you than it delivers — engagement, data, or money are harvested in ways that don't serve your interests.

41–60

Compromised

This product has genuine value but isn't fully on your side — trade-offs exist between what's good for you and what's good for the business.

61–80

Aligned

This product makes money when you get value — its incentives and your interests point in the same direction.

81–100

Principled

This product actively prioritises your interests, sometimes at cost to itself — the way it's built reflects values, not just incentives.

Profiles are always displayed with their spectrum position — e.g. Compromised (43) and Compromised (58) are both Compromised, but the number makes the difference visible.

How the spectrum position is calculated

Each scored dimension is mapped to a 0–100 value using fixed band midpoints: a score of 1 maps to 17, a score of 2 to 50, a score of 3 to 83. N/A dimensions are excluded. The base score is the average of all mapped scored dimensions.

Flags then adjust the base score. A flag on an anchor dimension adjusts by ±2 points; a flag on a non-anchor dimension (or one not mapped to any dimension) adjusts by ±1 point. The total flag adjustment is capped at ±8 in either direction. The final score is then clamped to the band determined by the base score — flags can move the position within a band but cannot change the label. The label is always determined by the structural base score.

Anchor dimensions ±2 per flag

Incentive alignment

Captivity

Engagement extraction

Algorithmic accountability

Non-anchor dimensions ±1 per flag

Revenue clarity

Multi-sided tension

Ownership pressure

Anchor status has no effect if a dimension is N/A — it drops out of the calculation entirely. Flags on N/A dimensions still apply at ±1.

The seven dimensions

Each company is scored across seven dimensions on a 1–3 scale. Dimensions that genuinely do not apply are marked N/A and excluded from the profile calculation.

Revenue clarity Can a user immediately understand how this company makes money?

Incentive alignment Does the company make more money when users succeed, or when they stay longer, spend more, or remain confused?

Captivity How easy is it to leave? Is data portable? Is cancellation straightforward?

Engagement extraction Is the product engineered to defeat the user's ability to disengage, or does engagement reflect genuine user value?

Multi-sided tension When the interests of different customer groups conflict, whose side does the company take?

Algorithmic accountability Does the company take responsibility for what its systems surface and amplify?

Ownership pressure Who owns this company, and what structural pressures are they under to extract value from users? (Trajectory dimension)

The scoring rubric

Each dimension is scored 1, 2, or 3. There are no half-points. A score of 1 means the product is actively working against the user on this dimension — a present-tense harm, not a missed opportunity or future risk.

1. Revenue clarity

Can a user immediately understand how this company makes money?

1Misleading. Revenue streams are actively disguised. Hidden fees surface at checkout; data monetisation is structurally concealed; "free" positioning deliberately obscures what is actually being sold.

2Partial. Primary model is broadly known or findable, but secondary streams are not surfaced to users.

3Clear. Revenue model stated plainly, including any secondary streams. A user can understand the full picture without research.

2. Incentive alignment

Does the company make more money when users succeed, or when they stay longer, spend more, or remain confused?

1Opposed. The business model structurally requires user failure or dependency. Revenue grows when users don't achieve their goal.

2Neutral. Revenue is not tied to user failure, but the company doesn't benefit from user success either. Flat subscriptions are the clearest example.

3Aligned. The company makes more money when users achieve outcomes — commission on successful transactions, outcomes-based pricing, or direct-purchase models.

3. Captivity

How easy is it to leave? Is data portable? Is cancellation straightforward?

1Deliberately trapped. Lock-in is engineered by design. Cancellation flows use dark patterns; data is locked in proprietary formats; deletion requests don't result in actual deletion.

2Friction. Leaving is possible but inconvenient. Data export exists but is incomplete; cancellation is accessible but not prominent.

3Portable. Data is exportable in usable formats; cancellation is self-serve and immediate; clear renewal notice; no meaningful penalty for leaving.

4. Engagement extraction

Is the product engineered to defeat the user's ability to disengage, or does engagement reflect genuine user value?

1Weaponised. The product is deliberately engineered to defeat the user's own ability to disengage. Psychological mechanisms are exploited to override user judgment — not to deliver value, but to extract time and attention.

2Extractive. Engagement mechanics exist and are tied to revenue, but stop short of deliberately exploiting psychological vulnerabilities.

3Value-driven. Engagement reflects genuine user value. The product is used when it is useful and does not employ mechanics designed to extend use beyond that.

5. Multi-sided tension

When the company serves more than one customer group, whose interests take priority when they conflict?

1Users subordinated. The company's primary commercial relationship is with a party other than the end user, and user interests are structurally subordinated. When interests conflict, the paying customer wins by default.

2Interests unresolved. The company serves more than one customer group and nominal protections exist, but the track record when commercial interests conflict is ambiguous or unresolved.

3Users defended. The company has a documented track record of siding with users when interests conflict — including at measurable commercial cost to itself.

6. Algorithmic accountability

Does the company take responsibility for what its systems surface and amplify?

1Unaccountable. The platform's algorithmic systems actively surface or amplify harmful content, and the company has chosen not to intervene despite awareness.

2Partial accountability. The company acknowledges responsibility and has moderation policies, but these are inconsistently applied or have a documented record of failing to prevent significant harm.

3Accountable. The company takes clear and documented responsibility for what its systems surface. Ranking signals are disclosed; paid placement is clearly labelled; moderation policies are published and consistently applied.

7. Ownership pressure Trajectory

Who owns this company, and what structural pressures are they under to extract value from users?

1Maximum pressure. PE-owned; public company with ad-supported model under quarterly earnings pressure; late-stage VC with active return pressure; or parent company that has publicly redirected investment away from the product.

2Moderate pressure. Public company (subscription or mixed model); mid-stage VC; or founder-controlled public company with some insulation from pure shareholder pressure.

3Low pressure. Bootstrapped and profitable; co-op or employee-owned; nonprofit or mission-locked structure; early-stage with genuine long runway.

Flags

Flags surface the most significant positive, negative, and trajectory findings from the research — specific documented incidents, structural commitments, or directional signals that are important enough to affect how a dimension score should be read or how it might change in a future version.

Tied to a specific dimension where the evidence allows — some trajectory flags may not map cleanly to a single dimension, but the connection to user interests must be made explicit
Significant enough to matter — if it is not significant enough to affect how a dimension score is read, or to qualify how that score might change, it is not a flag
Grounded in a cited source — flags require at least one source, assessed preferred
Specific in the report — flag categories are defined in the methodology; flag instances in reports are always named, dated, and sourced

Flags come in three types:

Negative — A documented incident or structural problem. Adjusts the spectrum position down by 1 or 2 points depending on dimension anchor status.

Positive — A documented commitment or action that genuinely distinguishes a company from the norm. Adjusts the spectrum position up by 1 or 2 points depending on dimension anchor status.

Trajectory — A directional signal that does not yet affect the current score but could affect a future one. Must point clearly toward a better or worse outcome — vague uncertainty does not qualify. Trajectory flags do not adjust the spectrum position.

Flags do not modify individual dimension scores. They adjust only the final spectrum position, through the formula described above. A flag is what makes the analyst's reasoning legible — particularly when a documented incident is what distinguishes a score of 1 from a 2, or a structural commitment is what earns a 3.

Full flag categories and trigger conditions on GitHub →

Confidence scoring

Every source is classified as either Assessed (primary sources: company websites, filings, terms of service, pricing pages, official statements) or Inferred (secondary sources: journalism, logical deduction from business structure, community reports). The confidence split is displayed per dimension and as an aggregate.

High inferred ratios are themselves editorial information. A company that is hard to score confidently is exhibiting opacity — and that is worth surfacing.

High confidence 70%+ assessed Well-supported. Suitable for citation.

Medium confidence 40–70% assessed Reasonable but has meaningful inferred components. Treat as informed analysis.

Low confidence Below 40% assessed Largely structural inference. Flagged for priority community verification.

How scores are produced

Phase 1

Research collection

An AI agent searches public sources systematically — company websites, filings, terms of service, press, regulatory records. It collects and organises all evidence relevant to the seven dimensions, tags each source as assessed or inferred, and produces a structured research document. The agent does not score anything.

Phase 2

Scoring

A second AI agent takes the Phase 1 research document and applies the scoring framework. It scores each dimension 1–3 (or marks it N/A with a rationale), writes a rationale grounded in the evidence, produces a confidence split per dimension, proposes a profile label and spectrum position, and identifies flags. The agent works only from the Phase 1 document — it does not search the web.

Phase 3

Human analysis and narrative

A human reviews the research and scorecard, checks sources, corrects errors, verifies the calculated spectrum position, and writes the published narrative. This final step is completed by a human, not generated by an AI. The analyst's judgment determines what gets published.

Versioning

Every score is a snapshot. Entries are dated and versioned. When a score changes — due to new information, community input, or company response — a new version is published with a changelog explaining what changed and why. The historical record is preserved.

Ownership changes, acquisitions, and major business model shifts trigger a rescore review. A product that scores Aligned under one owner may score very differently under another.

What The Rake is not

It is not a legal document or regulatory filing
It is not a claim about what companies do in private
It is not a campaign to get companies shut down
It is not affiliated with any company, investor, or advocacy group

It is analysis, applied consistently, in public, with sources. Scores reflect the public record at a point in time. They are not verdicts.

methodology.md on GitHub — fork it, critique it, improve it →