How Factrail works
A verifiable map of reality — what happened, who contributed, and what it did to public welfare.
Factrail is not a blog or a news feed. It is an analytical graph: actors create facts, facts move long-running drivers, drivers move welfare indicators, and each indicator's deviation from its normal line is read as benefit or harm — then traced back to the people responsible. Every serious conclusion is the best verifiable answer at publication time, not a claim to absolute truth.
- Evidence-first
- No source, no fact. Every claim carries its source, confidence, and verification status.
- Traceable, not labelled
- Actors are judged by measured impact on welfare through an explicit chain — never by moral labels.
- Updated, not hidden
- When evidence changes, conclusions are corrected in the open; past revisions stay visible.
- Method open, calibration not
- Every factor and rule is disclosed; the exact calibration is not — to stay honest and hard to game.
The causal chain
One chain, end to end
Responsibility and impact flow along a single path — and only along it. Hover any node to see what it means.
- Actor
- Fact
- Driver
- Indicator
Actor → Fact
via Contribution
direct · indirect · complex — with a size and a direction
Fact → Driver
via FactDriverImpact
initiating · strengthening · neutralizing · weakening
Driver → Indicator
via DriverIndicatorImpact
strengthening or weakening, with a strength and a time lag
Indicator → Score
via Normal line
deviation from the normal line, read as benefit or harm
Factrail never stores a direct Actor → Driver or Actor → Indicator link. Those are derived by walking this chain — traceable and reproducible, never a stored shortcut.
The entity model
Six things, each with its own colour
Almost everything on Factrail is one of these six entity types. The colour is consistent across every page.
- Person
A politician, official, business figure or other significant actor.
- Organization
The vehicle through which a person often acts indirectly.
- Fact
A concrete, dated event with a source. Neutral on its own.
- Driver
A sustained process that carries intensity over time.
- Indicator
An axis of public welfare, measured against its normal line.
- Article
Analysis that explains the graph and links out to it.
How the score is built
Many small impacts, summed
An actor’s score is not a hand-set number. It is the sum of many small, individually-recorded impacts — one per traceable path through the chain. Each impact is shaped by these factors. Hover any factor for the detail.
These combine into a single signed impact. An actor’s score is the sum of all such impacts — a positive net effect on welfare raises it, a negative one lowers it.
The method is open; the exact weights are not.
Every factor above is disclosed, along with the direction it pushes. What Factrail does not publish is the precise numerical weighting — that lives in a versioned, calibrated formula, is tuned and improved over time, and is recorded with every score. Publishing the exact numbers would invite gaming and would misrepresent tunable parameters as fixed truths. You can see what shapes a score and why, without a recipe to manipulate it.
Guardrails
What the score will never do
A few invariants protect every conclusion. They are enforced in the engine, not left to judgement.
Future facts don’t count
Scheduled or future events never move a score until they actually happen.
Forecasts never feed the rating
Projections live on a separate track; only measured, past reality affects scores.
Weak links shrink impact
Confidence chains multiplicatively — one shaky link discounts the entire path.
Outliers are capped
A single extreme reading can’t dominate a conclusion.
Promise & lie are editorial
Promise-keeping and truthfulness are tracked separately from the welfare score.
Public responsibility
What “Public Responsibility” means
Public Responsibility is an analytical estimate of the social harm attributed to a person under the Factrail methodology — and, where it can be benchmarked, a compensation-equivalent figure that puts that harm on a comparable scale. It is a way to read responsibility for public outcomes, not a verdict.
The compensation equivalent is not a legal debt.
It is not a debt, a fine, a court award, or money legally owed. It is an analytical figure that expresses estimated social harm on a comparable monetary scale, so that very different harms can be weighed against one another. It is never a court ruling or a legal claim, and it is only described in those terms when the underlying record is an actual judicial decision.
How affected people are estimated
Affected people are estimated from the documented record — the populations a fact touched and the share of that harm a person is responsible for. These figures are inherently uncertain: the monetary compensation equivalent is always shown as a range (a low and a high), and affected-people figures are presented as estimates — the headline shows a central estimate — rather than exact counts. The aim is to convey uncertainty honestly, not to imply a precision the evidence does not support.
How organisation responsibility is distributed
When a person acts through an organisation, the chain is explicit: the person, their role, the organisation’s contribution to the fact, the resulting social harm, and the share attributed to the person. Only part of an organisation’s contribution is attributed to any one individual.
Unallocated institutional responsibility is the part of an organisation’s contribution that is not assigned to any named person. It makes clear that not every member of an organisation is personally responsible, and that institutions carry responsibility that individuals do not absorb entirely.
Benchmarks, price year and inflation
A compensation benchmark is a reference value used to translate a unit of harm into a comparable monetary figure — for example, an award scale used in comparable cases. Because money changes value over time, every figure is tied to a stated price year, so amounts are comparable on the same basis rather than mixing values from different eras.
Separating harm from contribution
Public Responsibility represents gross negative impact. Concrete harms — deaths, serious injuries, unlawful detention, severe rights violations — are never reduced by unrelated positive contributions. Positive contributions are presented in their own right elsewhere on the profile; they are not netted against concrete harm.
Limitations and uncertainty
These are estimates, not findings of fact or law. They rest on the available record, carry real uncertainty, and change as evidence changes. Verified figures are shown by default; reviewed disputed estimates are kept separate and never inflate the verified total. Where the record is thin, the estimate says so rather than overstating its confidence.
Normal line & deviation
Measured against a baseline
Every indicator is read against its own normal line. The deviation — above or below — is what counts as benefit or harm.
- Higher is better
- — more is good — incomes, life expectancy
- Lower is better
- — less is good, with a normal level above zero — corruption, poverty
- Zero is norm
- — only zero is acceptable; any amount is worse — crimes, repression
- Dynamic norm
- — the normal level itself shifts over time
- Custom
- — needs its own dedicated logic
Confidence & verification
Three independent dimensions
Factrail keeps evidence confidence, verification, and publication status separate. Visibility and lifecycle are additional independent public gates.
Verification status
Whether evidence has been reviewed or remains publicly disputed.
Confidence level
Epistemic certainty in a claim or a link. On facts and edges — never on people.
Publication status
Editorial progression from draft to publication. Visibility and lifecycle still gate whether a published item is public.
From source to conclusion
How facts and articles are produced
Every published conclusion is built from cited public sources, reconciled against the existing graph, and reviewed before it goes live. Nothing reaches the public graph unreviewed.
- 01
Discovery
Topics surface from monitored public sources or from an editor’s request, and become candidates to investigate.
- 02
Research
Each candidate is investigated against public sources and assembled into a structured draft: facts, the actors and contributions behind them, the drivers and indicators they move, and a source for every claim.
- 03
Reconciliation
Proposals are matched to the real graph. Known entities are linked; genuinely new ones become drafts — never silently published. Indicators are only ever linked to real, existing ones.
- 04
Review
Every draft passes review before it goes public. An accusation needs at least two independent sources and review. Nothing reaches the public graph unreviewed.
Behind each article, every notable claim is tracked by kind (factual · causal · interpretive · predictive · contextual), salience (major or minor), and validation (supported → unsupported), tied to the exact entities it relies on.
Every published conclusion follows these same standards — applied consistently, with no hidden exceptions.
Forecasts
The future, as a forecast — never as fact
Some questions are about what may come. Factrail answers them as explicit scenarios with stated probability — kept strictly apart from the measured record.
A forecast is a scenario
Each carries an explicit premise and a probability level (low / medium / high). Projected values come with a range — a low and a high bound — never false precision.
Forecasts never move a score
Projections live on a separate track. Only measured, past reality — readings of events that have actually happened — feeds the rating. A forecast informs the reader; it never quietly becomes a verdict.
Reading the colours
Trust colours vs. entity colours
A second colour language sits on top of the entity colours: it shows whether something helped or harmed.
Promise kept — green
Broken / lie — red
Net welfare impact (the score)
Entity colour says what kind of thing you are looking at — a person, a fact, a driver, an indicator.
Trust colour says whether it helped or harmed — green for benefit and kept promises, red for harm and broken ones. The two are deliberately distinct.
Corrections & limits
When the evidence changes, so does the conclusion
A Factrail conclusion is the best verifiable answer at publication time — not a final verdict. Uncertainty is shown, not hidden.
Corrected in the open
When new evidence arrives, a conclusion is updated — not quietly rewritten. Past revisions stay visible, so the record shows how an assessment changed and why.
Disputed is not verified
When sources genuinely conflict, the item is marked disputed and the conflict is shown rather than resolved by fiat. A disputed conclusion is never presented as a verified one.
Limits, stated plainly
Factrail does not claim to capture every cause or to predict the future with certainty. Causation is modelled, not proven; confidence is stated, not assumed. Where the map stops, it says so.
You can follow any conclusion back along its chain — from an actor's contribution to a fact, the driver it moved, and the welfare indicator that shifted — and read the source, confidence, and verification status at each step. That trail is the point: not a verdict to take on faith, but a path you can check.