Credibility
Agency is the ability to influence changes of state. Credibility is the proof you've earned that ability.
The Book records every collision. Credibility is what survives when you make the book legible. Not reputation — what people say. Not authority — what a title says. Credibility: did you do what you said you'd do, measured across time?
The Metric
Credibility = commitments kept / commitments made
| Signal | What It Measures | How You Know |
|---|---|---|
| Delivered on time | Reliability | Promise vs actual date |
| Quality matched spec | Competence | Acceptance criteria met |
| Referred others who delivered | Network quality | Referral conversion + satisfaction |
| Paid on terms | Financial integrity | Invoice vs payment date |
| Communicated proactively | Respect for others' time | Updates before being asked |
The ratio is brutal. A person who makes 100 commitments and keeps 95 has a 0.95. A person who makes 10 and keeps 10 has a 1.0. The second person is more credible — but the first has more data. Both dimensions matter: the ratio AND the sample size.
Three Layers
| Layer | Question | Verification | Book Entry |
|---|---|---|---|
| Identity | Are you who you say? | Registry, KYC, attestation | Existence confirmed |
| Capability | Can you do what you claim? | Track record, portfolio, proof | Flow — attention invested |
| Integrity | Will you do what you promise? | Commitments kept over time | Alignment — actions match words |
Identity is table stakes. Capability is expected. Integrity compounds.
The Agent Parallel
Human agents and AI agents both need credibility. The same ladder applies:
| Question | Human | AI Agent |
|---|---|---|
| Did you complete the task? | Deliverable shipped | Task output verified |
| Did it match the spec? | Acceptance criteria met | Quality gates passed |
| Did you improve the template? | Better process for the next person | Legacy rule applied |
An agent with 100 successful runs and 0 manual interventions has earned L4 credibility. A human who delivers 10 projects on time has earned the same. The ledger doesn't care what kind of agent you are. It cares whether you kept your word.
The Graph
Credibility isn't individual — it's relational. A recommends B. B delivers. C trusts A more.
The referral chain IS the credibility graph. Each successful referral strengthens every node in the chain. Each failure weakens them. This is The Book made computable — the invisible ledger rendered as a graph where edges are kept promises and nodes are agents who earned their position.
| Graph Element | What It Represents |
|---|---|
| Node | An agent (human or AI) with a credibility score |
| Edge | A kept promise between two nodes |
| Weight | Number of successful interactions |
| Cluster | A trust network — people who vouch for each other |
| Orphan | No edges — no credibility data, not necessarily untrustworthy |
Predictions as Evidence
Credibility is not just about keeping promises on tasks. It's about the quality of your predictions — plans are predictions with deadlines.
| Element | What it measures | How it builds credibility |
|---|---|---|
| Forecasting | Can you see the future before others? | Correct predictions over time = earned conviction |
| Time horizon | How far out can you predict accurately? | Longer accurate horizons = deeper credibility |
| Conviction scoring | How confident were you, and were you right? | HIGH conviction + correct = earned trust. HIGH conviction + wrong = credibility cost |
| Kill criteria | Did you name what would make you wrong? | Naming failure conditions in advance = intellectual honesty |
| Update frequency | Do you revise when evidence changes? | Bayesian updating = compound credibility. Stubbornness = credibility drain |
A plan worth believing in has three properties:
- Specific predictions with resolution dates — not "this will work" but "3 inbound conversations by Month 4"
- Conviction tags — HIGH/MEDIUM/LOW/NONE on every claim, so others can calibrate your confidence
- Kill signals — what would make you abandon the plan, stated before you start
The BerleyTrails feedback loop is this pattern in practice: five KPIs, each with a target, each with decision-action processes for when results are bad or good. The venture's credibility grows not from succeeding — but from predicting what success and failure look like, then measuring honestly.
The Score
Over time, the prediction ledger produces a credibility score:
Credibility Score = (correct predictions / total predictions) × average conviction weight
A person who makes 10 HIGH-conviction predictions and gets 8 right has a score of 0.80 × HIGH. A person who makes 50 LOW-conviction predictions and gets 40 right has a score of 0.80 × LOW. Same accuracy. Different credibility. Conviction is the multiplier — it separates those who hedge everything from those who put their reputation on the line.
Context
- The Book — The invisible ledger credibility makes visible
- Agency — Credibility is what makes agency real
- Character — The foundation credibility is built on
- Predictions — The game credibility is scored on
- Forecasting — The discipline of seeing the future
- Behavioural Biases — What distorts your predictions
- Goodwill — Generosity without expectation that compounds into trust
- Data Trust — How credibility gets measured in systems
- Alignment — The audit: do the books match reality?
Questions
If credibility is commitments kept divided by commitments made, what happens when you stop making commitments?
- Is a person with 10/10 commitments kept more credible than one with 95/100?
- When an AI agent earns L4 credibility through receipts, is that the same kind of trust you'd give a human with the same track record?
- The Book says loyalty is "commitment held through difficulty" — is that the same as credibility, or does credibility require something loyalty doesn't?
- What's the minimum number of kept commitments before credibility compounds rather than accumulates?