Our method
Published before the findings, so anyone can check whether a number holds.
Sourcing
Every claim traces to a primary source: a filing, a regulator dataset, a court record, an audit, a peer-reviewed paper or a named on-record statement. Databases are used to find a fact and never to confirm one. Where sources conflict, the conflict is shown rather than resolved. Where something could not be confirmed, the page says so.
Evidence grades
| Grade | What it rests on |
|---|---|
| A | Independent audit, court finding or regulatory determination, with first-party statements from both sides |
| B | Filings or first-party disclosure, corroborated by contemporaneous reporting |
| C | Single-source reporting or an uncorroborated first-party account |
| D | Claimed and unverified. Recorded, never relied on |
Outcome tiers
| Tier | What it means |
|---|---|
| T1 Controlled | Holdout, control group, or staged rollout with comparison |
| T2 Before and after | Measured baseline, measured result, no control |
| T3 Estimated | Modelled or extrapolated by the organisation |
| T4 Asserted | Claimed, unmeasured |
Ceilings before estimates
Where no credible measurement exists we do not invent one. We decompose the work into its components, identify the share the system actually touches, and publish the arithmetic maximum. No model can save time on work it does not do. A ceiling with a decomposition behind it is defensible. A point estimate with a confident tone is not.
What we will not publish
- A cut with fewer than thirty responses. It appears greyed, with the count shown.
- An average across a range where the distribution is the finding.
- A chart without its base, its period and its source on the face of it.
- A benchmark default we cannot source.
- A forecast. We measure what is happening. We do not project market sizes.
The six failure modes
Every case in the Failures series carries one primary mode and any number of contributing modes. The modes describe structure rather than technology, which is why they survive changes in the technology.
| # | Mode | What it means |
|---|---|---|
| 1 | No workflow | The system functioned and was never wired to the moment a human decides. Output arrives beside the decision rather than inside it. |
| 2 | No data path | The system cannot reach the data it needs from inside the institution, so humans carry the data to it by hand. |
| 3 | No owner | No named person is accountable for the decision the system was built to change, so no one can authorise the change in practice. |
| 4 | Wrong bottleneck | Visibility, prestige or ambition selected the use case rather than a measured constraint. The system solves something that was not costing anything. |
| 5 | No baseline | Nobody measured what the humans did before, so improvement can never be demonstrated and the programme is defended on anecdote. |
| 6 | No stop condition | No agreed circumstance under which the programme would be halted, so it is extended instead of judged. |
The four maturity bands
| Band | Test |
|---|---|
| Production | Deployed at scale in named enterprises, verifiable from filings or customer disclosures |
| Early production | Live deployments, small numbers, no independent verification of outcomes |
| Pilot | Funded, demonstrated, no production reference we can confirm |
| Announced | Press release only |
Corrections
We correct in public, on the record, with a date, and we leave the correction visible. Superseded versions are preserved rather than silently edited.