Writing

40 essays on measuring AI systems honestly — what a benchmark has to be, whether a model can grade one, what the statistics will and will not support, and the defects we found in our own code while building the thing.

Most recent · 30 August 2026

Subscribe via RSS →

SERIES A

evaluation foundations

11 August 2026

What a benchmark actually is — and why yours probably isn't one

Three properties separate a benchmark from a demo. Most internal AI evaluations have one of them.

Level 1 · 5 minS01 · Foundations
11 August 2026

Can you trust an AI to grade an AI?

A practical guide to judge calibration: why raw agreement lies, what kappa corrects for, and why you gate on a lower bound.

Level 3 · 7 minS02 · The Judge
12 August 2026

The benchmarks are broken — and the audits are arriving late

What 2026's benchmark audits found, why every one of them was retrospective, and what gating at authoring time would have caught.

Level 2 · 6 minS03 · Broken Benchmarks
12 August 2026

Evaluating an agent is a different discipline

Four things need grading, not one — and the scaffold is often doing more work than the model.

Level 3 · 6 minS04 · Agents Under Test
13 August 2026

Your failures are your training data

A score isn't a plan. A failure map is — and the fastest way to get one is to cluster by the criterion that refused, not by topic.

Level 3 · 6 minS05 · From Failure to Fix
13 August 2026

How to write an evaluation that holds up

Scope from consequence. Weight by cost, not frequency. And write the oracle first — it's where you find out your experts disagree.

Level 3 · 7 minS06 · Writing the Test
14 August 2026

The statistics of evaluation: what your numbers actually support

Measure your noise floor first. Then intervals, multiple comparisons, pass@k, and where the variance really comes from.

Level 4 · 7 minS07 · The Statistics
14 August 2026

Red-team your rubric, not just your model

Reward hacking is a specification defect. Here are six cheat probes to write, and the afternoon that catches most of them.

Level 4 · 6 minS08 · The Adversary
15 August 2026

Evaluation doesn't end at launch

Offline and online answer different questions. Put the benchmark in CI, pin the model version, shadow before canary, and write postmortems back into the suite.

Level 3 · 6 minS09 · Evals in Production
15 August 2026

Trust that doesn't require trusting anyone

What a regulator actually asks, why provenance has to be recorded at authoring time, and why a result nobody outside can reproduce is a claim rather than a measurement.

Level 3 · 6 minS10 · Trust & Governance
16 August 2026

You're six months into an AI programme. Is it working?

The pilot went well, a team uses it daily, and nobody can tell you whether it's safe to widen. That gap isn't capability — it's evidence.

Level 1 · 6 minS11 · Your AI System
16 August 2026

Eight workflows, eight failure maps

The failure modes that actually hurt — by use case — and the expert data that fixes each one.

Level 2 · 9 minS12 · By Use Case
SERIES B

extended

17 August 2026

The evaluation maturity model

Five stages from vibes to verifiable. Most organisations are at stage one and believe they're at stage three.

Level 2 · 6 min
17 August 2026

How to run a model bake-off that produces a decision

Most vendor evaluations produce a slide. Here's the protocol that produces a defensible choice.

Level 2 · 7 min
18 August 2026

Fourteen questions to ask a vendor about their benchmark

A buyer's checklist. Most vendors can answer four of these. The gap tells you what you're actually buying.

Level 2 · 6 min
18 August 2026

Should you build your evaluation in-house?

An honest decomposition. Some of it you must build. Some of it you shouldn't. The line is not where most teams draw it.

Level 3 · 6 min
19 August 2026

RAG evaluation is two problems, and most teams measure one

Retrieval failure and generation failure look identical in the output and need completely different fixes.

Level 3 · 6 min
19 August 2026

Evaluating multi-agent systems

When work is split across agents, failure attribution becomes the hard problem — and the handoff is where it lives.

Level 4 · 6 min
20 August 2026

The economics of evaluation

What it costs, what it saves, and why the second workflow is the number that matters.

Level 2 · 6 min
20 August 2026

Running human annotation that doesn't fall apart

Your experts are the scarcest resource in evaluation. Here's how to spend them so the labels are worth having.

Level 3 · 6 min
21 August 2026

Synthetic evaluation data: when it works and when it lies

Generation is fine for volume and dangerous for ground truth. The line is sharper than most teams draw it.

Level 3 · 6 min
21 August 2026

Benchmarks with a date on them

The questions enterprises actually care about are time-indexed. A live API call makes them unmeasurable. Capture and replay are separable.

Level 4 · 6 min
22 August 2026

A starting checklist for evaluation in regulated industries

What examiners actually look for, and the artefacts worth having before anyone asks.

Level 3 · 6 min
22 August 2026

The 30-day evaluation

What you can genuinely accomplish in a month, in order, without a platform or a budget line.

Level 2 · 6 min
SERIES C

from the workbench

23 August 2026

The number nobody publishes

Every benchmark reports the fraction of tasks a model got right at least once. Almost none report the fraction it gets right every time. For anything going to production, the second number is the one that matters.

Level 4 · 6 min
23 August 2026

An empty checklist passes every audit

The most dangerous bug in an evaluation system is the one where silence and success are the same value. We found nine of them in one codebase, each behind a green test suite.

Level 3 · 7 min
24 August 2026

Test the grader, not just the model

Your evaluation harness is production code that nobody evaluates. Five properties — stated once, enforced against every grader you have — catch defects that thousands of example-based tests miss.

Level 4 · 7 min
24 August 2026

Your traces are a benchmark you haven't compiled

You already have thousands of recordings of real work. Turning them into tasks is mostly mechanical — except for the one part that isn't, which is where every naive attempt fails.

Level 3 · 7 min
25 August 2026

When the simulated user leaks

Half of real agent work is getting information out of a person. Simulating that person with an LLM is the obvious move, and the obvious implementation hands the model the answer key.

Level 3 · 6 min
25 August 2026

Say what you're predicting before you know if you're right

The question that decides whether your evaluation was worth building is whether a better score predicted a better outcome. You can only answer it if you wrote down the claim first.

Level 4 · 6 min
SERIES D

the refusals

26 August 2026

Audit the passes, not just the failures

Every evaluation team reviews the failures and ships the passes. But the pass is the claim you are selling, and a pass nobody interrogated is not evidence that anything worked.

Level 4 · 8 min
26 August 2026

Every number should resolve to a receipt

A figure you cannot click through to the query that produced it is a claim, not a measurement. The fix is to make the evidence executable by the reader, and to diff the prose against what it actually printed.

Level 3 · 8 min
27 August 2026

How do you know the gate would catch it? Seed the corruption

A quality gate with no measured detection rate is decoration. The only way to learn what yours catches is to inject the defects it claims to stop, at known density, and count.

Level 4 · 8 min
27 August 2026

The grader needs a grader — and permission to refuse

If a model is scoring your evaluation, its agreement with human judgement is a measured quantity with a confidence interval. And when there is not enough data to measure it, the honest output is a refusal with a name on it.

Level 4 · 8 min
28 August 2026

An unverified oracle is not an answer key

If the reference solution has never been run through the graders that will score the model, you do not have an answer key — you have a plausible document. And a closed-book claim enforced by reading the command line is not a boundary.

Level 4 · 8 min
28 August 2026

Silence is never success

The zero that means no measurement, the null that gets formatted into a report, and the security setting a backend accepts and never enforces. Three shapes of the same defect, and the rule that closes all of them.

Level 3 · 8 min
SERIES E

worlds and evaluation systems

SERIES F

evaluation immune systems