The evaluation set is the asset that survives the kill
Killing a prototype destroys the system and preserves the encoded definition of correct behaviour. Salvage is a forward comparison, not a rescue justified by what was already spent.
4 min read
Singular State / Perspectives
Perspectives
Practical thinking on what to build, how to make it work, and what it takes to own it.
Excel refresh timing and the meaning of a financial total
How refresh completion, calculation order, and business cutoffs determine what a workbook total represents.
3 min read
Extend, kill, or transfer
The decision the operate phase exists to produce. Three named outcomes, three others that get hidden inside them, and a default that applies when the evidence is missing.
5 min read
Extracting a VB6 function across a COM boundary
Choosing a coherent VB6 extraction boundary while preserving COM consumers, transactions, and deployment.
3 min read
Falsifiable success criteria
A success criterion is falsifiable when a stated observation would end the project. Most criteria have two states and need three, because inconclusive is the result that actually occurs.
5 min read
Fixed-price modernization when discovery is incomplete
Defining priced obligations and acceptance evidence when legacy behavior and dependencies are still being discovered.
3 min read
From FoxPro DBF files to a relational database: preserving meaning
Combining FoxPro file interpretation, target schema design, client connectivity, and business reconciliation.
4 min read
Golden, adversarial, and regression sets
Three purposes, not three folders. One case can carry all three tags and still be one case family, and three passing tags are not three independent successes.
5 min read
Guardrails are an architecture, not a filter
Controls belong at the boundary that can enforce them. Authorisation happens in the downstream system, not in the model, and autonomy level sets how much each layer must carry.
5 min read
How a strangler migration replaces a system incrementally
A strangler migration routes selected responsibilities from an existing implementation to replacements. Completion requires removing the old dependencies, not merely adding a facade.
4 min read
Idempotence: what a repeated operation is allowed to do
Idempotence concerns the intended effect of repeated requests. It is distinct from identical responses, broker acceptance, and exactly-once business execution.
4 min read
Incomplete inventories can favor the systems that are easiest to document
How missing evidence can shape portfolio selection before substantive risk and value are compared.
3 min read
Full-stack IP ownership, work-for-hire, and licence-back
Owning the copyright is not holding the file, and holding the file is not being able to run the system. Calling a contract work made for hire does not make it one.
5 min read
Kill criteria are a political instrument
A precommitted criterion works only where the sponsor survives triggering it. Escalation of commitment is the mechanism it has to defeat, and it was demonstrated in the people who made the original choice.
5 min read
Write the kill criteria before the demo
Recorded conditions for stopping, agreed before anyone sees the evidence they govern. Six trigger families, and the rule that a stop decision and a finding about value are separate statements.
5 min read
Liability after the last invoice
Which obligations survive handover, which insurance responds to a model's output, and why detection after transfer establishes nothing about when the defect arose.
5 min read
LLM-as-judge, and what calibrating it against humans actually requires
A judge is a measurement component with order sensitivity and no established authority. Swap the presentation order, and a judge that picks the first answer both times has produced two conflicting verdicts.
4 min read
Lotus Notes migration must preserve execution authority
Mapping Notes workflow state together with agent identity, authority, triggers, and in-flight work.
3 min read
Low-code applications still have a deployment lifecycle
Low-code packaging, dependencies, query limits, and restoration remain software lifecycle concerns.
3 min read
Masked data, real conclusions: the fidelity gap in your evidence
Every de-identification transform removes signal the model was going to use. What a pilot on treated data can and cannot establish, stated as a rule rather than a caveat.
5 min read