Foray 247

Business project benchmark

End

Find out whether successive AI models become more useful at answering ordinary business-project consulting questions. Hold a contained project case and twenty questions steady; retain answers so readers can inspect improvement, regressions and the work needed to make an answer usable.

Ways

The methods are open. Start with a fixed input pack, interactive browser answers, independent calculation checks and explicit practitioner judgements. Explore repeated runs, blind pairwise review, repair effort, uncertainty, evidence use and robustness to small input changes where they help answer the End. These are options, not an exhaustive research programme.

Use the latest suitable model at each execution and record its precise identity and conditions. Keep ordinary business questions primary; borrow methods from research when useful. Freeze a version before comparison, distinguish model performance from assistance, and preserve earlier answers that expose a meaningful difference.

First essai

A fictional two-hub repair-and-reuse service: twelve weeks, £240,000 and six staff. Twenty interactive answers address plans, exceptions, coordination, decisions, suppliers, evidence and information systems. This first developed baseline establishes a case and a readable answer set; later controlled runs can test whether newer models do better.

Full inputs and prompts · App007 source prompts · Repeat and review

The original Gimmer app was a prompt bank. Its neuroscience analogies and proposed targets are preserved as provenance, without treating them as measured performance. Research progress and discarded directions belong in the foray’s running notes.