Writing a scope doc with success criteria before seeing data
I drafted a scoping document for a small proof-of-concept test before a client commits to a full build. The whole point was fixing success criteria before looking at any results.
Modeling on a previous audit plan
I read an existing audit plan document first and used it as a style reference. It had fixed scope, explicit in/out boundaries, and decision criteria written down before any evidence was seen. I copied that discipline.
Three outcomes instead of two
The scope defines pass, fail, and inconclusive as separate outcomes. Inconclusive is defined explicitly so a weak result cannot be quietly rounded up to a pass.
Baseline set to majority-direction drift
For a drifting asset, beating 50% is not evidence of anything. The naive baseline is predicting the asset's majority direction over the whole period every time, ignoring the narrative. The pass threshold is about 15 percentage points above that baseline, plus a robustness check that removes the single largest contributing event.
Roles split stated bluntly
I own scope, structure, method, success criteria, and the analysis framework. The client's financial expert owns every market judgment: which asset class a narrative implicates, direction, horizon, what "priced in" means, and whether the mapping logic is financially coherent. I wrote a sentence saying explicitly that I am not making market calls and am not qualified to.
The financial expert is treated as a hard dependency. If the expert is unavailable the proof cannot run as designed.
Detection layer excluded
The proof tests only the translation layer. Events are hand-picked, not auto-detected. If detection and translation are tested together, a null result cannot be attributed to either.
No price, no timeline, no build commitment
The document contains none of these. The final section states each exclusion plainly so nothing is assumed.
The document has not been sent to the client yet. It goes to the user for review first.