A retro from a live gas engagement on how the clean 7-step SOP meets the mess of real work — and the iteration discipline we added because of it.
Our Standard Operating Procedure describes a clean seven-step waterfall: frame, decompose, gather, model, reconcile, review, publish. Real engagement work does not move like that. The honest description is: sketch first, then scaffold the governance retroactively.
You build a rough model to understand the problem — and the model itself surfaces the gaps that drive the framing. The framing and the model co-evolve. Trying to fully frame before touching the model is how analysis stalls. So we added an explicit Exploratory vs Production mode to the method: exploration is permitted and encouraged for ad-hoc analysis, but it must never feed a formal deliverable without being promoted — reviewed, provenance-stamped, pinned to a published assumptions snapshot.
A real learning that has nothing to do with code: partners hear “I changed X” as “you framed it wrong.” Iteration reads as criticism unless you name it as iteration. Make the mode explicit — “this is still exploratory” — and the same change lands as progress rather than a rebuke.
One-lever “what-ifs” go to the sensitivity tornado — they never quietly become a new scenario. Named scenarios are governed centrally; a stray sensitivity dressed up as a scenario is how two models end up running “the same world” with different numbers. Keep the exploration loud and the production layer strict, and the messy middle stops being a liability.
Internal field note — sanitised for cross-team learning. More field notes