Concept
Provenance is the property that every number in a model is traceable to where it came from — a citation, a quote, a comparable process, or an explicitly stated assumption — and defensibility is the resulting ability of the result to withstand outside scrutiny. A model is only as credible as the provenance of its weakest load-bearing figure.
What provenance is. Each input carries a record of its origin: a literature value with a citation, a vendor quote, a figure borrowed from a reference / comparable process, or a modeling assumption stated as such (“taking ~$X as a round anchor”). A number with provenance can be traced, questioned, and updated when a better source appears; a number without it is a floating assertion that no one — including its author later — can evaluate.
Why it is a communication property. A model travels to people who did not build it — investors, partners, a technical diligence team. They cannot re-derive it, so they test it the only way they can: by checking whether each load-bearing number is defensible when challenged. Defensibility is provenance made external — being able, when asked “where does this come from?”, to name the source and its basis. The handbook’s own data-sourcing and validation discipline (flagging every unsourced figure until it is checked) is provenance enforced during the build, before anyone outside ever asks.
Not all provenance is equal. A vendor quote, a peer-reviewed value, and a number from a press release sit at different tiers of the source hierarchy. Defensibility tracks the tier of the weakest load-bearing input, not the average across the model: a result resting on a single press-release figure is exactly as defensible as that figure, however well-cited everything around it is.
Provenance is traceability, not correctness. A well-sourced number can still be wrong, and a guessed one can be right. Provenance does not make a figure true — it makes it auditable: traceable to a basis a reader can find, evaluate, and challenge. This is distinct from false precision, which concerns the resolution a number is reported at; provenance concerns whether the value can be traced to a source at all. A figure can be precise and unsourced, or rough and fully traceable.
Defensibility concentrates on the drivers. Because the answer rides on a few high-leverage inputs, the defensibility of the result is set by the provenance of its drivers. A driver carried on a low-tier source is the model’s weakest defensible point; a non-driver’s provenance, however thin, barely affects whether the conclusion holds up.
Green ammonia’s $800/t result rests on a few load-bearing numbers, each with different provenance. The electricity intensity ($1.0bn) is an order-of-magnitude figure awaiting a quote. A diligence reader can challenge each on its own terms — the intensity against a vendor spec, the price as a stated assumption whose range must be justified, the capex as a figure still awaiting a quote. The defensibility of the $800/t headline is set by the weakest of these — the price assumption and the capex anchor — not by the well-grounded intensity.10 MWh/t) traces to electrolyzer efficiency data — a reasonably defensible tier; the electricity price ($40/MWh) is a stated round market anchor, defensible only as an assumption with a range, not as a contract; the total capex (
Separately, to show provenance of the value without provenance of the condition: a ~$540/t clean-hydrogen credit cited to the policy’s headline rate, but not to its eligibility rules (the carbon-intensity tier, the hourly-matching requirement), is sourced for the number and undefended on whether the plant qualifies. It collapses under the first diligence question that asks not “what is the rate?” but “do you actually meet the conditions to earn it?”