A shareable model is a version of a techno-economic analysis prepared to be handed to an outside party — investor, partner, diligence team — that preserves the model’s logic and defensible results while withholding the proprietary detail that is the team’s competitive advantage. It is the deliberately abstracted artifact that lets a model travel without giving away the trade secret.
Body
What it is. The working model contains the team’s IP: exact recipes, the novel step’s parameters, raw vendor quotes. The shareable version is a transformation of that model which keeps the structure, the headline number, the drivers, and the basis, but replaces the sensitive interior with abstractions. It is a separate artifact derived from the internal model for external use — not the working file with a few cells deleted.
The techniques it composes. A shareable model is built from three moves, each a tool in its own right:
Banding sensitive numbers — report a range or order of magnitude instead of an exact value, so the figure still informs without revealing (the countermeasure to a black box’s interface leaking).
Aggregating line items — roll detailed sub-costs into a single block so individual proprietary costs are not exposed.
What it must keep. For the shared model to be worth sharing it has to stay honest and defensible. The mass and energy balances must still close; the headline number and its basis must be intact; and the drivers must remain visible enough for the recipient to run their own sensitivity. Abstraction that breaks a balance, hides the drivers, or moves the answer is not sharing — it is a different, and misleading, model wearing the original’s conclusion.
The central tension. Every abstraction trades disclosure against usefulness. Hide more and the IP is safer but the model is less interrogable and less defensible; hide less and it is more credible but leaks more of the secret. A shareable model lives on that frontier — the most abstraction that still leaves a result the recipient can sensitize and defend. That balance is what makes building one a judgment call rather than a mechanical redaction.
Limits & typical error
Over-abstraction hollows out the model. Band and black-box so heavily that no driver is visible and the headline cannot be sensitized, and the recipient is left with nothing to evaluate; a model too protected to be tested is also too opaque to be believed.
Under-abstraction leaks the IP. Leaving exact proprietary numbers, raw quotes, or a too-revealing black-box interface exposes the very secret the exercise exists to protect — and the leak is usually in an unguarded place, a single stream composition or one line-item cost, not the obvious headline.
The shared and internal models drift apart. Maintaining two versions invites inconsistency: a correction made in the working model but not propagated to the shared one leaves the shared figure no longer matching the real analysis, so its provenance silently breaks even though every number looks sourced.
Abstraction that changes the answer crosses into misrepresentation. Banding to the favorable end of a range, or aggregating to bury an inconvenient cost, is no longer protection — honest abstraction preserves the central result and only blurs its detail; abstraction that shades the result is a false version of the model.
A band still implies a basis. Reporting ”~$800/t ±30%” does not free the figure from its boundary, capacity factor, and gross/net label; a banded headline on an unstated basis is as non-comparable as a precise one. Hiding the exact value does not excuse stating the basis.
Aggregation can conceal a balance break. Rolling line items together can hide that the underlying mass and energy balance no longer closes; aggregation has to sit on top of a model that still balances, not paper over one that does not.
See also
Black-boxing — the unit-op abstraction technique a shareable model applies to its proprietary steps.
Provenance & defensibility — the property a shared model must retain to be worth anything under diligence.
The headline number — the figure (with band and basis) a shared model must keep intact and sensitizable.
False precision — the discipline a banded figure must still respect, and the failure of shading a result while abstracting it.
Driver / key parameter — the inputs that must stay visible for the recipient to stress-test the shared model.
System boundary — the basis a banded headline still has to declare.
TEA in a pitch deck — the more compressed, persuasion-oriented sibling artifact, distinct from the diligence-oriented shared model.
Mini-example
Green ammonia, shared with an investor. The internal model holds the proprietary electrolyzer stack’s exact efficiency and a raw vendor capex quote. The shareable version applies the three moves: it black-boxes the stack (water + ~10 MWh/t → 0.18 t H₂/t at a banded cost), bands the headline to “$800/t ±~30%, power-to-ammonia basis, capacity factor ~0.90” rather than $805.27, and aggregates the proprietary balance-of-plant costs into the single installed-ISBL line — while keeping the drivers (power price, electrolyzer intensity, capex) visible so the investor can run their own sensitivity. The structure, the headline, the basis, and the drivers all survive; the stack recipe and the raw quote do not appear.
Separately, to show abstraction that changes the answer: banding the headline down to ”<$700/t” by quietly switching to the best-case capacity factor is not protection but a favorable-end misrepresentation — the abstraction moved the result, which a shared model must never do. Blurring detail is allowed; shifting the conclusion is not.