Concept

Reference / comparable process

structural

Overview

A reference (or comparable) process is a specific, real, well-characterized process — an existing plant, route, or unit operation — similar enough to the one being analyzed that its known performance and cost data can be borrowed as the starting anchor for the estimate. It is how a new or proprietary process is quantified by analogy to something already built and measured.

Body

Anchoring by analogy. At the maturity anchor’s level, most performance and cost figures are not derived from first principles — they are borrowed from a process that already exists and has been measured. A reference process is that source analog: a real plant, route, or unit operation close enough to the one being modeled that its numbers transfer with adjustment. Because the comparable has actually operated, its figures carry real-world effects — side reactions, true utility loads, practical inefficiencies — that a clean first-principles estimate tends to omit.

What a comparable supplies. Two kinds of anchor. Performance — the conversions and yields a route achieves, and its energy and feedstock use per unit of product. Cost — purchased-equipment costs and overall capex, quoted at the comparable’s size and cost year. The cost anchor is rarely usable as-is: it is taken at the reference’s scale and scaled to the target with the six-tenths rule, so the comparable supplies the reference cost C₁ that scaling starts from.

Matching and adjusting. Using a comparable is a two-step act. First match the target to an analog similar in chemistry, phase, materials, and operating regime; then adjust for the differences that remain — size (via scaling), cost year (via escalation), location, and any change in configuration or materials. The closer the match, the smaller the adjustments and the more the borrowed numbers are trusted; the further the analog, the more the estimate is really resting on the adjustments than on the anchor.

A comparable vs. a class of priors. A reference/comparable process is one specific analog — this plant, this route, this quoted cost. That is distinct from reference-class thinking, the broader estimation discipline of grounding a figure in a whole class of prior cases and their spread. The comparable is the concrete object you anchor to; reference-class thinking is the reasoning habit of anchoring to known cases at all, and of treating any single analog as one draw from a distribution. The two work together — a chosen comparable sits inside a reference class — but one comparable is a point, not a distribution.

Comparable vs. benchmark. A comparable process is an input anchor: its data seeds the model. Benchmarking uses characterized incumbents as an output check — comparing the finished model’s cost or intensity against known processes to test whether the result is plausible. The same incumbent plant can serve both roles, but seeding a model from it and checking a model against it are different uses.

Limits & typical error

Transfer error grows with the distance of the analogy. A comparable’s numbers are only as transferable as the comparable is similar. A close analog (same chemistry, phase, scale band, materials) transfers with small adjustment; a distant one (different feedstock, regime, or maturity) can be off by amounts that swamp the estimate, because the borrowed figure encodes conditions that do not hold for the target. The error is not in the comparable’s data but in the assumption that it applies.
The whole estimate inherits the comparable’s quality and its provenance. Scaling and balances are multiplicative on the anchor, so a reference cost or yield that is itself uncertain caps the accuracy of everything built on it (see the reference-cost point under the six-tenths rule). A figure pulled from a low-tier source — a press release, a rounded literature value — carries that uncertainty into every downstream number; the source hierarchy of the comparable’s data bounds the result.
Hidden basis mismatches travel with borrowed numbers. A comparable’s cost is tied to a cost year, a currency, a location, and a battery-limits scope; its performance is tied to a feed spec and a definition of capacity. Borrowing the number without its basis — an old cost without escalation, a nameplate figure where the target is on operating capacity — imports a silent error that looks like a genuine difference. The number transfers; its basis must transfer with it.
A single comparable understates uncertainty. Anchoring to one analog gives a point estimate with no sense of spread. Real processes of the “same” type vary widely in cost and performance, so one comparable can sit anywhere in that range; treating its figure as the value, rather than one sample, hides how much the answer could move with a different but equally valid analog — the spread that reference-class thinking is meant to expose.
Selection bias toward the available or the favorable. The comparable that is easiest to find — well-published, often a flagship or best-in-class plant — is not necessarily representative, and choosing the analog that flatters the target (the most efficient incumbent, the lowest quoted cost) biases the anchor. The similarity that matters is to the target, not availability or convenience.

Mini-example

Gray ammonia is itself the most common reference process for a new ammonia route, because it is mature and extensively characterized. In estimating a green ammonia plant, the synthesis loop is nearly identical to the conventional one, so its anchors transfer almost directly: the H₂:N₂ ≈ 3:1 feed ratio and ~20% per-pass conversion (the running-example baseline) come straight from established ammonia-synthesis practice, and the loop’s compressor and reactor costs can be anchored to known commercial figures and scaled with the six-tenths rule.

What does not transfer is the front end: green ammonia replaces the steam-methane-reforming section with electrolysis, for which a conventional ammonia plant is no comparable at all. That section needs its own reference — electrolyzer cost and performance drawn from water-electrolysis data — because borrowing the reformer’s numbers for an electrolyzer would import an unrelated process. The match is close for the synthesis loop and absent for the hydrogen source, which is exactly where the borrowed numbers can and cannot be trusted.

To show the edge, a separate one-line instance: anchoring a green plant’s whole capex to a single published gray-ammonia plant cost would inherit that one plant’s site, vintage, and scale as if they were the target’s — a point estimate masking a wide real-world spread, where the same nominal plant built in another place or year can differ substantially.