Last verified 19 August 2026.
The question that actually gets asked
Not “why this material”. Nobody asks that, because the answer is usually obvious from the requirement.
What gets asked, six months or three years later, in a design review, a failure investigation or a customer audit, is some version of:
Where did this number come from?
And the honest answer, most of the time, is that it came from a tab that is no longer open, in a session nobody logged, from a source that may since have been revised. The material choice was probably fine. The record of the choice is what failed.
That gap is not a discipline problem. It is a tooling problem: the information exists at the moment of deciding and there is no cheap place to put it.
The seven things a record has to contain
1. The requirement, as testable constraints. Not “corrosion resistant in seawater” but the property, the operator, the value, the unit and the condition it holds at. An adjective in a constraint list means that constraint was never actually screened — nobody can screen against an adjective, so it got waved through.
Requirements that genuinely are not numeric — weldability with existing procedures, availability, an approvals history, a customer restriction — are real and belong in the record. They belong as qualitative requirements with a named source, not converted into a score so they can be ranked. Turning “the shop can weld it” into a 7 out of 10 is how a preference gets laundered into an analysis.
2. What was considered, including what did not survive. The single most useful line in any selection record is the one that says what was rejected and why. A shortlist without its rejections is a conclusion; with them it is an analysis, and it is the difference between a reviewer being able to check your reasoning and having to take your word.
3. Where each number came from. The source, and specifically enough that somebody else can reach the number — a standard with an edition and a table, a database entry with its ID, a certificate with a heat number. “MatWeb” is not a source; a MatWeb entry with the date you read it is.
4. How strong each number is. A specification minimum, a datasheet typical value, a DFT result and a model prediction are four different kinds of claim. Recording which is which is the whole of the evidence ladder, and it costs one extra column.
5. The date you read it. Databases are revised, datasheets are reissued, standards get new editions. A value with no retrieval date cannot be reproduced, because the reader cannot know which version you saw.
6. Who decided. A named person. Not a department, not a tool, not a company. If the record ends without a name, nobody signed it, and its status in a review is somewhere between a note and a rumour.
7. What the analysis did not examine. Every selection has conditions it did not cover: fatigue, a joining method, a temperature excursion nobody mentioned, the supplier’s actual heat. Naming them protects you far more than omitting them does — an unstated limitation reads, later, as an oversight, while a stated one reads as scope.
The fifteen-minute version
If the full thing will not get written, write this. It fits on one screen and covers most of the value.
Illustrative: the constraints below are invented and the sourced values are the ones used in the 2205 example. Do not reuse it as engineering data.
DECISION Selected 2205 for the tubesheet.
REQUIREMENT PREN >= 32 at 40 C seawater; Rp0.2 >= 400 MPa, plate, 20 C;
thermal conductivity >= 14 W/(m*K).
CONSIDERED 316L, 2205, super duplex 2507, 6% Mo austenitic.
REJECTED 316L - fails strength, 170 MPa min vs 400 required.
2507 - passes everything, rejected on no qualified weld
procedure in the shop and single-source supply.
6% Mo - not evaluated, no sources consulted.
NUMBERS Rp0.2 450 MPa - ASTM A240 min, via secondary source, E3,
read 2026-08-19. Note: EN 10088-2 gives 500.
PREN 35 - Outokumpu Forta datasheet, E3, read 2026-08-19.
k 15 W/(m*K) - same datasheet, typical value, E3.
DECIDED BY [name], 2026-08-19.
ACCEPTED Rp0.2 used as a design allowable at E3 rather than E5.
Superseded when the 3.1 certificate arrives.
NOT COVERED Crevice corrosion. PREN is not a seawater qualification.
Fatigue. Galvanic coupling to the carbon steel shell.
Tube-to-tubesheet joint.Nine lines of substance. Everything a reviewer needs, and it can be pasted into a design note as it stands.
Note the ACCEPTED line, because it is the one people leave out. Using a datasheet value to
size a part is normal and often correct. Doing it without recording that you chose to is
what turns a judgement into an invisible default — and when the certificate arrives, that line
is what tells you which number to go back and replace. See
reading a mill certificate.
Three habits that make it survivable
Write it while deciding, not after. The information is free at the moment of decision and expensive an hour later. A record reconstructed from memory is a reconstruction, and it will be missing exactly the rejections you no longer remember making.
Record the date, always. It is one field. It is also the difference between a record that can be reproduced and one that merely looks thorough.
Keep the rejections. They are the part of the record you will want most and the part that feels least worth writing. Three months later the only question anyone has is why you did not use the obvious alternative.
Where a tool helps and where it does not
A tool can assemble candidates, carry the sources through to the output, apply the evidence rules consistently, and produce the document — that is what our research agent is for, and what the open record format specifies.
A tool cannot decide, and should not appear to. It cannot know that your shop has no qualified procedure for 2507, or that this customer has a history with a supplier. And it cannot be the name at the bottom. Whatever produced the analysis, the record ends with a person, because a record nobody signed is not a decision.
What this page does not cover
- Design codes. ASME, EN 13445, Eurocode and their equivalents impose their own documentation requirements, which govern over anything here.
- Configuration management. Where the record lives, how it is versioned and who can change it is a systems question, not a content one.
- Qualification and testing programmes. Documenting a selection is not the same as qualifying a material.
- Legal weight. What a record proves in a dispute depends on jurisdiction, contract and sector, and this page is not advice on that.