What the metal seed backup tests actually say
Search for "best metal seed backup" and most results point to one public set of stress tests: Jameson Lopp's Metal Bitcoin Seed Storage Reviews. This page is our dated summary of what those tests show, what they do not show, and the ways they are most often misread. It links to his work rather than copying it: read his table and individual reviews for the detail.
Key takeaways
- On 25 October 2026 Lopp's table listed 75 devices: 45 graded A, 11 graded A−, and 19 below that. An A means he could not destroy the data, so "top of the tests" is a crowded group, not one winner (the numbers).
- The designs that do well are simple: stainless steel or titanium, one solid piece, with the data centre-punched or deeply marked and the template marked just as deeply (what makes a backup survive).
- Price does not track the grade. Plain punched plates sit in the A group alongside devices costing many times more (how to read a grade).
- Some roundups and vendor guides recommend devices with low grades, or overstate a device's place in the results. Check the grade yourself (commonly misreported).
- These are informal tests by one person on single samples. They are strong evidence about designs, not a certificate for the product you buy (limits of the tests).
Summarised from this page only. Follow each link for the detail and the sources behind it. Educational information, not advice.
How many devices passed, and with what grades? Beginner
Evidence: Grade 1 Independent test Lopp publishes his method, photos and a review for each device, and has no stake in the products he grades.
We counted the overall grades in Lopp's public table on 25 October 2026. His page says he tested 75 devices and does not plan to test many more, because most new products resemble designs he has already covered.
| Overall grade | Number of devices | What it broadly means |
|---|---|---|
| A | 45 | He could not cause data loss in the heat, acid or crush test |
| A− | 11 | Every one of these had A in two tests and B in the third |
| B+ or B | 3 | Mixed results, with at least one weak test (C or D) |
| C or C− | 6 | At least one test graded D or F |
| D | 5 | At least one test graded F |
| F | 5 | Failed heat and acid; four of the five have no crush grade |
| Total | 75 | 45 A, 11 A−, 19 below A− |
By individual test, the most common failure was heat: 10 devices scored F for heat and 8 scored F for corrosion. Five scored F for crushing, and 6 have no crush grade at all (shown as a dash); each of those had already failed the corrosion test.
Checked on · counts made by us from the grades shown in Lopp's public table on this date
Counts are our own tally of the overall and per-test grades on Lopp's table. He also links a sortable spreadsheet (Google Sheets). The project is open source and takes corrections through GitHub, so grades and prices can change: check his page before relying on a number here. Our physical storage guide describes selected devices and what happened to them.
Who runs the tests, and how? Beginner
Jameson Lopp is a Bitcoin security engineer who co-founded Casa, a company that sells multisig key-management services. His method page sets out three tests, each harsher than most homes will see:
- Heat: a propane torch flame of about 1,093 °C (2,000 °F) for up to 10 minutes, then a plunge into water to mimic a fire being put out. His background page puts typical house-fire peaks at around 704 °C.
- Corrosion: 12 hours in 16% muriatic (hydrochloric) acid, or until bubbling stops. He says strong acid is probably not a real threat for most people; the point is to separate the strongest designs.
- Crush: a 20-ton hydraulic press, after which he checks whether the data can still be read.
Fuller detail, including what happened to selected devices, is in our stress-test evidence section, and the house-fire context is in how hot is a house fire?
How should you read a Lopp grade? Beginner Intermediate
- The grade is about durability only. It does not score privacy, how easy the device is to fill in correctly, or whether the seller could know your words. Lopp separately warns against devices where the maker may know your private key, naming Keyois Capsule, Cold Storage Coins and Ballet Crypto; one of those, the Keyois Capsule, still has an A− for durability.
- A is a group, not a rank. 45 devices share the top grade. Within that group the table does not put one device above another.
- Read the three test grades, not just the overall one. A device with an F for crush may still suit a backup that is unlikely to be crushed, while an F for heat matters for almost everyone. His method page says the grading is "arbitrary": a relative ranking against the rest of the market, not a pass mark.
- A dash is not a pass. It means no grade was given for that test.
- Price does not follow the grade. The prices he lists for A-graded devices run from about €13 to over €360, in mixed currencies and without dates. For current prices see our dated price check.
- The tested unit may not be today's product. Makers change materials and designs; a review describes the version Lopp tested.
What makes a metal backup survive? Beginner Intermediate
Lopp drew these lessons together in his design treatise (August 2022, after testing about 70 devices). Here is our summary, matched to the evidence behind each point.
| Factor | What the evidence says | Evidence |
|---|---|---|
| Material | Stainless steel (he prefers austenitic grades such as 316L) or titanium. On his background page he calls stainless steel the best metal that is easy to get hold of; for most people it is generally sufficient, and titanium is not automatically better (the titanium ColdTI scored C because of its engraving and screws). He warns against aluminium, and a thin copper sheet (Copper Seed Safe, C−) failed the crush test. | Grade 1 Independent test |
| How the data is marked | Centre-punched dots or electrolysis etching are the two methods he recommends. Stamping letters works but is hard to do well. Shallow engraving can be lost to heat or acid: on the ColdTI plate, engraved words were lost where the flame hit, while stamped words survived. | Grade 1 Independent test |
| One piece or many | Fewer parts means fewer ways to fail. Designs that hold loose letter tiles in a tray or rail lost letters when heated or crushed (for example Billfodl, C, and Keystone Tablet Plus, C). If you do use tiles, use one tile per word with the word number on it. | Grade 1 Independent test |
| Fasteners | Screws and bolts melted, dissolved in acid, stripped or seized crushed devices shut. Avoid them where you can; data was sometimes intact but hard to reach. | Grade 1 Independent test |
| Shape | Small but thick resists bending: he suggests at least 5 mm of metal, ideally over 1 cm. Big stacks of plates hold heat and can warp when cooled quickly. | Grade 1 Independent test |
| Template | The grid or guide that tells you what the marks mean should be marked as deeply as the data. Printed or lightly etched templates were erased by heat or acid. | Grade 1 Independent test |
| Format | Avoid novel or proprietary encodings: you or your heirs may need to read the backup decades from now. Plain BIP39 words (or their first four letters) survive partial damage better than word numbers, because one lost character leaves fewer possible words. | Grade 1 Independent test (his reasoning from testing) |
| Who sees your words | Never use a device that comes with words already on it or where the maker generates the key. Assume a seller's customer list could leak, so do not have backup products delivered to where you will keep them. Generic hardware-shop blanks avoid both problems. | Grade 1 Independent test (his advice; see also safety) |
Checked on · each row compared with Lopp's treatise, method and background pages and the linked reviews
A maker's own "fireproof" or "indestructible" wording is Grade 3 Maker claim until someone without a stake has tested it. Details and more examples: design lessons from the stress tests and materials compared.
What gets misreported about the tests? Beginner Intermediate
- "The top performer in Lopp's tests." No single device tops the table: 45 share an A. For example, an affiliate roundup dated 1 October 2026 (it carries an Amazon Associate disclosure) calls the Cryptosteel Capsule Solo the strongest all-round performer in Lopp's tests. Lopp grades the Cryptosteel Capsule A− (B for crush), below the 45 A-graded devices. The same roundup says its sources include no Lopp grade for the CryptoTag Zeus, but his table lists the CryptoTag Zeus with A in all three tests (roundup, read 25 October 2026).
- Low-graded devices described as "a great option". Ledger Academy's recovery phrase article (updated 19 June 2026) names Billfodl, Cryptosteel Capsule Solo and Cryptotag Zeus as examples of metal storage that is "a great option" against fire. Lopp grades Billfodl C overall (C for heat, D for crush) (Ledger Academy, read 25 October 2026). Ledger is a hardware wallet maker.
- "Fireproof" claims treated as test results. Some products that failed his heat test had been sold as fireproof; see the example in our stress-test evidence. A maker's rating is Grade 3 Maker claim, not a test.
- "Titanium beats steel." Both can score A. What failed was usually the design (tiles, screws, shallow marks), not the metal.
- "Not tested means as good as tested." Absence from the table is not a pass. Lopp has stopped testing most new devices, so judge an untested one against the criteria above.
We give these as examples of a pattern, checked on the date shown, not as a verdict on any publisher. Pages change: if one has been corrected, tell us and we will update this list (see corrections).
Checked on · each example re-read on the publisher's page and compared with Lopp's table on this date
What can't the tests tell you? Intermediate
- One tester, one sample. Each device was tested once, by one person, using his own equipment. He calls the tests "not highly scientific". A second sample might do better or worse.
- Specific conditions. A torch flame for up to 10 minutes is not the same as hours in a burning building, a long soak in salt water or years buried in damp soil. The tests do not cover those.
- Durability is only one risk. The tests do not measure theft, loss, a backup nobody can find, a wrong word written down, or a missing passphrase. For those risks see where to keep backups and the case files.
- Designs change. A grade is a snapshot of the unit he tested, sometimes years ago.
- Disclosure. Lopp co-founded a company that sells multisig services; his device reviews do not sell plates, but his wider advice favours multisig. We class his tests as an independent source because he has no stake in the devices graded.
So what should you buy or make? Beginner
- Pick a simple design that meets the criteria: one solid stainless steel or titanium piece, centre-punched or deeply marked, with a deep template. Lopp names single plates with centre-punch grids as the best value, giving Bitplates, Blockplate, CodlKeys, Smallseed and Steelwallet as examples, and the Quadrat Register devices as the most durable he tested. These are his examples, not our endorsements.
- Or make one yourself from generic hardware-shop parts, so no seller knows what it is for: see our DIY metal backup steps. Two of his DIY designs scored A.
- Check the backup on your device after you mark it, and keep a second copy in a separate place (see location strategy).
- Compare prices on dated figures, not on roundup tables: see the price check and the directory, where each metal backup shows the evidence grade of its sources.
Primary sources for this page
- Lopp: Metal Bitcoin Seed Storage Reviews (table, read 25 October 2026)source type: Independent · stress test methodsource type: Independent · why stress testsource type: Independent
- Lopp: A Treatise on Bitcoin Seed Backup Device Design (August 2022)source type: Independent · Lopp: stress test round IIIsource type: Independent
- Individual reviews: Cryptosteel Capsulesource type: Independent · CryptoTag Zeussource type: Independent · Billfodlsource type: Independent · ColdTIsource type: Independent · Keystone Tablet Plussource type: Independent · Copper Seed Safesource type: Independent · Keyois Capsulesource type: Independent
- Project repository and corrections (GitHub)source type: Independent · sortable spreadsheet
- Examples in the "commonly misreported" box: Ledger Academy: what is a recovery phrasesource type: Vendor · affiliate roundup of metal seed backups (1 October 2026)
Last reviewed: · Educational information only — verify current details with official sources.