Verify: download → compute → enter your verdict → compare

This page does not ask you to believe a single conclusion. It gives you three inputs and one algorithm note, so that you work out the verdict yourself and then compare it with our "expected verdict".

Step 1Download

Three offline vectors that open and save with no network. Write for the originals (address in section 7). Start with group A alone, get it through once, then take B and C.

Group A input — expected: independently verifiable (PASS)
Group B input — expected: altered (FAIL), naming entry 3 and the "quantity" field
Group C input — expected: this book is incomplete (INCOMPLETE), entry 4 missing
Recomputation notes (formulas / intermediate values / verdict) + self-test script (standard library only, runs offline)

Step 2Compute it yourself (three steps, standard library is enough)

Count the entries: how many entries the ledger actually holds.
Sum: group by "metering item + unit", convert the quantity field from a decimal-integer string to an integer and add. No floating point — once a decimal enters, the rounding drift enters the ledger.
Compute the digests: serialise each entry as "keys sorted, no spaces, UTF-8", hash it once; then hash "entry count + totals + each entry's id and digest" the same way to get the whole-ledger digest.

Then compare those three numbers with the numbers declared in the input file. Do not read our verdict first and work backwards — that is copying the answer, and a copied "match" is worth nothing.

Step 3Enter your verdict

Which group you checked Your verdict (choose one) If FAIL or INCOMPLETE: name the entry and the field (e.g. the quantity field of entry 3, or entry 4)

The comparison is entirely offline front-end computation: this page does not go online, does not write any of your input anywhere, and keeps no record.

Step 4See whether it matches

Two things must match; missing either one counts as "no match":

The verdict matches: the same one of verifiable / altered / incomplete;
The naming matches: an "altered" verdict must state "which entry + which field + original value to current value"; an "incomplete" verdict must state "which entry is missing". Saying only "it is wrong" is not a match.

Why "naming" counts as a criterion: in reconciliation, "one entry missing" and "one entry altered" are two completely different actions — one is replenished, the other is investigated. Saying only "the digest does not match" answers neither.

5. Four falsifiable claims, each mapped to a contract clause number

These four are not attitudes; they are four designs that you can check and can falsify. Each points to a specific clause in the contract template (the template can be supplied on request, see Connect). Only what can be checked in the contract counts as having been said.

#	The design (a claim that can be falsified)	Which clause you check	What falsification looks like
One	No licence, no spectrum, no user data	Clause 2 subject matter and exclusions; clause 9 data terms; annex A boundary statement	A licence, spectrum or data asset appears in the subject matter, or the data terms lack "data plane deployed locally"
Two	The service identity and the shareholder identity are kept separate	Clause 3 signing party and authorisation chain; annex C authorisation chain note	The signing party and the shareholder are the same legal entity, or the chain cannot say who authorised whom
Three	The technology can switch between constellations	Clause 4 constellation list; clause 5 switching responsibility; annex B asset register	Fewer than two, the two are related, no comparable quotation, or nobody bears the switching cost
Four	The standard is governed by more than one party	Clause 6 composition of the certification committee; clause 7 publication of the standard; annexes D and E	The committee has no target-country representative or no independent technical expert, or the published scope cannot be stated

6. Verification on the API side: four more, equally falsifiable

The four above check the contract. There are four more on the API side, checking "whether the door is really locked". All four have been measured, and the test records can be supplied on request — write and ask.

#	The claim that can be falsified	How you check it yourself	What falsification looks like
One	No credential, no entry	Call the three write endpoints once without a credential	200 without a credential ⇒ falsified (in that case, write and tell us at once)
Two	No entry beyond your permission	Take a credential with only usage-reporting permission and call device registration	200 ⇒ falsified. The correct answer is 403 — not 401: the identity is fine and retrying will not help
Three	Revocation takes effect immediately	Mark a credential as revoked and re-sign, then call the same request again without restarting the service	Still 200 ⇒ falsified. The correct answer is 401 on the next request
Four	An altered ledger is refused	Change one character in one field of one ledger line, then call the request for that line	Still 200 ⇒ falsified. The correct answer is that that line is void and answers 401 (only that line, not the whole book)

The measured readings for these four, and "what the API answers today", are on neutralorbit.org/en/status (measured 2026-11-01).

7. Where to obtain the material and test records

This page and this site do not host the originals: the three vectors, the recomputation notes, the self-test script, the contract template, the records for the four tests, and the six "inspectable materials" of the contract.

To obtain them, write to partners@neutralorbit.com (the address is live; see section 7 of Connect). Say which item you need and what you want it for.

Next step

If the numbers hold, see what makes the standard count and who sets it.

Go to Standards →