device-identity: third adversarial review and its repairs (with the release left out of a bundle, the firmware the device attests is still judged against the manufacturer's policies; a verifier clock and every time read strictly as RFC 3339 UTC, days that do not exist refused; C1, DEL and direction marks escaped on the terminal; a key file is never overwritten); probe 372/372, negative control 220/220. conformance: the halves of the device verifier mapped again (263/263 under the negative control) and the present-tense counts brought to the measurement of 2026-10-14 (467 vectors, 467/467 with the AIP-23 verifier as served)
This commit is contained in:
1 parent
ed4d3d79d8
commit
86779c8334
8 files changed
+263
-100
No files matched your search
+30
-10
@@ -94,9 +94,9 @@ not rule ids. Their own probes and negative controls stay what checks them.
|
||||
|
||||
## The vectors
|
||||
|
||||
`vectori/` holds 422 vectors (413 offline, 9 that read testnet 28001; 16 MB, since every device and identity vector carries its ML-DSA-65 keys and signatures), listed with their digests in `vectori/INDEX.json`:
|
||||
`vectori/` holds 467 vectors (458 offline, 9 that read testnet 28001; 19 MB, since every device and identity vector carries its ML-DSA-65 keys and signatures), listed with their digests in `vectori/INDEX.json`:
|
||||
envelope 64, payment-risk 11, payment-receipt 10, market-timeline 26, dataset 22, ai-trust-bundle 24, audit-log 19, audit-inclusion
|
||||
10, audit-consistency 4, agent-ledger 18, identity-presentation 65, device-bundle 149. By expected verdict: 59 `VALID`, 47 `PARTIAL`, 316 `INVALID`. None is typed by hand
|
||||
10, audit-consistency 4, agent-ledger 18, identity-presentation 65, device-bundle 194. By expected verdict: 64 `VALID`, 66 `PARTIAL`, 337 `INVALID`. None is typed by hand
|
||||
(`node genereaza-vectori.mjs --verifica` regenerates them and compares byte for byte):
|
||||
|
||||
- the 16 AIP-23 kinds come from their producers on fixed inputs (the same envelopes as the published AIP-23 kind vectors); the audit
|
||||
@@ -121,10 +121,10 @@ envelope 64, payment-risk 11, payment-receipt 10, market-timeline 26, dataset 22
|
||||
from the chain (two read calls).
|
||||
|
||||
An `INVALID` vector names its `base` and its `mutation`: **one operation** (set, remove or add at one JSON path) applied to the input
|
||||
of a vector that is not invalid, and it must be refused for exactly the rules in `expected.failed`, not merely refused: 209 of the
|
||||
316 are made this way. Where the rule sits under a hash or a signature, one place is not reachable (changing an amount under the
|
||||
of a vector that is not invalid, and it must be refused for exactly the rules in `expected.failed`, not merely refused: 221 of the
|
||||
337 are made this way. Where the rule sits under a hash or a signature, one place is not reachable (changing an amount under the
|
||||
instruction hash trips the binding first; a false decision in a ledger must still be signed by the agent; a disclosure the issuer never signed must still be signed for by
|
||||
the holder; an authorization by a previous owner must still be signed by that owner): the other 107 are made at
|
||||
the holder; an authorization by a previous owner must still be signed by that owner): the other 116 are made at
|
||||
the producer, the commitments above the change recomputed with the published formulas, and every path that moved is declared
|
||||
(`mutation.changed`). Eleven vectors expect more than one rule: rules that guard the same field cannot be separated (a looser policy shown
|
||||
under a receipt also makes the risk verdict not reproduce; a Proof of AI envelope that is changed is also no longer the one the
|
||||
@@ -136,9 +136,9 @@ key breaks both the id derivation and the match with the manufacturer's credenti
|
||||
|
||||
```
|
||||
AERE_VERIFY_PROOF=<verify-proof.mjs from aere.network/tools, with verify-anchor.mjs, mpt.mjs, falcon512.mjs, registry-v3.mjs and package.json beside it and npm install done>
|
||||
node conformance.mjs --impl "node implementare-aere.mjs" # the published verifiers behind the one verdict form: 413 offline vectors
|
||||
node conformance.mjs --impl "node implementare-aere.mjs" # the published verifiers behind the one verdict form: 458 offline vectors
|
||||
node conformance.mjs --impl "node implementare-aere.mjs" --online # and the 9 that read testnet 28001 (nothing is sent)
|
||||
node conformance.mjs --impl "python3 implementare-integritate.py" # an independent implementation: 62 vectors, 360 reported NOT COVERED
|
||||
node conformance.mjs --impl "python3 implementare-integritate.py" # an independent implementation: 62 vectors, 405 reported NOT COVERED
|
||||
node conformance.mjs --impl "<your command>" # yours
|
||||
node implementare-aere.mjs <artifact.json> [--online] # one artifact { artifact, input }: its verdict
|
||||
```
|
||||
@@ -161,12 +161,32 @@ refusal. Without `--online` an implementation must not open a connection.
|
||||
`implementare-integritate.py` is the second implementation: Python, standard library only, written from the rules and not from the
|
||||
JavaScript. It claims 6 of the 181 rules (the form and the statement hash of an envelope; the form, the commitment, the salted field
|
||||
commitments and the RFC 9162 inclusion of a dataset record) and passes all 62 vectors it claims. It cannot check an ML-DSA signature
|
||||
or read a chain, says so in its capabilities, and the runner reports the rest as not covered: 360 vectors, among them every vector of
|
||||
or read a chain, says so in its capabilities, and the runner reports the rest as not covered: 405 vectors, among them every vector of
|
||||
the trust chain, the audit log, the agents' ledger, the identity presentations and the device bundles. For those six verifiers the
|
||||
suite has one implementation, ours.
|
||||
|
||||
## How the suite itself is checked
|
||||
|
||||
Measured on 2026-10-14 (UTC), after the third review of the device verifier (Node.js 24.14.1; the AIP-23 reference verifier fetched
|
||||
from aere.network/tools with its four sibling files and `npm install` done in an empty folder, as a stranger would; testnet 28001
|
||||
read, nothing sent):
|
||||
|
||||
- `node proba-conformance.mjs --online`: 12 rows green, 0 red (in this published copy the comparison of the 16 kind envelopes with
|
||||
the published AIP-23 kind vectors is printed as information; the twelfth row is green where their generator is). The rules and the
|
||||
vectors are what the code derives now: 181 rules of 11 verifiers, 467 vectors. Every rule has a vector that expects it (181 of 181);
|
||||
every artifact kind has a valid vector (15); every invalid vector is its base with one operation (221) or one declared producer
|
||||
change (116); the published verifiers pass 467/467; the independent implementation passes all 62 it claims and the other 405 are
|
||||
reported not covered; 96,757 malformed inputs given to the module verifiers each got a verdict and none threw (one vector,
|
||||
`device-bundle/chain-of-64.partial`, is larger than 300,000 characters and is not damaged this way); 183 malformed envelope files
|
||||
given to the reference verifier each got a verdict.
|
||||
- `node proba-conformance.mjs --control-negativ --online`: 263 rows green, 0 red: 181 of 181 rules, 76 of 76 conditions of the device
|
||||
verifier (under them 96 distinct device vectors fail, each exactly under the conditions written for it), six controls of the
|
||||
instruments (an implementation that calls everything valid fails 426 of the 458 offline vectors, every one that is not plainly valid).
|
||||
|
||||
The finer counts in the paragraphs below (how many plantings of the device verifier no vector reaches, how many device vectors the
|
||||
purpose separation breaks at once, and the `--pe-loc` measurement) are those of the measurement of 2026-10-05; they were not counted
|
||||
again on 2026-10-14.
|
||||
|
||||
Measured on 2026-10-05 (Node.js 24.14.1, Python 3.14.4, testnet 28001 read):
|
||||
|
||||
- `node proba-conformance.mjs --online`: every row green, 11 in this copy (a twelfth, the comparison of the 16 kind envelopes with
|
||||
@@ -247,7 +267,7 @@ On 2026-10-05, while the identity verifier was being enrolled, one more, fixed t
|
||||
## What it is not, and what is not measured
|
||||
|
||||
- **Not a certification, not an audit, not a mark.** Passing says that an implementation agrees with the published verifiers on these
|
||||
422 inputs. It does not say either of them is correct, and no third party has reviewed the verifiers, the vectors or this suite.
|
||||
467 inputs. It does not say either of them is correct, and no third party has reviewed the verifiers, the vectors or this suite.
|
||||
- **The vectors and the verifiers have the same authors.** The expected verdicts were written by the people who wrote the verifiers.
|
||||
The second implementation is independent in code and language, not in authorship, and it covers 6 rules of 181; the other
|
||||
175 rules, among them every rule of the trust chain, the audit log, the agents' ledger, the identity verifier and the device
|
||||
@@ -289,7 +309,7 @@ On 2026-10-05, while the identity verifier was being enrolled, one more, fixed t
|
||||
survive a sale: after a transfer it is named and not applied (the manufacturer's is final). A retirement by an owner that holds
|
||||
across a sale needs a witness of when it and the transfer were signed, which is not built; every time in a transfer, an
|
||||
authorization and a status record of these vectors is the statement of the key holder that signed it.
|
||||
- **What a damaged artifact can still pass.** Of the 70,358 malformed inputs, 3,889 were still accepted, at 192 distinct paths (81 of them in device bundles: a status record that decided nothing, such as a suspension already lifted, a stranger's record, a record not in effect yet or one of a previous owner; a recall list or a firmware policy that could not be read and stood beside a clean one, which damaged is no longer signed by the manufacturer and counts like one that was not handed; the envelope and its notarization claim, which are optional; and the signature of a Proof of Software attestation, which the device verifier does not read): parts a
|
||||
- **What a damaged artifact can still pass.** Of the 96,757 malformed inputs (measured 2026-10-14), 4,280 were still accepted, at 187 distinct paths (76 of them in device bundles; the kinds that follow were listed on 2026-10-05, when the count was 81 and the device verifier had not had its third review: a status record that decided nothing, such as a suspension already lifted, a stranger's record, a record not in effect yet or one of a previous owner; a recall list or a firmware policy that could not be read and stood beside a clean one, which damaged is no longer signed by the manufacturer and counts like one that was not handed; the envelope and its notarization claim, which are optional; and the signature of a Proof of Software attestation, which the device verifier does not read): parts a
|
||||
verifier treats as optional (the transaction facts of a receipt, the envelope of a timeline, the head of an audit log), a policy
|
||||
field removed that falls back to its default, a log or a ledger without its last entry (still a valid shorter one), a revocation that was not in effect (damaged, it is
|
||||
ignored as before), and fields no rule reads
|
||||
|
||||
Reference in new issue
Block a user