A Proof Log: What It Proves, and What It Does Not
· 7 min read · Updated
Lire en français
We readily write that every grant, every denial and every bypass attempt produces an entry that is timestamped and justified, that each entry is chained to the previous one, and that the head of the chain is signed at every upload. The phrase is accurate, but it raises a question our most rigorous counterparts never fail to ask, because one needs to know signed by whom, timestamped by which clock, and above all what such a proof log actually establishes on the day it must be produced before someone who has no reason to take our word for it.
This article is for security managers, lawyers and auditors who must assess what an event log produced by a device is worth, ours or anyone else's.
What an audit trail is for, when all goes well
The audit log, or audit trail, is often imagined as an exhibit for a trial, brought out the day things go wrong, whereas that is in fact its rarest use, since in the vast majority of cases it answers mundane questions, who used this bench last week, how many times was this machine started outside its slot this month, did the planned maintenance take place. It then serves, which is already less mundane, to show an auditor that a control exists and works, since an access policy that leaves no trace cannot be told apart, from the outside, from a policy that is not applied.
Only last does it become evidence in the strong sense, produced in a dispute with a contractor, before an insurer after a loss, or in proceedings, but it is this last use that dictates how it is designed, because a log that would not withstand adversarial examination is worth little more than a handover notebook.
What the law expects of an electronic record
We are not lawyers and we report here what we understand of it. Under French law, the Civil Code gives an electronic record the same evidential value as a paper one, on two conditions, which are that the person it comes from can be duly identified and that it is created and kept in conditions that guarantee its integrity (Article 1366). Between merchants, and in criminal matters, evidence is free, meaning a judge may rely on any element whose weight they assess, provided it was obtained fairly and debated by both sides, and most European legal systems follow, as far as we know, comparable principles.
At European level, the eIDAS Regulation defines so-called qualified electronic signatures, seals and timestamps, issued by supervised trust service providers, which benefit from a presumption of integrity and accuracy, and we want to be clear on this, because the signature produced by a field device is not a qualified electronic seal under eIDAS and does not carry that legal presumption, so its value rests on what can be demonstrated about how it is produced, and that is exactly what interests us here.
What a well-designed proof log establishes
A proof log does not need to be perfect to be useful, but it must answer four questions precisely.
Who produced this entry? Each Cube, our on-site decision unit (overview), has its own cryptographic identity, meaning a key pair generated in the device and confined to a secure component, and at every retrieval the Cube signs the head of its log chain with this key, and since each entry is linked to the previous one, that signature covers every entry before it, so a valid signature establishes that the log comes from that Cube, and from no other device or intermediate server, provided its key has not been extracted. That is the form non-repudiation takes here.
Has it been altered since? Each entry carries a fingerprint of its own content and of the entry before it, a fingerprint being a digest computed from the content whose value changes as soon as the content changes, which makes the log a hash chain, so changing a single character of an entry changes its fingerprint, breaks the chain from that point and invalidates the signature, and the result is a tamper-evident log that a third party can check with the Cube's public key, without having to take our word for it.
Are entries missing? This is the most often forgotten question, and yet it is where conventional logs seem to us weakest, because one can rarely modify a line unnoticed, but one can often delete one. Chaining answers this in part, since a deletion in the middle of the log breaks the chain detectably, but a cut at the end of the log would not break it, and that is why every entry carries a sequential number and why, at each retrieval, the head of the chain is handed to the authority, which acknowledges it, because rewriting the past would then also mean rewriting what the server has already received. We preferred the log to be able to say by itself that something is missing at a given point, rather than leave the reader to guess.
When? The timestamp comes from the Cube's clock, set by an authenticated holder during synchronisation, and every clock setting is itself a log entry, whereas an order arriving while the clock is not deemed trustworthy is refused rather than executed with a doubtful date, so accuracy depends on the last synchronisation, but the uncertainty is known rather than hidden.
What it does not prove
It would be tempting to stop there, but we think, on the contrary, that the value of evidence lies as much in the list of what it establishes as in the honesty with which its limits are described, and that is, in our opinion, what an expert would look at first.
It proves that a credential was presented, not that a person was there. The log establishes that an enrolled token, badge or phone, answered a cryptographic challenge correctly, and if the token was lent, stolen or used under duress, the log will not say so, which is true of any access control, including the building badge, and that is why we never claim to identify a person in the strict sense.
It proves what the device saw, not what happened. A machine powered through an extension lead from another socket, outside the device's perimeter, will not appear in the log, or only indirectly as an anomaly, because the log describes the protected perimeter, and nothing beyond it.
It depends on the integrity of the device producing it. A signature guarantees that an entry comes from the Cube, but it does not guarantee that the Cube was not deceived or compromised, and that is why the key is confined to a secure component rather than ordinary memory, why physical tampering attempts are themselves logged events, and why we publicly describe the attacks we know how to counter and those we cannot yet counter.
Its timestamp is not a qualified timestamp. It is consistent, monotonic and documented, but it carries no legal presumption of accuracy, and when a date must be asserted against a third party with certainty, it remains possible to have the exported log timestamped by a qualified provider, for instance when it is uploaded to the server.
What makes the difference before a third party
In the cases we have encountered, what gives a proof log weight before an auditor or an expert has less to do with cryptography than with three very concrete qualities. The first is continuity, because a log covering the whole period, with no unexplained gaps, including during a network outage of several days, is worth more than a richer but patchy one. The second is symmetry, because a denial must be recorded as carefully as a grant, otherwise the log only tells what went well. The third is reproducible verification, because a third party must be able to check signatures and chaining with standard tools, without depending on the vendor, which requires formats and public keys to be documented.
These are the three qualities we prioritised, and we invite you to judge us on them rather than on the word "signed", which on its own says very little.
A few precautions on your side
A proof log remains data you keep, and its value also depends on how you handle it, so we would suggest setting a retention period suited to your obligations and to the personal nature of the data it contains, keeping exported logs in a space whose access is itself logged, and never reworking the original files, since analysis is done on a copy and the original must remain verifiable. These precautions are more common sense than technology, but they are what make it possible, eighteen months later, to say without hesitation that the document produced is the one the device wrote, and we say what this log contains that is personal, and how to limit it, in Detecting Presence Without Monitoring Employees.
Sources
- French Civil Code, Articles 1366 and 1367 (electronic records and signatures).
- French Commercial Code, Article L110-3, and Code of Criminal Procedure, Article 427 (freedom of evidence).
- Regulation (EU) No 910/2014 (eIDAS), as amended by Regulation (EU) 2024/1183. EUR-Lex
- ISO/IEC 27037:2012, Guidelines for identification, collection, acquisition and preservation of digital evidence.
- RFC 3161, Internet X.509 Public Key Infrastructure Time-Stamp Protocol (TSP). IETF
