Platform / Limits
The ledger
Everything this platform cannot do, in one place.
Every page of this site ends with its own honest limits. This page collects them verbatim, bar one sentence that carries live counts, with a link back to where each one is published. Nothing here is softened for being read all at once — the point of writing them down was that you find them now rather than during an incident.
Why it is a page and not a footnote
A vendor who publishes limits per page can still be read selectively. Collected, they are checkable: you can count them, and you can notice if one quietly disappears.
Last collected 29 August 2026. Every limit below still appears, word for word, on the page that publishes it.
What is not here
Nothing has been added, and only one limit is reworded — a single sentence that opens with live counts read from the build, marked on the row itself. Where a limit reads awkwardly out of context, that is the original wording, and the source page is one click away.
If one of these is a dealbreaker
Say so early. Several are dated rather than permanent — no certifications, no published amounts — and the honest answer to whether one is close is worth more than a discovery call.
The ledger · drawn from its own rows
Sixty-one limits from fifteen pages. Sorted by kind.
A vendor who publishes limits per page can still be read selectively. Collected, they are checkable: you can count them, and you can notice if one quietly disappears.
Sixty-one numbered limits start in the columns of the fifteen pages that publish them and settle into four groups by kind: eighteen “Not ours yet”, ten “What it cannot do”, twenty “Measured, not claimed” and thirteen “Yours to carry”. Pick a kind below to read only those.
Kind of limit
Published on
Search limits
61 limits · 15 pages
What it cannot do
01
Tampering before the seal is outside its reach
The seal proves the record has not changed since the platform sealed it. It says nothing about a source that lied before that boundary. That is addressed by attestation and source trust, and it is a different problem with different limits.
Measured, not claimed
02
Duplicate delivery is detectable, not automatically removed
Content-derived identifiers make a replayed frame recognisable. Nothing silently discards a second copy at ingestion; de-duplication happens on the read path. Silently dropping records would be a decision we would have to make deliberately and tell you about.
Measured, not claimed
03
A verifier that cannot check an algorithm flags it
Seals name the algorithm that signed them, so a store can hold more than one generation at once. A verifier that does not implement a newer algorithm marks those records unverified rather than vouching for them. Fail closed, then say so.
What it cannot do
04
A graph edge is only as good as the field that made it
Causal pointers are recorded by the component that produced the event. Where a source cannot tell us what caused what, the path has a gap, and the interface shows a gap rather than inferring a plausible one.
Yours to carry
05
A machine with no agent is not a quiet machine
It is an unobserved one, and the platform draws those two states differently everywhere. Nothing here infers safety from silence.
What it cannot do
06
We cannot normalise a field a source never sent
An absent field stays absent rather than becoming a zero or a default. Boards above will say a value was never stamped instead of inventing one.
What it cannot do
07
Sealing proves integrity, not truthfulness
If a compromised host lies to us, we can prove nobody altered the lie after it reached us. We cannot make it true.
Yours to carry
08
Retention is a budget, and it is yours
Questions that reach past your retention window have no answer here. We would rather say that plainly than return a confident partial one.
Measured, not claimed
09
Severity describes the pattern, not your estate
A rule cannot know whether this is normal for you. That is what the entity's own history is for, and disagreeing with a rule's ranking is a legitimate outcome of triage rather than a fault.
Measured, not claimed
10
A baseline needs time before it means anything
In the first weeks of a deployment, unusual and unfamiliar look identical. Boards say which they are showing you rather than presenting a young baseline as a confident one.
What it cannot do
11
A graph can only hold what layer one recorded
A gap in coverage becomes a gap in the chain, drawn as a break rather than closed with a plausible guess.
Measured, not claimed
12
Big searches are bounded, and say so
A query that hits its ceiling returns the newest results and tells you it stopped. Counts from a bounded scan are floors, labelled as floors — never presented as totals.
Measured, not claimed
13
An allow is permission to begin, not to finish
Allowed actions go to one machine first, then a check, then the rest. A failed check rolls back without asking anyone — which means a green verdict is not a promise the action succeeded.
What it cannot do
14
Not everything can be undone
Isolating a host is reversible. A killed session is not resumable, and a deleted file may be gone. Guards weigh reversibility, and irreversible actions are held more often by design.
Measured, not claimed
15
“Affects nothing” usually means we could not tell
A blast radius of zero is the dangerous reading. Where a machine's owner and dependencies are unmapped, the number is unknown rather than small, and the board says so.
Yours to carry
16
A refusal has no appeal button
Looking for one is the commonest mistake made on this board. Changing a refusal means changing the policy that produced it, deliberately, somewhere else.
What it cannot do
17
We can only revoke what we are the authority for
Where another identity provider issues the credential, we can tell you it exists and stop it reaching your machines — we cannot end its session inside a third party's product.
Yours to carry
18
An unmapped machine cannot enforce a decision
Enforcement happens where the agent runs. Somewhere with no agent is not covered by it, and coverage is reported as a count of machines rather than a percentage that flatters us.
Measured, not claimed
19
A permission we never read is not enforced
If a policy declares something nothing consumes, the platform says so rather than accepting it quietly. Desired state you believe is in force but isn't is worse than a gap you know about.
Measured, not claimed
20
Some questions have no answer, and get none
Where a field was never recorded, a report says it was never measured rather than returning a zero. A confident number is worth less than a stated gap.
Not ours yet
21
No customers, and no invented ones
There is no deployment in production at a named organisation. We considered writing a composite case study and decided a fictional customer is worse than an empty page.What you can verify today ·Run the 30-day evaluation on your own telemetry and inspect the live Evidence Standard via GET /v1/assurance/standard.
Measured, not claimed
22
Coverage of a catalogue is not coverage of an attacker
A fraction of a public technique list is a useful measure of breadth and says nothing about whether the one technique aimed at you next week is in the covered set.
Measured, not claimed
23
Detectors need their feed
A covered technique is covered when the stream behind it is actually arriving. Coverage measured against our build is an upper bound on coverage in your estate, never a promise about it.
Not ours yet
24
No independent audit
Everything here is self-measured. It is reproducible on your own hardware, which is better than a claim and not the same thing as somebody independent having checked it.What you can verify today ·Run the 30-day evaluation on your own telemetry and inspect the live Evidence Standard via GET /v1/assurance/standard.
Not ours yet
25
No certifications, and no mapping dressed up as one
A control mapping exists internally. A mapping is not an audit, and we will not put a standard's name on this page until somebody independent has signed against it.What you can verify today ·Inspect the machine-evaluated control framework mappings, rights-engine workflows, and signed evidence bundles in the trial deployment.
Yours to carry
26
On-premise cuts both ways
On an on-premise deployment, your data staying with you also means the operational burden stays with you: patching, backups, key custody and availability are yours. We think that is the right trade for this kind of software, and it is a trade.
Yours to carry
27
Privacy behaviour depends on your declarations
Purposes, retention windows and monitoring scope are things you declare. The platform enforces what you declare; where you have declared nothing, it does not constrain — it cannot decide on your behalf what is lawful where you operate.
Not ours yet
28
No status page yet
We do not publish a status page yet — including for SaaS. An on-premise deployment is yours to watch.What you can verify today ·Inspect the machine-evaluated control framework mappings, rights-engine workflows, and signed evidence bundles in the trial deployment.
Not ours yet
29
We are not a CVE numbering authority
Our advisory identifiers are ours. An advisory names a CVE only once one has been assigned, which means the CVE can lag the advisory.What you can verify today ·The live disclosure programme, safe-harbour status, and advisories are served directly from /.well-known/security.txt, GET /v1/trust/vdp, and GET /v1/trust/advisories.
Not ours yet
30
No bounty
There is no payment for reports. An advisory can credit you if you want it. Whether safe harbour applies is what the published programme declares — this page renders that answer rather than promising one.What you can verify today ·The live disclosure programme, safe-harbour status, and advisories are served directly from /.well-known/security.txt, GET /v1/trust/vdp, and GET /v1/trust/advisories.
Not ours yet
31
We hold no security certifications
A control mapping is published, and a mapping is not an audit. We will not list a standard here until somebody independent has signed against it.What you can verify today ·The live disclosure programme, safe-harbour status, and advisories are served directly from /.well-known/security.txt, GET /v1/trust/vdp, and GET /v1/trust/advisories.
Yours to carry
32
On-premise means you patch
On an on-premise deployment an advisory tells you what to upgrade, and the upgrade is yours to schedule. As SaaS, how upgrades are applied is described in your customer agreement, not assumed from this page.
Measured, not claimed
33
Everything above is generated from the programme itself
The contacts, the windows, the safe-harbour answer, the scope and the advisory list are read from the declared programme rather than typed into this page; the one typed line is the address at the top, which says where to write and nothing about what happens next. That is why the page can honestly say nothing is published when nothing is — the alternative is a static page that keeps promising a window after the promise has lapsed.
Not ours yet
34
Platform-line amounts are published. The other lines’ are not.
The platform line’s plans and prices are on this page. The verification line’s price list is not published yet, and the lines that meter on seats, nodes or hosts carry no published rate either; those are quoted against your own estate.What you can verify today ·Platform-line plans and overage rates are published on Pricing and served live by GET /v1/metering/rate-card; every product line runs in a 30-day trial.
Not ours yet
35
Lines without a published rate are still in the trial.
They are built, they are in the package, and you will use them in a trial. We are not going to publish a rate for them before we can invoice for them.What you can verify today ·Platform-line plans and overage rates are published on Pricing and served live by GET /v1/metering/rate-card; every product line runs in a 30-day trial.
Not ours yet
36
How verification checks are grouped is not published.
The unit is a resolved check. Checks are grouped into families that are priced differently, and that grouping is a quote conversation today rather than a public table.What you can verify today ·Platform-line plans and overage rates are published on Pricing and served live by GET /v1/metering/rate-card; every product line runs in a 30-day trial.
Yours to carry
37
A ceiling is only enforced where one is set.
A licence that sets no quota for agents or seats admits without limit, by design. The daily ingest ceiling is never a valve: going over it is metered and billed, and ingest is never paused.
Measured, not claimed
38
It is a planning estimate, not a quote
The per-event sizes are measured, the events-per-second defaults are judgement. Two estates with identical hardware can differ widely on audit policy alone.
Not ours yet
39
Sizing gives the volume, not the bill
This gives you the volume, not the amount. The platform line’s prices are on the pricing page; the other product lines are quoted against your own figure.What you can verify today ·Enter your estate counts on Sizing to compute daily ingest volume, Core node cells, and storage against measured event sizes.
Measured, not claimed
40
Above a certain load it shards
One Core node is planned to about 10,000 average events per second. Past that the calculator divides the estate into identical cells.
Yours to carry
41
The archive is yours to place
The archive figure is compressed roughly threefold and holds the balance of your compliance period beyond what stays searchable on the node. Where it lives, and what restoring from it takes, are yours to plan.
Measured, not claimed
42
A source in the list is not a promise about your version of it
Vendors change their log shapes. Each decoder is written against a documented format, and where a product emits something the documentation does not describe, that variant is a gap until somebody hits it.
Measured, not claimed
43
Connecting a source is not the same as covering a technique
A detector needs the specific field its rule reads, and plenty of feeds arrive configured to omit it. Measured detection coverage, and what blocks the rest, is on the proof page.
What it cannot do
44
One industrial protocol is deliberately not decoded
A widely deployed factory protocol has no published specification, so decoding it would mean working from reverse-engineered material. We refused to write that decoder; its traffic and connection envelope are still recorded from the open transport standards, with the protocol itself treated as opaque.
Measured, not claimed
45
This list is a snapshot of a build
A running deployment can be asked for its own catalogue, and that answer is the authoritative one for the version you hold.
Measured, not claimed
46
Compiled is not proven
A compiled cell is enforced at build time without a cited run on real hardware. That is the true state of that cell, and publishing it as a tick would be the lie.
The source sentence opens with two counts read from the build, so the opening clause is paraphrased here. The rest is verbatim.
Measured, not claimed
47
A capability is not a detection
Reading a signal on a platform is a precondition for detecting something with it. What is actually detected, and where that stops, is on the proof page.
What it cannot do
48
Old hosts are refused, not degraded
Operating systems below the toolchain floor get no agent. Collecting from them through another route is possible; pretending the agent runs there is not.
Measured, not claimed
49
Your build is the authority
This is a snapshot of one build. Each build can generate its own matrix, and where this page and your build differ, your build is right.
Not ours yet
50
We hold no certifications yet
No SOC 2, no ISO 27001 as a vendor. We are early, and claiming otherwise is the one thing that would end this conversation permanently. What we can give you is the platform's own evidence properties and a direct line to the people who built them.What you can verify today ·Walk any alert or control finding back to its Ed25519-sealed event record and export a deterministic, hash-verified evidence pack.
What it cannot do
51
Tamper-evidence is not tamper-proofing
We can prove a record changed. We cannot prevent someone with sufficient access to your infrastructure from destroying it — and a vendor who tells you otherwise about your own hardware is selling you something.
Yours to carry
52
Evidence stops at your retention window
A question reaching past it has no answer here. Legal hold exists precisely so an investigation can outlive the ordinary window, and it has to be applied before the window closes.
Yours to carry
53
A machine without the agent cannot be enforced on
Coverage is enforcement, and enforcement happens where the agent runs. An estate is never reported as compliant on the strength of the machines that happened to be instrumented.
Yours to carry
54
Capacity is yours to plan
You are hosting it, so retention and event volume are your infrastructure decisions. We will size it with you honestly, including telling you when a customer's volume makes a deal unattractive.
Not ours yet
55
Six lines are not sold yet
They are in the package and they work, but you cannot resell what has no commercial terms. Build your service catalogue on the two lines that are generally available — see pricing.What you can verify today ·Multi-tenant partner cells, white-label posture, and signed WebAssembly decoder plugins ship in the platform today.
Not ours yet
56
There is no marketplace to sell into
Signed third-party modules are part of the architecture and not part of the business yet. If you are betting on revenue share from your own content, that door is not open today.What you can verify today ·Multi-tenant partner cells, white-label posture, and signed WebAssembly decoder plugins ship in the platform today.
Not ours yet
57
We are early, and you would be too
There is no partner programme with tiers and badges. What exists is a direct line to the people who built it, which is better for the first few partners and worse for the fiftieth.What you can verify today ·Multi-tenant partner cells, white-label posture, and signed WebAssembly decoder plugins ship in the platform today.
Not ours yet
58
We are early, and that is the whole risk
No production deployment at a named organisation, no certifications, no independent audit. Joining early means the platform's credibility is still being established, and your work is part of establishing it.What you can verify today ·Read the engineering essays, coverage-gate proofs, and live assurance endpoints to see how the system is built and tested.
Not ours yet
59
There is no programme around the job
No levelling framework, no structured onboarding, no learning budget policy to point at. What exists is direct access to the people who built the thing, which is better for the first few hires and worse for the fiftieth.What you can verify today ·Read the engineering essays, coverage-gate proofs, and live assurance endpoints to see how the system is built and tested.
Yours to carry
60
Wide scope cuts both ways
Working across this many capabilities means less depth in any one of them than you would get elsewhere. If you want to become the definitive expert on a single subsystem, this is the wrong place for a few years yet.
Not ours yet
61
No compensation bands published
Compensation is not published yet: the numbers are not settled, and posting a range we would have to revise is worse than saying so. It is discussed in the first conversation rather than the last.What you can verify today ·Read the engineering essays, coverage-gate proofs, and live assurance endpoints to see how the system is built and tested.
The list you should trust is the one you produce yourself.
Thirty days on your own hardware, with nothing switched off, will tell you which of these matter in your estate and which never come up. Ask the deployment what it covers and compare its answer to what is written here.