up:: Migration Architecture MOC

The Evidence Ladder

The evidence ladder is the three-tier ranking of the proof that a post-quantum migration actually worked, ordered by how independently each source can be confirmed: rung 1 is the wire, the record of what real connections actually negotiated; rung 2 is the certificates, signed and dated and machine-readable, carrying their own algorithm and validity inside them; rung 3 is vendor attestations, for the surfaces you don’t control. The order is the whole framework. Evidence of what happened outranks evidence of what a signed record claims, which outranks a supplier’s word about a surface you can’t observe. And the discipline that makes the ladder honest is that a vendor’s silence goes on the record as a dated finding, because “this surface is unconfirmed” is itself a true and auditable fact about the estate.

The short version:

• Proving a migration worked means handing an auditor evidence they can re-derive without trusting you, and the three sources that hold up rank in a fixed order.
• Rung 1, the wire: what real connections actually negotiated, counted over time, is evidence of what happened and outranks everything.
• Rung 2, the certificates: signed, dated, machine-readable records that carry their own algorithm and validity dates as structured fields, close to self-proving.
• Rung 3, vendor attestations: written statements about surfaces you can’t observe, graded as the vendor’s claim, with every non-answer recorded as a dated finding.
• Four properties make any rung hold (dated, sourced, reproducible, traceable), and a fifth outranks them all: current. Documented evidence records that someone said a thing is true; verified evidence was independently confirmed and can be confirmed again.

Think of how a courtroom weighs proof. The security-camera footage of the event outranks everything, because it records what actually happened and anyone can watch it back. A signed, dated contract ranks below the footage and above a person’s memory, because it’s a fixed record you can examine, though it only proves what was written down. A witness statement ranks below both, because it’s one person’s account, valuable but the weakest of the three. And when a witness refuses to testify, the court doesn’t leave a blank space in the record. The refusal is entered, with a date, because the fact that a witness would not speak is itself part of the case. The evidence ladder is that hierarchy, applied to a cryptographic migration, and the refusal-on-the-record rule is what keeps the whole thing trustworthy.

What is the evidence ladder?

The evidence ladder is a scoring rule for migration proof, and it has three fixed rungs plus a governing discipline.

1. Rung 1, the wire. A capture or a stream of negotiation telemetry showing the share of real handshakes that used the post-quantum group, counted over time. This is evidence of what your systems actually did rather than what they were configured to do, which is why it sits at the top. The instrument that produces it is Wire Over Config: the configuration is the menu, and the wire is what was ordered.
2. Rung 2, the certificates. Signed, dated, machine-readable records whose signature algorithm and validity dates are structured fields of the certificate itself, so an export can’t misremember them the way a spreadsheet can. A certificate is close to self-proving, because reading it is re-deriving it.
3. Rung 3, vendor attestations. Written statements from suppliers about the post-quantum posture of surfaces you can’t instrument. Evidence of what the vendor claims, dated and recorded, and graded honestly as the weakest of the three.

The governing discipline sits underneath all three rungs: a supplier who won’t answer still produces a recorded finding, because a documented “this surface is unconfirmed, as of this date” is a true fact about the estate that an auditor can act on.

The ladder applies when you’re proving a migration after the fact, to an auditor, a regulator, a board, or your own next review. It’s the post-deployment companion to Testing as the Control Surface, which proves a deployment is correct before it ships. What the ladder does NOT do is decide which systems to migrate first, or how much any of it costs, or whether a given surface is exposed in the first place. It ranks the proof once the work is claimed done, and the same three instruments you used to find your cryptography during discovery are the three you now turn toward proving you fixed it.

Why does the order matter?

Because the rungs differ in one property an auditor cares about above all others: how far each piece of evidence sits from someone’s word. An auditor’s whole job is to prefer evidence they can confirm over claims they have to trust, and the ladder is that preference, made explicit and ranked.

The wire outranks the certificates because it records the outcome of a real negotiation with a real client, so it captures the gap between what a system permits and what it actually does. A configuration can list a post-quantum group as enabled while a meaningful share of connections still complete on classical key exchange, because an old client never offered the group or a middlebox forced a classical retry. The wire shows that gap; the certificate inventory doesn’t, because a certificate describes what’s installed and stays silent about what got negotiated on the day in question. The scale of that gap is measured: a 2026 study of Nginx configurations published to public GitHub repositories found 28.9% of TLS-enabled contexts still using RSA key exchange with no forward secrecy, a pattern the paper attributes to copy-paste and template inheritance rather than to client demand.

Source: Balaji, H., et al., “Operationalising Post-Quantum TLS,” arXiv:2605.17955, 2026, arxiv.org.

The certificates outrank the vendor attestations because a certificate is a structured, signed artifact you can parse and re-derive yourself, while an attestation is prose a supplier wrote about their own product. RFC 5280 defines the signature algorithm and the validity period as fields of the certificate, so an export carries its own proof of what algorithm signed it and when it’s valid, with nothing to take on faith.

Source: RFC 5280, “Internet X.509 Public Key Infrastructure Certificate and CRL Profile,” §4.1 (signatureAlgorithm, validity, and subjectPublicKeyInfo as certificate fields), datatracker.ietf.org.

The vendor attestations sit at the bottom because they’re a supplier’s claim about a surface you can’t observe, and the grade for a claim is documented rather than verified. They still belong on the ladder, because for a surface nobody outside the vendor can instrument, a dated written statement is the best evidence obtainable, and the alternative is a blank.

Rung	What it is	Why it ranks there	Sample artifact
1, the wire	Negotiation telemetry, counted over time, showing the share of handshakes that used the post-quantum group	Records what actually happened, capturing the gap between permitted and negotiated; re-derivable by re-reading the logs	A 30-day export of negotiated cipher suite, key-exchange group, and certificate signature algorithm, with counts, from the edge and load balancers
2, the certificates	Signed, dated, machine-readable records carrying algorithm and validity as structured fields	Close to self-proving, because reading the certificate re-derives the claim, though it describes what’s installed rather than what negotiated	A certificate export from the internal CA showing signature algorithms and validity dates per certificate
3, vendor attestations	Written statements from suppliers about surfaces you can’t instrument	The vendor’s claim about their own product, graded documented until independently confirmed; still the best evidence for an unobservable surface	A dated, signed supplier statement plus a CBOM naming the algorithms in use

What does each rung look like as an artifact?

Each rung produces a concrete deliverable, and knowing what a good one looks like is how you tell proof from a screenshot.

Rung 1, the wire, is a counted export. Not a single packet capture and not a claim, but every distinct cipher suite, key-exchange group, and certificate signature algorithm negotiated over a real window, with counts, pulled from the tier that terminates your traffic. The counts are what make it evidence rather than an anecdote: they turn “the hybrid group is reachable” into “61% of 40 million handshakes used it this month, and here’s the trend against last month.” The mechanics of the pull, the exact three fields, the 30-day window, and where the logs already live, are the whole subject of Wire Over Config, and the ladder inherits that instrument wholesale for its top rung.

Rung 2, the certificates, is a parsed export. A dump of the certificate estate where each row carries the signature algorithm and the validity window read straight out of the certificate’s own fields, so the artifact proves itself on inspection. The public infrastructure for this already exists at internet scale: Certificate Transparency logs are append-only, publicly auditable records of issued certificates, and the same structured, dated, machine-readable quality that makes those logs auditable is what makes a private certificate export rung-2 evidence. A certificate that says it was signed with an ML-DSA algorithm and is valid across a stated window is testimony that can’t quietly change its story between the export and the audit.

Rung 3, the vendor attestation, is a dated written record. A statement from the supplier naming the algorithms their product uses for key establishment and signing, ideally delivered as a CBOM in a standard format so the answer is structured rather than marketing prose. The date is load-bearing, because an attestation is only ever true as of when it was made, and a supplier’s posture shifts when they ship an update. The artifact is the statement plus its date plus, where the vendor provides it, the CBOM that lets the claim be checked against a named standard.

Source: CycloneDX, “Cryptography Bill of Materials,” cyclonedx.org.

What makes any rung hold?

Four properties make a single piece of evidence defensible, and a fifth outranks all four. An auditor trusts an artifact when they can re-derive it themselves, and these are the properties that let them.

1. Dated. A cryptographic posture is only ever true as of a moment, so an undated capture proves nothing about the day in question. Every artifact carries the date its evidence describes.
2. Sourced. The artifact traces back to the exact configuration line, the specific log, or the named capture that produced it, so anyone can follow it to its origin rather than to an assertion.
3. Reproducible. Someone else running the same check against the same system gets the same answer. Reproducibility is what separates evidence from a one-time screenshot nobody can regenerate.
4. Traceable. Every entry points back to where it came from. The federal-grade version is that each field in a cryptographic inventory records what it was derived from, down to the file and the line, which a CBOM captures as a provenance field on every row.

Those four are the difference between documented evidence and verified evidence. Documented evidence records that someone wrote down that a thing is true. Verified evidence was independently confirmed and can be confirmed again. An auditor is paid to prefer the second, and the four properties are what turn the first into the second.

Property	What it means	The failure it prevents
Dated	Carries the date its evidence describes	An artifact that proves a posture but not the day it held
Sourced	Traces to the exact config line, log, or capture	Evidence you can’t follow back to an origin
Reproducible	The same check run again yields the same answer	A one-time result nobody can regenerate
Traceable	Every entry records what it was derived from	An inventory that rests on assertion rather than provenance
Current (outranks all four)	Describes the estate as it is now, not last quarter	Proof that was accurate the day it was filed and stale by the audit

The fifth property, current, is the one a written report can never hold, because the estate doesn’t stop moving the day you finish. A cryptographic inventory goes stale within roughly 6 months: new systems arrive with old defaults, a troubleshooting change re-enables a weak setting, a vendor pushes an update, a certificate lapses and auto-renews under the wrong algorithm. So the proof you assembled last quarter is describing an estate that has since changed. Being current outranks being dated, sourced, and reproducible, because a perfectly dated and sourced artifact from 6 months ago proves a posture the estate has already drifted out of. Keeping evidence current is the standing discipline of Continuous Cryptographic Assurance, and it’s why the deliverable at the top of a mature program is a living record rather than a filed report.

Why is a vendor’s silence a recorded finding?

Because a gap is a promise that future work will happen with nobody accountable in the meantime, and a finding is an accountable record with a date, an owner, and a consequence. Rung 3 is the one rung where you depend on someone else’s cooperation, so it’s the rung where silence has to be handled deliberately rather than left blank.

A vendor round splits cleanly into two outcomes, and both produce evidence. A supplier who answers hands you a dated attestation you can file at rung 3. A supplier who declines, or points at a compliance page and stops, hands you a documented open finding, and the move is to write it down with today’s date, because a refusal you can date is evidence and an undated one is a rumor. Filing the attestation as a finished item, while quietly leaving the non-answers as empty cells, hides the exact surfaces an auditor goes looking for first. Recording the silence as a finding does the opposite: it puts the unverifiable surface on the map, with a date that marks when the organization knowably knew, and an owner who can act on it through the one lever that works on a vendor, which is procurement and renewal terms.

This is the same rule that governs Zone 3 in The Three Zones, where the cryptography that stays invisible no matter what you do is reported as a dated, owned finding rather than a blank on the map. The two frameworks rhyme because they’re solving the same problem from two directions: the three zones sort the estate by who controls each surface, and the evidence ladder ranks the proof each surface can yield, and both land on the rule that an unverifiable surface is a finding rather than an omission. Federal practice already treats it this way. OMB M-23-02 requires agencies to inventory their cryptographic systems including contractor-operated ones, and keeps the accountability with the agency rather than the vendor, so a supplier’s silence is a risk the agency owns on paper from the day it’s dated.

Source: OMB M-23-02, “Migrating to Post-Quantum Cryptography,” November 18, 2022, whitehouse.gov.

The joint CISA, NSA, and NIST guidance points every other organization down the same road, telling them to engage vendors proactively about post-quantum roadmaps rather than waiting to be told, which only works if the non-answers are recorded as they arrive.

Source: CISA, NSA, NIST, “Quantum-Readiness: Migration to Post-Quantum Cryptography,” August 2023, cisa.gov.

How do you rank real migration evidence?

You collect what each surface can yield, then place every piece on its rung and grade it. Take a generic mid-size estate that has just declared its first migration phase complete: a public API behind a managed load balancer, an internal PKI, a payments processor reached over an API, and an HR SaaS. The team assembles what it has to prove the work.

Evidence offered	Rung	Grade	Why it lands there
30-day wire pull from the edge: 61% of handshakes on X25519MLKEM768, with counts and a trend line	Rung 1	Verified	Records what actually negotiated, dated, sourced to the load-balancer logs, reproducible next month
Certificate export from the internal CA showing ML-DSA signatures and validity windows	Rung 2	Verified	Structured, self-proving fields read straight from the certificates
Config screenshot captioned “TLS 1.3 hybrid enabled, preferred”	Not on the ladder	Not evidence of outcome	The config is the menu; it states what’s permitted, not what ran (see Wire Over Config)
Payments processor’s signed statement plus a CBOM, dated this quarter	Rung 3	Documented	The vendor’s claim about their own surface, dated, checkable against a standard
HR SaaS: vendor pointed at a compliance page and declined to name algorithms	Recorded finding	Dated finding	A non-answer, written down with today’s date and an owner, rather than an empty cell

Reading the ranked stack top to bottom tells the true story the config screenshot hid. The API edge is 61% migrated with evidence under the number, a denominator, and a trend, which is a defensible rung-1 claim. The internal PKI is proven at rung 2. The config screenshot drops off the ladder entirely, because it proves what the system was allowed to do rather than what it did, and offering it as proof is the single most common way a team mistakes a claim for evidence. The payments surface is documented at rung 3, honestly graded as the vendor’s word until the wire or a CBOM confirms it. And the HR SaaS becomes the most important line on the page: a dated finding that names an unverifiable surface, assigned to whoever owns third-party risk, framed as a renewal-negotiation item rather than a blank the auditor discovers first.

What the estate walks out holding is a proof stack ordered by strength, with each item graded and dated, and its weakest surface named on the record instead of hidden. That’s a defensible answer to “prove it worked,” and the config screenshot alone was not.

Common misconceptions

1. “A configuration screenshot proves the migration.” A configuration states what a system is permitted to do, and a screenshot of it proves the menu rather than the meal. The proof of what ran is the wire, the negotiated record on real traffic, which is why rung 1 exists and a config screenshot doesn’t appear on the ladder at all. The full argument is Wire Over Config.
2. “A point-in-time audit is permanent truth.” An audit result describes the estate on the day it was taken, and the estate drifts within months as systems arrive, configs change, and certificates renew. Current outranks dated for exactly this reason, and a report filed once ages into fiction, which is the standing problem Continuous Cryptographic Assurance exists to solve.
3. “A vendor’s word ranks alongside my own wire evidence.” A supplier’s attestation is a claim about a surface you can’t observe, and your wire pull is a record of what your own systems actually negotiated, so they sit on different rungs by construction. Grading a vendor statement as verified because it’s signed confuses documented with confirmed; a signature proves who made the claim, not that the claim is true.
4. “We’ll assemble the evidence when the auditor asks.” Evidence is only as good as the moment it describes, and you can’t reconstruct last quarter’s wire after the fact, because the handshakes that would have proven your posture happened and are gone. A scramble on request produces a snapshot of today, which says nothing about the date the auditor is actually asking about, so rung-1 proof has to be captured continuously and dated as it happens.
5. “A signed vendor statement closes the gap.” An attestation is evidence of what the vendor claims rather than proof the surface is migrated, and its real value is that it gets recorded either way. Treating the signature as a finish, instead of as a dated rung-3 record whose silences are themselves findings, hides the surfaces an auditor probes first.
6. “If everything still connects, everything is still compliant.” A cryptographic regression is silent: a fallback to classical key exchange or a re-enabled weak suite still completes the connection and still shows a padlock, so the service runs while the posture has drifted. Availability tells you nothing about which cryptography is in use, and only watching the wire continuously catches a quiet regression.
7. “More evidence is better than ranked evidence.” A thick binder of mixed-strength artifacts buries the one rung-1 export that actually proves the outcome under a stack of screenshots that don’t. The ladder’s value is the ordering: an auditor wants the strongest evidence per surface, clearly graded, rather than volume.

Pro tips

1. Lead with the wire, and keep the counts. The single most persuasive artifact in the stack is a counted, dated wire pull with a trend against the previous window, because it answers “what actually ran” with a number and a denominator. Report shares, never a bare list of what was seen at least once, so the second pull becomes a trend instead of a repeat.
2. Grade every artifact out loud. Label each piece verified or documented on the page itself. A reviewer who sees the grades trusts the whole stack more, because self-grading is the tell of a team that knows the difference, and it pre-empts the exact question an auditor would otherwise open with.
3. Date the non-answers the day they arrive, and file marketing replies as non-answers. A vendor reply that names no algorithms is a non-answer, and two dated non-answers across consecutive quarters is escalation material at renewal. The dates are what make it escalation rather than a grievance.
4. When a vendor reply is evasive, narrow the question rather than repeat it. “Which algorithms for key establishment, which for signing, and can you provide a CBOM in a standard format” is answerable by any engineering team in an afternoon, and a supplier who can’t answer the narrow version has told you where the surface really stands.
5. Ask which rung a claimed proof sits on. When anyone offers evidence that a surface is migrated, the second-layer question is “is that the wire, a certificate, or a vendor’s statement?” The answer places it on the ladder and grades it in one move, and it surfaces a config screenshot masquerading as proof immediately.
6. Turn the discovery baseline into the before-and-after. The inventory you built to find your cryptography is the baseline your proof measures against, so keep it: the migration story is the delta from that baseline plus the running confirmation that the new posture still holds today.

Where does the evidence ladder break?

The ladder breaks wherever there’s no wire to read, and it’s built to say so rather than paper over it.

1. Surfaces with no wire access. A SaaS platform hosted by someone else, a vendor-sealed appliance, or a managed surface whose handshake lands on infrastructure nobody on your side can instrument yields no rung-1 evidence, because there’s no negotiation available to observe. For those surfaces the ladder honestly starts at rung 2 where a certificate is obtainable, and at rung 3 where only an attestation is, and the defensible sentence is “we have wire-grade proof for what we terminate, certificate-grade for what we can export, and dated attestations for the rest, with every unanswered surface a finding.” Announcing that the ladder starts lower on a given surface is the posture, and the surfaces where it starts at rung 3 are precisely the Zone 3 findings that belong in front of a board.
2. Cryptography that never negotiates. Data-at-rest encryption, code signing, firmware trust anchors, and key wrapping run fixed algorithms with no per-connection choice, so there’s no wire outcome to capture. Their proof runs on the other rungs, chiefly certificate and key inventories and the provenance a CBOM carries, and stretching rung-1 language over them overstates what the evidence covers.
3. The window bounds the claim. A 30-day wire pull proves 30 days, so clients on quarterly or annual cycles and cold standby paths can sit outside any single window. The cadence mitigates it and honesty covers it: the claim is always “in this window, on these termination points,” and the running feed is what extends the proof past a single snapshot.

Within its lane, one limit deserves naming. The ladder ranks the strength of proof, and it stays silent about which surface matters most. Deciding where to aim the strongest evidence first takes the risk models, blast radius for dependency weight and HNDL for what’s already exposed, so the ladder grades the proof and the risk sort tells you which surface to prove first.

How do you use the evidence ladder in the boardroom?

One level up, the ladder converts a vague “we migrated” into a graded, defensible position a board and a regulator can both stand on. The deployment is a single beat: “We can prove our migration three ways, ranked by strength. On the surfaces we control, we have the wire, the record of what actually negotiated, at 61% and climbing. On our certificates, we have signed, dated proof of the algorithms in use. On the surfaces outside our control, we hold dated vendor attestations, and every vendor who hasn’t answered is a recorded finding with an owner and a renewal date.” Every clause in that beat is a verifiable fact drawn from the estate’s own evidence, which is what makes it hard to argue with and easy to minute.

The same framing does two more boardroom jobs. It sets the acceptance standard for the whole program, because once “migrated” means “the wire shows it,” a config screenshot stops being accepted as proof anywhere in the organization. And it authorizes the vendor pressure campaign, because a dated non-answer recorded as a finding is what turns a renewal into a leverage point, and those levers live at the level the board controls. The broader briefing pattern, including how to frame quantum risk without theatrics, is in Brief Your Board.

Questions people ask

Why does the wire outrank the certificates? Because the wire records what actually negotiated on real traffic, which captures the gap between what a system permits and what it does, while a certificate describes what’s installed and stays silent about what got chosen on the day in question. A meaningful share of connections can complete on classical key exchange even when the post-quantum group is enabled, and only the wire shows that.

Isn’t a signed certificate stronger than a log? For proving what algorithm signed a given certificate, yes, and that’s exactly what rung 2 is for. For proving what your systems negotiated with real clients, the certificate is silent and the wire is definitive, so they answer different questions and the ladder ranks them by which question an auditor asks about migration outcome.

What do I do with a vendor who won’t answer? Record it as a dated finding: the surface, the date you asked and got no usable answer, the named owner (usually whoever holds third-party risk), and the consequence. That finding is both your honest evidence for the surface and your leverage at the next renewal, and it’s far stronger than leaving the cell blank.

What’s the difference between documented and verified evidence? Documented evidence records that someone said a thing is true, like a vendor attestation. Verified evidence was independently confirmed and can be confirmed again, like a wire pull anyone can re-run. An auditor prefers verified, and the four properties, dated, sourced, reproducible, and traceable, are what move an artifact from documented to verified.

Why can’t the proof just be a report I file once a year? Because the estate drifts within months as systems arrive, configs change, and certificates renew, so a report is accurate the day it’s filed and stale by the audit. Current is the property that outranks all the others, and only a living feed holds it, which is the discipline Continuous Cryptographic Assurance describes.

Do I need to prove every surface at rung 1? No. You prove each surface at the highest rung it can yield: rung 1 where you can read the wire, rung 2 where you can export a certificate, rung 3 where only an attestation exists, and a dated finding where even that’s refused. A stack that’s honestly graded per surface is stronger than one that pretends every surface reached rung 1.

How is this different from testing before deployment? Testing proves a deployment is correct before it ships, as a gate at the moment of change. The evidence ladder ranks the proof after the fact, for an auditor asking whether the migration actually worked in production and stayed working. Testing is the pre-flight check; the ladder is the flight recorder.

Can the wire pull come from a scanner’s clean report? A scanner is one well-behaved client making a handful of connections, so its report shows the best available outcome rather than what your thousands of real clients negotiate. A clean scan proves the strong option is reachable, and the counted wire pull proves how often it’s what actually ran, which is the evidence the ladder ranks at rung 1.

---

Everything here is the map, given freely. When your team needs its own wire pulled, its certificates parsed, its vendors pressed, and all three wired into one dated, sourced record that answers an auditor at any moment, that’s a working session with your team.

Last verified 2026-07-26 · Updated 2026-08-25 · Maintained by Addie LaMarr, LaMarr Labs.