Thesis & writing
Thesis · 02Architecture

Designing for replayability

For shared work, current state should be backed by a reconstructable, attributable history that other participants can check.

On this page

To audit or dispute shared work, we need more than its current state. We need to reconstruct how that state was reached.

Current state is only a claim.

It tells you what the system says is true now. It does not tell you how that truth was formed. It does not tell you who caused it. It does not tell you what evidence existed at the time. It does not tell you whether the path to that state was valid.

Answering what exists now and explaining how it came to exist are different requirements. We are designing for both.

That is replayability.

We treat replayability as a starting constraint: the record must retain what a later reviewer needs to reconstruct a decision.

Current State

Asset status Issued
Approval Approved
Owner Current holder
Risk tier Low

Replay Timeline

Summary only — how did it get here?

No replay available — slide to reconstruct the path

Current state is a summary. Replayability preserves the path that made the summary meaningful.

Event history is not enough

A log can preserve timestamps and events without preserving the rules and evidence needed to interpret them.

The sequence alone does not answer the harder question.

Because replayability is not just about storing events. It is about storing enough structure for those events to be interpreted correctly later.

What was the state before the event.
Who acted.
Under what authority.
What evidence existed.
What was signed.
What algorithm was used.
What changed as a result.
What other actors had already signed or approved.
What was anchored publicly and what remained local.

Without that, you do not have replayability. You have chronology.

Chronology is useful. Replayability is stronger.

Chronology

Replayable Record

Ordering: yes Authority context: no Evidence binding: no Replayable: no
Ordering: yes Authority context: yes Evidence binding: yes Replayable: yes

A list of events tells you that something happened. Replayability tells you whether the state transition can still be reconstructed and checked.

Replayability needs proof of origin

A replayable system does not just preserve change. It preserves origin.

That means each meaningful action must remain attributable to an actor, tied to a proof, and verifiable later without relying on institutional memory or private assurances.

Shared work needs to preserve who asserted what, when, under which key and with what evidence. That does not require making every part of the record public.

A database row saying "approved by Alice" is not proof of origin. It is a label.

Proof of origin begins when the system can preserve the actor, the signing context, the content that was signed, and the signature evidence in a form that survives beyond the current application.

That is what makes replay meaningful rather than narrative.

Transition: Awaiting review → Approved

Select a layer to see what breaks without it

↗

Eight proof layers. Each one load-bearing.

Hover or tap any layer above to see what breaks when that proof element is missing.

Replayable history preserves the attribution, authority and evidence attached to each meaningful change.

Preserve the contributions, not just the final state

Real systems are rarely shaped by one actor signing one thing once.

They involve multiple actors contributing over time. A reviewer signs. A verifier adds evidence. An approver countersigns. An external authority attests. A system anchor confirms ordering. A later actor revokes or supersedes a prior state.

If only the final event remains, the earlier contributions and their context are missing.

A replayable system should preserve not only the final output, but the chain of attestations that formed it. Not one signature replacing another. A structured history of who asserted what at each stage.

That matters for workflow. It matters for authority. It matters for provenance. It matters for dispute.

Full signature chain — replay from origin to anchor

Chain integrity: Full
Replay confidence: Reconstructable

Evidence and approvals accumulate through a process. Replayability preserves the chain in which they were recorded.

Algorithm agility is part of replayability

This is often missed.

A signature is not enough on its own. The system must preserve how that signature should be interpreted. Which algorithm was used. Which key was used. Which challenge or verification context applied. Whether the proof can still be verified later as cryptographic expectations change.

That means replayability depends on algorithm agility.

A record can be reviewed after its original application has gone, its keys have rotated or its verification policy has changed. Replaying it means preserving the assumptions and supporting evidence that applied at the time, as well as the means to evaluate them later.

Preserved context keeps the original signature interpretable. Continued verification also depends on the evidence retained and the migration decisions made as cryptographic assumptions change.

At issuance

Verified

After migration

Context lost

Preserved context

Signature bytes only

Retaining algorithm and key context gives future verifiers something to assess. Cryptographic migration also requires supported verifiers and an explicit trust policy.

Replayability has to survive boundaries

Most real systems do not live in one place. They touch application databases, object storage, APIs, external registries, public ledgers, message systems, and downstream indexers. They may also cross chains, move between trust domains, or carry proofs from one environment into another.

That means replayability cannot be local only.

Replay confined to the original application depends on that runtime, storage model and operator remaining available. Our direction is to make the record independently interpretable.

A stronger design separates the system of record from the views built on top of it, and preserves proof packages that can travel across boundaries.

Not because every component is on one chain. Not because every proof is public. But because state transitions, signatures, evidence references, and anchor points remain interpretable across system boundaries.

Single point of failure — replay depends on one system

Single point of failure

Yes

Survives system removal

No

Cross-boundary replay

Not possible

Portable records support cross-system replay when their contents and verification dependencies remain available.

Replayability belongs to the record, not the view

Views can change. Dashboards can change. Queries can change. Applications can be replaced. Interfaces can be rebuilt.

Separating the record from its views lets us change an interface without redefining the history it presents. The work is in preserving the schemas, rules and evidence needed to rebuild those views.

Derived views should remain derived. The record should preserve the inputs and rules needed to rebuild its recorded state rather than merely inherit it.

Replayability is what makes the difference between a system that looks complete and a system that can still prove itself when challenged.

Operator dashboard

Status: Issued. Last action: Final approval by Actor B. Next: Distribution.

Record unchanged

Replayable Record

State Asset issued
Prior state Awaiting final approval
Actor Actor B (delegated authority, valid until 22 June)
Evidence 4 documents, 2 signatures attached
Proof Ed25519 signature, content hash SHA-256
Anchor Block 14,291,003

Replayability belongs to the record. Views only expose slices of what the record preserves.

Where serious replayability starts

It does not start with a log table. It does not start with a dashboard. It does not start with a chain transaction alone. It does not start with a current state document.

It starts with a harder question.

What must remain reconstructable years from now if this system is challenged.

Who acted.
What they acted on.
What state existed before and after.
What authority they had at the time.
What evidence was bound to the transition.
What proof of origin was preserved.
What signature context must remain verifiable.
What ordering or anchoring mattered.
What has to survive system change, chain change,
or cryptographic change.

That is the starting point. Storage, interfaces and verification tools have to serve that requirement.

Replayability is not about storing more data. It is about preserving enough structure that the system can still explain itself later.

Who acted. What changed. Why it was valid. What proof existed. How the state was formed. Whether that path can still be checked when the system is challenged.

A current state compresses a path into an answer. Our bet is that shared work becomes more dependable when the path remains available for inspection.

That is where we are putting the architectural effort: records that can be carried forward, reconstructed and checked as the surrounding systems change.

Current state is the answer. Replayability is the working.