In this article

Define what must survive

Imagine recovering an examination and finding every observation in the list. The notes are readable, the conclusion is present and the record count matches. Yet the image file has been replaced, an audio attachment is absent or a cited feature now resolves to its latest revision rather than the submitted one. The interface may appear intact while the basis of the examination has changed.

PROFAS treats recovery as a question about connected material. The database contains accepted records and their history, but source bytes, canonical analytical representations, audio and attached reports also contribute to what a reader can inspect. Recovering only the current database projection would not necessarily recover the examination that supported a particular conclusion.

That gives a practical starting point for a rehearsal: identify the artifacts and relationships whose survival can be checked. Source hashes answer a byte-identity question. Submission identities and citations answer a historical-basis question. Access checks answer a different question again. None can stand in for all the others, and matching totals alone cannot establish any of those relationships.

Restore somewhere deliberately separate

The discipline rehearsals use isolated recovery destinations. The DNA reconciliation script is constrained to its designated loopback environment, including the source-file URLs it will fetch. This makes the intended destination part of the exercise rather than relying only on an operator remembering which terminal or backend is active. The restored data can be inspected without replacing the working application.

The research library follows the same principle with different storage. Its recovery preparation requires explicitly named recovery D1 and R2 targets and emits an assertion that the relevant tables are empty. It verifies the private bundle’s checksums before preparing restoration statements and an original-file manifest. A recovery bundle is not an instruction to merge old rows into whichever database happens to be available.

These constraints serve a bounded purpose: they make restoration safer to rehearse and its result easier to inspect. They do not perform an institutional failover, choose a live hostname or decide that recovered accounts should immediately regain access. Moving from an isolated restoration to an operating service remains a separate, deliberate action.

Compare meaning as well as counts

The DNA rehearsal reconciled original FSA files, canonical signals and associated analysis-export bytes against their retained hashes. It also retained the native signal frame. These are distinct artifacts: a trace rendered successfully on screen would not, by itself, show that the original acquisition file or the imported analytical material had survived unchanged.

The script then checked the independent baseline, the submitted target and the verifier’s report by their recorded identities, together with disclosure ordering. It checked audio bytes and their link to the accepted note and signal view. An externally supplied engineering report was retrieved under frozen-target access and compared by hash. The point was to recover the associations, not merely recover files with recognisable names.

The histories in that exercise were scripted engineering examples using real public research source files. They demonstrated the specified software behaviour after restoration. They were not observations of human examiner independence or evidence that an interpretation was scientifically correct. Recovery can preserve a conclusion and its basis faithfully without validating the conclusion itself.

ENGINEERING NOTE

What does “the same examination” require?

  1. Source identity

    Original and derived bytes retain their recorded hashes and native frames.

  2. Historical basis

    Baseline, target, disclosure and cited revisions keep their identities.

  3. Reproduced output

    Frozen research and teaching receipts return the same package bytes.

These checks answer complementary questions. A successful database import alone does not cover the whole chain.

Reproduce the frozen output

An export provides another useful boundary. The DNA rehearsal repeated the accepted research export requests against the restored system and compared their receipts and complete package bytes with the originals. The teaching export received a corresponding check. This tests whether the stored request, its frozen basis and its output remain connected after restoration.

The important qualification is the exact accepted request. Re-running an old export is not the same operation as creating a new dataset from today’s current records. A faithful retry should reproduce its original result. A new extraction may legitimately select a different basis. Recovery checks need to make that distinction explicit or they risk treating expected change as corruption, or unexpected change as normal.

Earlier-head replay adds a related check from another direction. In the rehearsal, an earlier analysis view retained its native frame and continued to withhold reference material. A system can retain every later record and still fail to reconstruct what was available earlier. Checking that earlier view makes the historical access boundary part of recovery rather than only part of normal navigation.

Recover access deliberately

The examination app’s backup arrangement separates the consistent data-and-file snapshot from an encrypted environment capture and a source recovery pack. The decryption key is held separately from the backup executor. Signing configuration and other environment requirements therefore remain part of recovery planning; a restored data archive alone does not demonstrate that browser authentication or a live cutover has been completed.

The library makes a different recovery choice explicit in its prepared data: restored memberships and agent grants are disabled, identity bindings and work leases are cleared, queued authoring work is paused and maintenance remains enabled. Historical authorship survives without automatically reviving the authority of old credentials. Reauthorising access is an owner operation after inspection.

There is also a clear boundary around Git. Metadata and extracted text provide useful portable research assets, but the deliberate Git export omits original files, private identity mappings and credentials. It cannot replace a complete private recovery bundle. These rehearsals demonstrate the artifacts and conditions they checked; full operational disaster recovery, live reauthentication and larger deployments require their own scoped exercises. The useful habit is to specify what “recovered” means before accepting the reassuring appearance of a restored screen.