Research and version note This is a version 0.1 research draft in Who Maintains the World? It uses Australian Government information management as its main institutional case. It distinguishes public records, working knowledge, and individual memory, and does not argue for retaining all information or assign all organisational memory to archives.
Who Maintains the World? · Article 8
Quick read
Staff turnover need not cause institutional amnesia. Public bodies should not depend on a long-serving employee remembering every decision. Records, procedures, handovers, data, and professional communities allow knowledge to cross individual terms. Yet the presence of a file does not mean an institution remembers. Information that cannot be found, lacks context, depends on obsolete formats, or is not consulted in decisions is stored residue.
Institutional memory contains at least three different things: records that provide evidence and accountability, operational knowledge required to perform work, and contextual knowledge about exceptions and historical reasons. Records management supports all three but cannot fully convert tacit skill into documents. A stable institution needs formal information and continuity of people to correct each other.
The National Archives of Australia’s Building Trust in the Public Record policy requires agencies to create, manage, and use information assets effectively and assigns responsibilities to agency heads and everyone performing government work. Information should remain findable, understandable, and usable, while accountable destruction occurs after value ends. Retaining everything increases privacy, security, cost, and retrieval risks.
Outsourcing, system migration, and restructuring are dangerous transition points. A supplier can hold configuration and decision context, then hand back files without working knowledge. Data can move from a legacy system while field meanings disappear. “Handover complete” should be tested by whether successors can reconstruct an actual decision and continue service.
More retention is not always more memory. Excess material makes the significant record harder to find and increases privacy and security exposure. Lawful disposal, classification and access control are part of maintenance, as is preservation. The standard is whether evidence with continuing value remains authentic, understandable and available to the people entitled to use it—not whether storage volume keeps rising.
Institutional memory is not storage of the past. It is the maintained capability for past evidence to support present judgment and future accountability. Significant decisions should preserve reasons, evidence, authority, and disagreement; critical operations require handovers that have been exercised. Memory must also permit old rules to be overturned. It should prevent repeated error, not turn “we have always done it” into authority.
An institution has no remembering mind
It is convenient to say that a department knows something. In practice, facts are dispersed among records, databases, procedures, employee experience, and suppliers. No individual has all of them, and people transfer, retire, and leave.
The consequences of a decision commonly outlast its participants. Years later, an institution may need to explain why a licence contained a condition or a system used an exception. If reasons remained in a meeting memory or private mailbox, successors see a rule without the risk it addressed.
Institutional memory is therefore not an enlarged psychological memory. It is an arrangement that carries evidence, meaning, and ways of acting across personnel change. Records are central but incomplete.
Records, operational knowledge, and context
Public records establish what happened: who made a decision, on which evidence, under what authority, and with what notice. They support rights, audit, judicial review, and history. Without records, an institution can redescribe the past when challenged, while an affected person cannot prove an entitlement or experience.
Operational knowledge explains how work is done. It includes workflow, system fields, inspection methods, and emergency steps. Some can be written into manuals and training, while practice supplies judgment about which anomaly needs escalation.
Contextual knowledge explains why a procedure has its present form and which assumptions may have changed. A form field may derive from repealed law. An awkward workaround may compensate for a persistent defect. Without context, successors retain a meaningless step or remove an important control.
These forms require different maintenance. Legal records emphasise integrity and authorised retention. Operational knowledge needs testing and updates. Context depends on decision reasons, handover, and professional discussion. Assigning everything to an archive system does not explain how knowledge re-enters action.
The public record as infrastructure
The National Archives of Australia’s Information Management Standard treats government information as an asset supporting decisions, transparency, risk management, rights, and entitlements. Information should be accurately created, findable and interpretable, preserved in suitable systems, and available for use.
The Building Trust in the Public Record policy places governance responsibility on agency heads and states that everyone working for government, including contractors and consultants, has information duties. Its extension to 2028 reflects an ongoing rather than transitional need to govern digital and non-digital information.
Calling records infrastructure does not make a server equivalent to a road. Records provide common conditions for many decisions. Entitlements, contracts, environmental approvals, and personnel accountability depend on usable past evidence. Failure may not cause one dramatic collapse, but it increases delay, repeated evidence gathering, and inconsistent treatment across cases.
A saved file can still be forgotten
Cheap digital storage encourages organisations to equate preservation with backup. A file exists without metadata, authority, or version relationships. A document named final_v7_revised does not show whether it was approved. Spreadsheet codes have no data dictionary.
Long-term use is also affected by format, software, and links. A page remains while attachments disappear. A database export loses field relationships. Scans are technically present but not searchable. Preserving bytes does not preserve evidential function.
Findability matters in current work. Under pressure, staff use the information they can retrieve. An official guide that cannot be found loses to an old personal template. Information management must be embedded in working systems rather than performed only as retrospective filing.
Access and change histories may matter as much as content. Security controls have to protect privacy and sensitive information without preventing legitimate users from obtaining what they need.
Why everything should not be retained
The intuitive response to amnesia is to keep everything. That creates another form of failure. Information exceeds classification and retrieval capacity; important evidence is buried in duplicates. Personal data retained indefinitely creates risks of breach and unjustified reuse. Migration costs continue to grow.
Archives therefore include accountable destruction. Under legislation and records authorities, the National Archives regulates disposal and selects records of enduring governmental or community value as national archives. Destruction can be a component of structured memory rather than its opposite.
Disposal must be reviewable. A person cannot delete relevant records once controversy appears, and automated retention needs litigation, investigation, and rights exceptions. Generative AI adds drafts and derivative material at scale, requiring institutions to distinguish evidence of a decision from transient output.
Why handover is more than a meeting
Handover is commonly compressed into a folder and one conversation near departure. The successor may not yet be appointed, leaving knowledge with an interim manager. At contract transition, commercial incentives may not favour full transfer. Files arrive but capability does not.
Critical roles need early identification of single-person knowledge. Joint work, rotation, exercise, and documentation reduce dependence. A manual should be tested by someone unfamiliar with the task. A disaster plan that has never been exercised proves only the existence of a document.
Handover should include unresolved conditions. Organisations prefer completion, but temporary controls, accepted risk, and dissent are often more important. Successors need to know where apparent normality should not be trusted.
The practical test is performance by the successor. Can that person locate the governing authority, reproduce a routine operation, recognise a known exception and identify whom to contact when the manual no longer fits? A signed checklist records that material was delivered; an observed exercise shows whether capability crossed the boundary. High-risk roles may need overlap and simulation rather than reliance on a final interview.
There is also a timing problem. Knowledge transfer that begins after resignation or contract termination is already constrained. Institutions need ordinary habits of recording reasons, pairing staff and reviewing single-person dependencies. Succession is more reliable when it is part of work rather than an emergency extraction from someone who is leaving.
Outsourcing, migration, and memory fracture
A supplier can accumulate system configuration, customer exceptions, and workarounds. At contract end, the agency receives source files or reports without rebuilding internal understanding. A new provider rediscovers the same faults while users bear transition.
Contracts should specify ownership, format, metadata, change history, knowledge transfer, and exit testing. The client must retain enough technical intelligence to assess delivery. If a public body relies entirely on a supplier’s account of its own performance, accountability weakens.
System migration creates similar risks. Successful import does not prove that meaning moved. Legacy codes, null fields, and old business rules need documentation, while relationships among records remain traceable. Sampling actual cases and reconstructing their decisions is more valuable than comparing record counts alone.
Restructuring moves functions and can separate records from people. Old links and contact paths stop working. Machinery-of-government change is an information and service-continuity project, not only an organisational chart.
Users often detect a failed migration before management. They are asked to resubmit evidence, receive contradictory histories or cannot establish a past entitlement. Complaint data and case reconstruction should therefore be part of migration assurance. A technically successful transfer is incomplete when the people whose rights depend on the record must rebuild it themselves.
Exit capability deserves funding throughout the contract, not only at its end. If the client cannot interpret exports, verify configuration or continue essential service without the incumbent, contract renewal is not fully voluntary. The retained internal knowledge need not duplicate every supplier skill, but it must be sufficient to govern dependency.
How memory can obstruct reform
Institutional memory can preserve obsolete bias and informal prohibitions. Experienced staff close proposals with “we tried that”. A procedure continues after its purpose disappears. Past success becomes evidence against new conditions.
Forgetting is not the remedy. Reform needs evidence of earlier consequences to avoid repetition. Memory should retain evidential force without becoming automatic authority. What happened constrains invention of the past, but does not determine what should happen now.
Records can also preserve only management’s view. If complaints, frontline dissent, and the experience of affected groups do not enter formal systems, the future receives a curated institutional autobiography. Information governance should determine which differing voices form part of decision evidence.
Will AI improve institutional memory?
Search, summarisation, and knowledge assistants can reveal relationships in large collections and return old records to use. They can also misstate, merge versions, or expose information beyond permission boundaries. A fluent response is not a substitute for the source and provenance chain.
Use of AI also needs records: sources, versions, retrieval configuration, and how outputs entered decisions. If only the final outcome is kept, future review loses a material part of the process.
Convenient assistants may reduce employees’ incentive to learn institutional context. Search should therefore operate with training and source verification, not outsource understanding to an interface unable to accept responsibility.
AI-generated summaries create a further preservation question. A useful summary may be replaced when the model or prompt changes, while later users assume it is a stable institutional statement. Important summaries should link to their source, identify their status and remain distinguishable from authorised records. Otherwise the system repeatedly rewrites the remembered past without making the revision visible.
Access control must apply at retrieval as well as storage. A model able to combine individually permitted fragments may expose a sensitive pattern that no ordinary search revealed. Memory becomes easier to use and potentially easier to misuse. Institutions need tests based on real roles and queries, followed by incident and correction routes.
Provisional judgment: a usable and challengeable past
Institutional memory does not mean keeping the greatest volume. It is the capability to find, understand, authenticate, and use past evidence. It relies on information systems, people, professional communities, and exercises. No repository can perform all of it.
Significant decisions should preserve evidence, reasons, authority, disagreement, and future obligations. Critical work needs tested succession. Outsourcing and migration should be complete only when a successor can continue service. Retention and destruction both require rules and oversight.
Memory must remain challengeable. It explains why an institution acted and allows new evidence to show that the old reason has expired. Without this corrective capacity, continuity becomes inertia.
Requirements will vary by risk and volume, and excessive documentation can remove time from service. Better automated classification may change cost. The minimum remains: where institutional power affects rights and public resources, its continuity cannot depend on whether one individual still remembers what occurred.
Primary sources and further reading
- National Archives of Australia, Building Trust in the Public Record, current policy extended to 2028.
- National Archives of Australia, Information Management Standard for Australian Government.
- National Archives of Australia, How we select national archives.
- Australian Government, Archives Act 1983.
- National Archives of Australia, Information management legislation.
Series navigation: Who Maintains the World? series overview
Discover more from Geoffrey Chen
Subscribe to get the latest posts sent to your email.