SWANK AI Guidance Note 18
Chronology · Organisational Memory · Record Integrity

Core Standard

TIME IS PART OF THE EVIDENCE

Organisational memory should preserve not only:

what happened

but also:

when it happened

what was known at the time

and

what changed later.

AI may help organisations retrieve and organise information.

It should not flatten chronology in ways that change meaning.

Purpose

Organisations depend upon memory.

In practice, institutional memory is rarely stored in one place.

It is often distributed across:

emails;
reports;
databases;
meeting records;
case notes;
policies;
internal correspondence;
decision records;
individual staff knowledge;
summaries;
and automated systems.

AI can help retrieve, organise and summarise this material.

It can also distort chronology if records are condensed without preserving sequence and context.

The relevant governance question is not simply:

What information exists?

It is also:

When did it exist, who knew it, and what happened next?

Chronology Is Operational Information

The order in which events occur may materially affect interpretation.

For example:

Decision → new evidence → review

is different from:

New evidence → decision → no review.

Both sequences contain:

a decision;
new evidence;
and a review.

But the institutional meaning is different.

A summary that preserves the events while losing their sequence may therefore preserve content while changing meaning.

Chronology is not decorative background.

It is part of the evidence.

Different Dates Mean Different Things

AI-assisted systems should distinguish, where relevant, between:

Event Date

When the underlying event occurred.

Record Creation Date

When someone documented the event.

Knowledge Date

When relevant information became known to the organisation.

Decision Date

When a decision was made.

Review Date

When the decision or information was reconsidered.

Correction Date

When an earlier record was amended, qualified or corrected.

These dates may be different.

A system that collapses them into one timeline may create a misleading institutional history.

What Was Known at the Time?

Later information can make an earlier decision appear obvious.

But assurance should consider what information was actually available when the decision was made.

A later reviewer may need to distinguish between:

what is known now

and

what was reasonably knowable then.

This matters because retrospective summaries can accidentally introduce later knowledge into earlier events.

AI-generated chronologies should therefore preserve temporal boundaries between:

contemporaneous evidence;
later interpretation;
subsequent correction;
and retrospective analysis.

Organisational Memory Is Distributed

Institutional knowledge may be fragmented across teams and systems.

Relevant information may sit in:

one team’s email inbox;
another team’s database;
meeting minutes;
case notes;
archived reports;
a staff member’s personal knowledge;
or a later summary.

No single record may contain the complete picture.

AI may help reconnect fragmented information.

But this is useful only if the system preserves:

source;
provenance;
chronology;
uncertainty;
and context.

Retrieval without temporal structure can create a more searchable record without creating a more accurate one.

Organisational Memory Failure

Institutional memory can fragment when:

staff change;
teams operate separately;
records sit in different systems;
summaries replace originals;
corrections are not propagated;
earlier context disappears;
or later users see only selected extracts.

Over time, an organisation may remember:

the conclusion

while forgetting:

how the conclusion arose.

This can weaken accountability and reassessment.

AI and Memory Reconstruction

AI can be useful for reconstructing:

chronologies;
decision histories;
communication sequences;
policy changes;
incident histories;
and evidence pathways.

But AI-generated reconstruction should not be treated as infallible.

A useful chronology should allow a reviewer to determine:

which source supports each event;
which date is being used;
whether an event remains disputed;
whether information was recorded retrospectively;
and whether relevant records may be missing.

An AI-generated chronology is still an analytical product.

It should remain traceable to evidence.

Recency Bias

AI and human reviewers may over-weight recent information.

Recent records may be:

easier to retrieve;
more visible;
better indexed;
or written more clearly.

Older information may nevertheless remain important because it establishes:

chronology;
baseline conditions;
earlier decisions;
prior understanding;
historical context;
previous corrections;
or recurring operational issues.

A governance system should not equate:

most recent

with

most important.

Outdated Information

The opposite risk also exists.

Historical information may remain prominent after circumstances change.

A record may have been accurate when created but no longer reflect the current position.

Systems should therefore distinguish between:

historically recorded

and

currently applicable.

An old classification should not silently remain operational forever because it continues to exist in the record.

Current Status Should Be Visible

Where circumstances change, the system should make the relationship between past and present clear.

For example:

Historical record: X was recorded in January.

Later evidence: Y emerged in March.

Current position: the earlier conclusion was revised in April.

This is stronger than simply deleting the historical record.

It preserves both:

institutional memory

and

current accuracy.

Corrections Are Temporal Events

A correction does more than change a fact.

It changes the history of what the organisation should rely upon.

A useful record should preserve:

the original statement;
the date it was made;
the date it was challenged;
the evidence supporting correction;
the date correction occurred;
and whether downstream conclusions were reconsidered.

If a correction is visible but the system cannot show when it occurred, later users may misunderstand which decisions were made using the inaccurate information.

Chronology and Reassessment

Reassessment depends upon knowing what changed.

A system should be able to identify:

the original evidence;
the original conclusion;
the new evidence;
the point at which circumstances changed;
whether review occurred;
and whether the institutional position changed.

Without chronology, reassessment can become difficult to distinguish from repetition.

A reviewer may see two different conclusions without understanding why the change occurred.

Chronology and Escalation

Escalation may also depend upon sequence.

For example:

Concern → further evidence → escalation

is different from:

Concern → intervention → consequences of intervention → further escalation.

The second sequence may require analysis of whether later records are independent evidence or partly consequences of the original institutional response.

Chronology can reveal feedback structures that are invisible in a static summary.

AI Summarisation and Temporal Drift

Repeated summarisation may gradually remove temporal detail.

An original record may say:

“At the time, the organisation had not yet received the later report.”

A later summary may say:

“The organisation had the report.”

A subsequent record may then imply that the report informed the earlier decision.

This creates temporal drift.

AI-assisted systems should preserve material phrases such as:

at that time;
subsequently;
later;
before;
after;
following receipt of;
prior to;
and after correction.

These words may carry significant evidential meaning.

Duplicate Events

Organisational records may describe the same event multiple times.

AI systems should avoid treating repeated references as separate events.

A chronology should distinguish between:

one event recorded in several places

and

several independent events.

Otherwise, repeated documentation may create:

false recurrence;
inflated significance;
or apparent corroboration.

Source mapping is therefore important alongside chronology.

Missing Dates

Real institutional records are often incomplete.

A system may encounter:

undated notes;
uncertain event dates;
retrospective records;
approximate dates;
or conflicting timestamps.

The appropriate response may be:

Date uncertain.

The available records do not establish the precise sequence.

A chronology should not invent temporal certainty merely because the output format expects an ordered list.

Conflicting Dates

Sources may disagree about when something occurred.

Where that conflict is material, the chronology should preserve it.

For example:

Source A records 12 March.

Source B records 14 March.

The available evidence does not presently resolve the difference.

This is stronger than choosing one date without explanation.

Time can be disputed evidence too.

Record Creation Is Not Event Occurrence

A common analytical error is to assume that the date of a document is the date of the event it describes.

For example:

A report written on 20 June may describe an event from 4 May.

An AI chronology that assigns the event to 20 June may materially distort the sequence.

Systems should therefore distinguish between:

when something happened

and

when it was documented.

Decision Chronologies

For consequential decisions, organisations should ideally be able to reconstruct:

what evidence existed;
what evidence was missing;
who knew what;
what role AI played;
when the decision was made;
whether challenge followed;
whether new evidence emerged;
and whether the decision was reconsidered.

This allows a later reviewer to understand not merely the outcome but the decision pathway.

Organisational Memory and Staff Turnover

When staff leave or roles change, institutional memory may weaken.

Important context may disappear because it existed partly in:

personal understanding;
informal communication;
or knowledge never fully documented.

AI may help preserve institutional continuity by making records easier to retrieve.

But it cannot recover knowledge that was never documented.

Governance should therefore consider what information needs to be recorded before it becomes dependent upon individual memory.

Organisational Memory and Policy Change

Policies also change over time.

A reviewer may need to know:

which policy version applied;
when a procedure changed;
whether staff were informed;
whether systems were updated;
and whether historical decisions were made under different rules.

AI systems should avoid applying current policy retrospectively to events governed by earlier arrangements unless that comparison is deliberate and explicit.

Version Control

Where records or policies are revised, version history supports accurate memory.

Relevant information may include:

version number;
publication date;
revision date;
author;
approval status;
superseded version;
and reason for change.

Without version control, AI retrieval may surface outdated material as though it were current.

AI-Generated Chronologies

Where AI generates a chronology, organisations should consider whether:

dates are verified;
events are ordered correctly;
disputed events are labelled;
source references are preserved;
omissions are visible;
uncertainty is retained;
corrections update the chronology;
duplicate references are recognised;
and current status is distinguished from historical status.

A chronology should make the institutional history easier to understand without pretending that incomplete records are complete.

Chronology and High-Stakes Decisions

Chronology becomes especially important where records may affect:

healthcare;
safeguarding;
education;
employment;
complaints;
disciplinary action;
public services;
regulatory intervention;
or legal rights.

In these environments, sequence may determine:

whether evidence was available;
whether review occurred;
whether an intervention was proportionate;
whether correction came before or after a decision;
and whether later information was properly considered.

High stakes require temporal discipline.

Reconstruction

A well-governed institutional system should enable an independent reviewer to answer:

What happened?

When did it happen?

When was it recorded?

When did the organisation know?

What decision followed?

What changed afterwards?

Was the record corrected?

Was the decision reconsidered?

That is organisational memory in operational form.

Questions for Organisations

Where AI assists with organisational memory or chronology, organisations may ask:

Are event date and record date distinguished?
Can we identify when information became known?
Is the policy version applicable at the time visible?
Are corrections shown in sequence?
Can later evidence be distinguished from contemporaneous evidence?
Are repeated records being mistaken for repeated events?
Are old classifications still being treated as current?
Can disputed dates remain visible?
Does AI retrieval over-weight recent records?
Can an independent reviewer reconstruct the decision history?
Are source references preserved in AI-generated chronologies?
Does the system show not only what happened, but what changed?

SIAAF Relevance

This Guidance Note principally relates to:

Domain 03 — Evidence & Traceability

Can the organisation reconstruct the evidence and chronology supporting an important conclusion?

Domain 04 — Communication & Feedback Integrity

Does new information move through the institution in a way that updates organisational understanding?

Domain 05 — Escalation, Challenge & Contestability

Can later evidence, corrections and challenges alter the institutional record?

It may also engage:

Domain 01 — Governance & Accountability

Where decision ownership and historical responsibility need to remain visible.

Domain 06 — Risk, Harm & Operational Resilience

Where outdated, incomplete or temporally distorted information may produce consequential error.

SWANK AI Standard

TIME IS PART OF THE EVIDENCE

Organisational memory should preserve:

what happened

when it happened

when it was recorded

when it became known

what decision followed

what changed

what was corrected

and

what remains current.

AI should help institutions remember more accurately.

It should not produce a cleaner chronology by removing the distinctions that give the chronology meaning.

Related SWANK AI Guidance

Guidance Note 02 — AI Summarisation and Record Integrity

Guidance Note 05 — Preserving Disagreement in AI-Assisted Records

Guidance Note 08 — Reassessment Alongside Escalation

Guidance Note 14 — AI Incident Reporting and Error Correction

Guidance Note 17 — Source Traceability by Design

Guidance Note 19 — Correction Rights in AI-Assisted Systems

SWANK AI

Independent AI & Institutional Assurance

We do not just review AI. We review the institutional systems responsible for governing it.