A decision record is not meeting minutes

Meeting minutes describe what was discussed. A decision record proves the operating position: the question that required resolution, the basis, the person with authority, the selected option, conditions, affected work, and the point at which the decision became effective.

The format can be simple. A controlled register, issue sheet, workflow record, or signed note may be enough. The strength comes from consistent fields, links to source evidence, revision control, and an owner who keeps the record current.

A useful record allows an independent reviewer to understand the decision without relying on memory, status, or personal reputation.

Documentation should support work, not paralyse it. The right level of detail depends on consequence, reversibility, uncertainty, interface complexity, and the cost of reconstructing the decision later.

The minimum defensible record

For every consequential decision, capture the following fields.

FieldRecordWhy it matters
Decision IDStable unique identifierLinks later records without relying on subject lines
QuestionThe issue that required a decisionPrevents the record from becoming a vague summary
TriggerRequirement, risk, condition, request, or eventExplains why the decision was necessary
OwnerNamed role with decision authoritySeparates advice from approval
InputsControlled records and material observationsReconstructs the information state
AssumptionsUnverified facts treated as trueShows where the decision could fail if reality differs
OptionsCredible alternatives and why rejectedDemonstrates deliberate judgment
DecisionChosen action and effective dateDefines the operating position
ConditionsLimits, checks, expiry, escalation triggersPrevents conditional approval becoming unconditional
ImpactAffected scope, records, interfaces, cost, schedule, riskControls downstream change
ApprovalAuthenticated authorizationShows the decision became effective
ClosureOutcome, validation, handover, supersessionCompletes the lifecycle
ProportionalityA reversible low-impact choice may need six lines. An irreversible safety, security, public, financial, or operational decision may require a signed assessment with specialist inputs and independent review.

A decision record should point to authenticated sources: issued drawings, calculations, requirements, field records, test results, correspondence, instructions, approvals, logs, photographs with provenance, and applicable standards. It should not overwrite those sources with an untraceable narrative.

For each linked item, preserve:

  • title and stable identifier;
  • author or producing system;
  • issue date and revision;
  • status and intended use;
  • where the authoritative copy is held;
  • whether authenticity or completeness has been verified;
  • the precise point the record supports.

Do not label an inference as evidence. “The log shows the account authenticated at 09:14” is an observation from a record. “The project manager authorized the change” is a different conclusion that needs a valid link between identity, authority, action, and context.

Evidence continuityThe ability to produce the authoritative record, explain its origin and status, connect it to the decision it supports, and show what happened to it through revision, handover, retention, and later review.

Request the record that changes the decision.

A large document collection can still leave the next decision unanswered. Work backward from the choice in front of you: what must you decide, what would change that choice, and which record could establish the missing fact? This is a practical research method, not a rule about anyone’s legal entitlement to records.

Write the question as a decision, not an accusation. “Can we reconcile this change against the approved scope?” gives the reviewer something to test. “Send everything proving wrongdoing” supplies a verdict but no useful search boundary.

Decision-to-record matrix: starting points, not a compulsory document list
Decision in front of youRecords to locate firstWhat still needs testing
Which scope was in force?Executed scope, amendments, exclusions and issue register for the relevant dateWhether a later instruction changed the working scope, and who could authorize it
Was this change authorized?Change identifier, request, technical disposition, approval and delegation limitWhether the approval covers this exact item, amount, revision and purpose
Which revision reached the work?Issued revision, transmittal, point-of-use record and implementation or field logSending a revision does not alone show that it was received or used
What caused the extra work?Original requirement, observed condition, inspection or test, instruction and priced changeWhether design, changed scope, installation, maintenance or another cause explains the work
What money actually moved?Invoice, payment allocation, credit and reconciliation against a named work itemWhether the amounts overlap, use the same tax basis and refer to the same period
Is another paid search worth commissioning?Existing record index, missing discriminator and the proposed search boundaryWhich answer would change the next step, and whether a narrower or free source can answer it
The useful next-record testComplete this sentence before a request: “If this record shows ___, we will ___; if it shows ___, we will ___.” If both answers lead to the same action, reconsider the priority before spending on retrieval.

One document can answer several questions. It can also answer none of them. Record the exact page, clause, item or event supporting the decision so the next reviewer does not have to repeat the whole search.

Give every statement a source status.

A confident sentence and a dependable source are different things. Use the following labels in a working brief. They describe how the statement was produced, not a numerical score or a guarantee that it is correct.

A small vocabulary for statements that can be checked
StatusWrite it this wayKeep attached
Source states“Invoice INV-07 describes item CH-02 as additional work.”The record ID, exact location, issuer, date, revision and authenticity limits
Independently checked“The amount in INV-07 matches the amount allocated to CH-02 in the payment record.”The second record, comparison method and any limits on independence or completeness
Inferred“These records suggest the same item was being discussed, but the identifiers differ.”The reasoning, credible alternative and record that could resolve the mismatch
Unresolved“No approval for CH-02 was located in the reviewed set.”The reviewed boundary, search terms, custodian response and next search

“Not found” is a search result. It is not the same as “never existed.” “The source states” does not mean that the source’s account has been independently verified. Two summaries repeating one invoice are still dependent on that invoice; counting them as three independent confirmations exaggerates the support.

The smallest useful source ledger

  1. Source ID. A stable identifier for the exact file, page, message or entry, including relevant attachments.
  2. Origin and custody. Who supplied it, the issuing organization or system, the collection date and the location of the unchanged copy.
  3. Record state. Date, revision, status, intended use and any uncertainty about authenticity or completeness.
  4. Statement supported. The narrow point it establishes, with an exact locator rather than a link to a folder.
  5. Limit and next test. What it cannot establish, any conflict, and what would confirm or weaken the statement.

Keep a correction trail when extraction or interpretation changes. If an unreadable scan initially appears to say 18,000 and a better copy says 13,000, preserve the initial extraction as a working error, correct the amount and identify every table or conclusion that depended on it. Do not silently replace the source.

An owner-ready records-request worksheet.

Complete the fields below before sending a request to the appropriate custodian. Remove fields that are not relevant. The worksheet helps explain the search; it does not create access rights, override confidentiality or authorize a fee.

A bounded request that the recipient can understand and route
FieldYour entry should identify
PurposeThe next decision and the one fact needed to make it
MatterProject name, address or system, contract or file number, and the exact item identifier if known
BoundaryStart and end dates, relevant decision or event, and known exclusions
Record categoriesA short named list, such as scope amendments, change approvals and payment allocations for CH-02
Likely locationKnown custodian or system, with “unknown” where you cannot identify it
Related materialRelevant attachments, revision identifiers and linked approval records, rather than all correspondence
Requested formatA usable existing electronic copy or an index first; preserve available dates, identifiers and issue status
Authority and handlingYour role, the basis for the request and a suitable transfer channel; avoid unrelated personal information
Cost gateWhether any retrieval fee is authorized; otherwise ask for an estimate and pause before chargeable work
Response logRequest date, reference, recipient, response, fee status and remaining gap

A request you can adapt

We are reviewing [project / file / item] to decide [specific next step]. Please identify the appropriate process for obtaining [named record categories] covering [start date] to [end date], including relevant attachments and revision identifiers. If this scope is too broad, an existing index or a suggested narrower first stage would help. No chargeable retrieval is authorized by this message. Please provide any proposed fee and search boundary for written approval before proceeding. If a requested category is not held or cannot be supplied, please identify the category and the appropriate next route where known.

Do not send the entire dispute file to every recipient. Start with identifiers and a non-sensitive explanation. Confirm the recipient, authority, permitted use and transfer method before sharing confidential material. Ask for qualified advice if the request concerns restricted records or a formal dispute process.

When the reply asks for a different route or more money

  1. Log the response. Retain its reference, named process, quoted boundary and fee without treating the reply as evidence about the underlying issue.
  2. Separate the options. Identify existing public records, an index-only stage, the narrowed paid search and any records held by another authorized custodian.
  3. Price the discriminator. Compare the proposed search with the precise missing fact. A cheaper search that cannot distinguish the explanations is not necessarily useful.
  4. Name the cost owner. Record who approves the fee, the cap, the deliverable and the stopping point. Do not commit another person or organization to funding it.
  5. Update the gap. After receipt, record which question was answered and which remains open. More pages do not automatically mean a stronger conclusion.
Check the money crosswalk

Control revision without erasing history

Projects weaken their own evidence when they silently replace files, edit old minutes, or allow multiple “final” versions to circulate. Correction is necessary; erasure is not.

  1. Preserve the issued state. Lock or snapshot the version that actually informed work.
  2. Issue the correction. Create a new revision with a visible reason and effective date.
  3. Map affected records. Identify drawings, schedules, instructions, models, procedures, and decisions that depend on the changed item.
  4. Notify by role. Send the change to the people who must act, not only a broad distribution list.
  5. Verify withdrawal. Confirm superseded copies are marked or removed from points of use while remaining preserved in the archive.

A revision log should distinguish authoring, review, approval, issue, receipt, and implementation. Those moments can be separated by hours or months, and the difference may control which information was available when an action occurred.

Turn informal communication into controlled decisions

Urgent decisions often begin in a call, chat, walk-through, or meeting. The communication channel is not the problem. The problem is allowing the temporary conversation to become the only record.

Close the loop with a short confirmation that states:

  • the decision or instruction understood;
  • the person believed to have authority;
  • the immediate basis and material assumptions;
  • the affected work and any hold points;
  • what remains unresolved;
  • where the formal decision record will be stored;
  • the time by which corrections must be raised.

Silence should not automatically be treated as approval unless the project’s agreed process explicitly gives it that effect. Where the consequence is high, obtain affirmative approval from the named authority.

Avoid self-serving reconstructionDo not create a retrospective record that pretends to be contemporaneous. If a later reconstruction is necessary, label its author, date, sources, gaps, and confidence. Preserve the underlying evidence unchanged.

Design retention around reconstructability

A record that cannot be found is operationally similar to a record that never existed. Retention must cover the decision register and its linked evidence, including the systems, exports, metadata, and access needed to interpret the material later.

Minimum controls

  • one authoritative repository for issued records;
  • role-based access and a documented record owner;
  • version history or immutable issued snapshots;
  • defined retention rules aligned with contractual, regulatory, operational, and legal requirements;
  • tested export and recovery, including metadata;
  • an offboarding process that preserves records when staff, vendors, or systems change;
  • a hold process that suspends ordinary disposal when a relevant event emerges.

Do not collect everything forever. Over-retention increases cost, privacy exposure, and search burden. Retain intentionally, protect sensitive content, and be able to explain what was preserved, what was disposed of under policy, and when.

The best time to improve a decision record is before anyone needs to defend it. The second-best time is now, while the sources and context still exist.