Evidence vs Interpretation in Audit Findings

"Tooling control not effective."

"Staff were unaware of the procedure."

"Occurrence follow-up was inadequate."

Each of these audit notes may point at a real problem. None of them would survive a challenge from the auditee. What was checked? Which record was missing? Who was interviewed, and what did they actually say? Which procedure applied? Was the problem seen once, across a sample, or inferred from something else?

The notes sound confident because they are doing too much at once. Each packs a fact, the evidence for it, an assumption and a judgement into one sentence. A defensible finding pulls those apart, and keeps the decision that follows in a separate place again.

Five things that should not share a sentence

Messy capture is normal. An auditor scribbles after an interview; a safety manager takes notes on an occurrence call; a CAMO engineer jots a concern about repeated deferred defects. The trouble starts when those notes become official wording without being unpacked.

"Engineering records are incomplete and staff do not understand the procedure." Before that sentence goes into a report it needs pulling apart. Which records? What sample? What was missing? Which procedure, which staff, and what did they say? Is the conclusion based on a document review, an interview, a system check or the reviewer's judgement? Left blended, it invites an argument about tone instead of a fix, and the action may target training when the real weakness is where the record is kept.

Facts: narrow enough to test

A fact is specific enough that another competent person can see exactly what is being claimed.

"Three sampled continuation training records did not show completion dates" is a fact. "Training control is poor" is an interpretation. "The occurrence was closed before the investigation action was completed" can be a fact, if the closure date and action status were checked. "The safety process is ineffective" is a conclusion that needs much more work.

Good fact statements name the object, the sample, the date or reference and the mismatch:

• two of five sampled supplier evaluation files did not contain approval evidence;
• the work pack for MSN-EXAMPLE-0412 did not include the referenced independent inspection sign-off;
• the management review minutes recorded the action, but no owner or due date;
• the training matrix listed the role, but the sampled employee file did not show the recurrent item;
• the occurrence report contained a recommendation, but no closure evidence was provided.

They are not dramatic. They are testable.

Objective evidence: how the fact is known

Objective evidence is the material behind the fact: a record sample, a screenshot, a checklist entry, a manual paragraph, an interview note, a system export, a technical log entry, a training record, meeting minutes, or something observed on the floor.

Keeping it visible stops overreach. If the only support is one interview, the note should not read like a verified organisation-wide failure. If the sample is three records, the finding should not imply every record is affected. If the material came from someone's notes rather than the controlled system, say so until it is verified.

A working note might look like this:

• fact: two sampled tools had expired calibration labels;
• evidence: audit photo references and tooling register extract provided by the tool store;
• scope: line station sample on the audit date;
• unknown: whether the central system had updated calibration data;
• reviewer action: confirm tool status before classification.

That structure does not weaken the issue. It makes it harder to dismiss.

Assumptions: label the leap

Assumptions are not the problem. Hidden ones are.

A missing record may mean the task was not done. It may also mean the record sits in another system, the file was incomplete on the day, or the sample was not representative. A late corrective action may show poor follow-up, or an approved extension that was not in the pack. An interview answer may show lack of awareness, or a badly asked question.

The discipline is to write the leap down:

• known: the sampled file did not contain the required record;
• unknown: whether an approved record exists elsewhere;
• assumption: the sampled file should be the official location;
• question: confirm the record location and retention rule;
• stop point: do not classify until the owner responds.

Interpretation: professional judgement, stated as such

Interpretation is where the auditor explains what the facts may mean, and it is valuable. It is also where the same evidence can lead to different conclusions depending on context, exposure, recurrence and operational impact.

A missing signature might be administrative. It might also remove the only proof that a required check was done. A procedure gap may be harmless wording, or it may produce inconsistent shift handovers. That is why interpretation sits after the evidence, not inside it:

• evidence: three occurrence reports had no documented effectiveness check after closure;
• interpretation: closure criteria may be applied before learning actions are verified;
• risk note: if confirmed, the SMS may not be demonstrating feedback from occurrence management;
• decision needed: safety manager to confirm classification and follow-up route.

The language stays cautious without going vague. The reader can see exactly where judgement begins.

Recommendations are proposals; decisions are accountable acts

A recommendation suggests a next step. A decision commits the organisation.

An auditor can recommend that a procedure owner define record retention. A compliance manager can propose that a finding stays open until evidence is sampled. Closure, acceptance, risk acceptance, procedure release and authority submission belong to the roles and processes the organisation has defined. Keep the words apart. "Recommended action" should not drift into "approved action". "Proposed closure evidence" should not become "finding closed".

Before and after: tooling control

Weak:

Tooling control is not effective.

Defensible:

• requirement: tooling control procedure, section as referenced by the auditor;
• evidence: two sampled calibrated tools showed expired labels; one register entry could not be reconciled;
• gap: local control did not demonstrate current calibration status at point of use;
• scope: one line station, audit date sample;
• uncertainty: central calibration status not yet confirmed;
• proposed classification: subject to quality manager confirmation.

The second version is less dramatic and much stronger. The auditee knows what to answer and the reviewer knows what to confirm. Written as a single finding statement:

At the ExampleAir line station, two of the calibrated tools sampled on the audit date carried expired calibration labels, and one tooling register entry could not be reconciled with the tool in use. The referenced tooling control procedure requires calibration status to be demonstrated at point of use. Central calibration records have not yet been checked.

EASA's own example of a level 2 finding against an observation, on the same subject, is quoted in the tool control guide.

A one-screen template

Copy this into the audit notes before anything is worded as a finding:

REQUIREMENT     Source, section, revision
SCOPE           Area, activity, date, sample size
FACTS           What was seen or not found, one per line
EVIDENCE        Record refs, photos, extracts, interviews
ASSUMPTIONS     What is taken as true but not checked
UNKNOWNS        What still needs confirming, and by whom
INTERPRETATION  What it may mean (auditor judgement)
GAP             Requirement vs evidence, in one sentence
DRAFT WORDING   Neutral finding statement
DECISION        Classification and issue: reviewer name

If a line cannot be filled, that is the next question to ask, not a gap to smooth over.

Separated findings produce better corrective actions

Vague findings create vague actions. If a note mixes evidence, opinion and solution, the auditee may fix the wrong thing. A training course will not solve a missing system control. A checklist will not fix unclear accountability. A procedure rewrite will not help if the procedure is fine and the record is filed in the wrong place. A reminder email may close the paper trail and leave the process unchanged.

When the finding shows what happened, how it is known, what it may mean and what is still unconfirmed, the organisation can tell containment from correction, root cause from symptom, and completion evidence from effectiveness. The next step is covered in how to write an aviation corrective action plan, and a worked root cause analysis in Root Cause Analysis and CAPs for EASA Audit Findings.

Where AI fits

The same separation is what makes AI drafting safe or unsafe. A tool that smooths "two tools had expired labels" into "tooling control is not effective" has done the damage in one sentence. A tool that asks for the requirement, sample and evidence, and labels assumptions and interpretation, is doing useful preparation. The guardrails are covered in can AI draft aviation audit findings safely?

In Avioverse, findings are worked in that shape inside the Audits module, and classification, issue and closure stay with the auditor and the organisation's approved process. Separation is not paperwork for its own sake. It keeps confidence from outrunning proof.