How to analyse open-ended event feedback without losing the detail

You've got the survey replies, and everyone wants the headline. Was the event good? What do we need to fix? Can we stop paying for the thing people barely mentioned?

Then you start reading. One attendee loved the programme but couldn't hear the final session. Another wants more networking. Someone else says the step-free entrance was locked. A summary saying "mostly positive, with some logistical issues" would be easy to write. It would also hide the decisions you actually need to make.

To analyse open-ended event feedback, keep the original comments, label the different experiences they describe, and count people consistently within each subject. Then review how serious each problem is, including the exceptions, before choosing what to investigate or change. A complaint can deserve immediate attention even when it appears only once.

The aim is a fair account that someone can check. You should be able to move from a recommendation back to the comments behind it, see where people disagreed, and explain what remains uncertain. That makes the analysis useful when your production lead, venue or supplier asks why you're proposing a change.

We'll work through a fictional paid event, including a comment-label template, example counts and a completed improvement decision. The figures and attendee wording are invented to explain the method. They aren't Loopyah customer results, research findings or targets for your own event.

Start with the decision the comments could change

Before opening a spreadsheet, write down what the team needs to decide. "Understand attendee sentiment" is too loose. "Decide whether entry needs a different layout before renewing the venue" tells you which experiences and records matter.

Don't restrict the analysis to that decision, though. You still need a way to notice something you didn't ask about. A survey intended to improve networking might reveal an access problem. Read every submitted comment for issues needing attention, then organise the remaining work around the decisions you can influence.

For our fictional example, imagine a recurring paid conference with 600 checked-in attendees. The team receives 180 survey submissions. Everyone submitting the survey was shown the optional question, "What most affected your experience of this event?" There are 120 substantive answers, 50 blank fields and ten answers such as "No comment". These are deliberately simplified figures.

The immediate commercial question is whether to approve a more expensive entry arrangement for the next edition. Other decisions concern sound coverage, catering and which parts of the programme to retain. The team needs enough detail to distinguish a staffing problem from an unclear entrance, and a disappointing meal from difficulty collecting it.

If you haven't collected the feedback yet, our guide to event survey question types covers how to choose the questions. Here, we're dealing with the words you've already received. Don't repair a weak question by pretending respondents answered a more precise one.

Write down the question exactly, including any instructions and routing. A reply to "What would you change?" is being invited to name a problem. A reply to "What did you enjoy?" is being invited to name a positive experience. Combining both into one mood score would lose that context before the analysis even begins.

Preserve the original and make a working copy

Save the original export with the event name, collection dates and questionnaire version. Work on a separate copy. If someone later asks whether "the queue" referred to lunch or entry, you need to read the original wording rather than a summary somebody polished during the review.

Use a simple response reference, such as R017, to connect a working entry with its source. That reference isn't proof that the data is anonymous. If you hold a lookup connecting it to a ticket holder, or the comment itself identifies somebody, handle it accordingly. Give access only to the people who need it for the agreed work.

For each answer, keep the question, original comment and any relevant context you legitimately collected. That might include the session selected in the form or whether the respondent bought a premium ticket. Don't add personal details just because they might make a future breakdown interesting.

Separate text that describes an experience from contact details offered for a reply. Your analysis team may need to know that somebody couldn't access a room without needing their phone number. The person handling the complaint can retain the contact information through your established process.

The UK Data Service's guidance on anonymising text explains why removing names may leave people identifiable through combinations of narrative details. A distinctive job, location or incident can reveal somebody. Check extracts before circulating them, and record any redactions without altering the experience being described.

Don't quietly rewrite grammar in the working source column. If you need a shorter version for the report, create a separate summary. Keep a difference between what the attendee wrote and what your team understood it to mean. That distinction matters most when the wording is awkward or the conclusion is disputed.

Agree how long you need the identifiable material and who will remove it when that purpose ends, consistent with your privacy arrangements and any relevant retention duties. The team needs a usable record, not a permanent archive of personal complaints shared with every future supplier.

Read complete answers before creating labels

Start by reading a varied set of complete submissions, then continue through the whole comment set. The early read helps you notice language people use and the connections they make. The full read prevents the first few memorable replies from defining everything that follows.

If you sort only by low ratings, you'll hear mainly from dissatisfied respondents. If you read only long comments, you'll favour people who had the time and confidence to explain themselves. Include short, positive, mixed and unclear answers. None of those groups is automatically less useful.

A comment such as "Great talks, but the room emptied into a queue and I missed the first part of the next session" describes more than a queue. It describes a sequence. A break, a service delay and lost programme time may be connected. Keep that sequence available while you label its parts.

The Office for Health Improvement and Disparities' guidance on qualitative analysis describes developing themes and revisiting them while reading the material. It specifically warns against treating repeated words as the whole analysis. The event application is straightforward: "waiting", "line" and "couldn't get served" may describe the same experience, while "queue" can refer to completely different services.

Write brief notes about possible subjects as you read. Entry, sound and catering are reasonable starting points. Don't decide yet that every entry comment proves you need more staff. A subject tells you where to look; it doesn't establish the cause or prescribe the purchase.

Keep a list of comments that resist your early categories. Perhaps somebody describes feeling unwelcome without naming a particular staff interaction. Perhaps a positive comment praises an unplanned conversation in the corridor. Those answers may require a new category, more context or an explicit "unclear" label.

This is also when you identify urgent reports. If the event is still operating, a safety concern goes to the responsible team immediately. If it has finished, follow your incident and complaint process promptly. Survey analysis mustn't become the reason a report waits for a monthly presentation.

Build labels that preserve what happened

Coding is simply assigning labels to passages of text. For an event team, a small written set of definitions usually works better than a long list of categories nobody can remember. The definitions should tell another reader what belongs in each group and what doesn't.

Start with the part of the event, then add a more specific experience when it changes the decision. Under entry, separate finding the entrance, waiting after arrival, ticket scanning and treatment by staff. Those problems can happen together, but their fixes are different.

For sound, distinguish difficulty hearing speech from discomfort caused by volume. "Bad sound" is broad enough to conceal opposite requests. Someone at the back may want clearer speech while somebody beside a speaker wants lower volume. Buying a louder system could make one person's experience worse.

Here's a short fictional starting set you can copy and adapt:

• Entry location. The attendee describes difficulty finding or identifying the correct public entrance. Exclude waiting once they have reached it.

• Entry waiting. The attendee describes a wait to enter after arrival. Keep their reported time separately, without treating it as a measured queue time.

• Speech audibility. The attendee describes being unable to hear or understand spoken content. Record the room or area if supplied.

• Sound discomfort. The attendee describes excessive or uncomfortable volume. Don't infer this from a generic low sound rating.

• Meal collection. The attendee describes finding, waiting for or receiving an included meal. Exclude judgments about taste unless the answer also mentions them.

• Programme usefulness. The attendee describes whether a session met their purpose or matched the advertised level. Keep topic suggestions separate where possible.

For every label, include an example and a boundary. An example helps somebody recognise it; a boundary stops them using it for every vaguely related complaint. Add a rule for unclear wording, and identify who can approve a new label.

You don't need a separate code for every unusual phrase. Several descriptions may express the same experience. Equally, don't force a new issue into "other" because changing your list would be inconvenient. Review that category before finishing. If it contains a coherent issue, define it and revisit the affected answers.

The CDC's chapter on qualitative data describes analysis as repeated reading, codebook development, discussion and revision. It also discusses examining alternative explanations before accepting a conclusion. Use those checks when reviewing your own interpretation.

Keep the definitions with a version date. If you split entry into location and waiting halfway through, check earlier answers against the new definitions. Otherwise, later respondents get more detailed treatment simply because the team learned how to label the data while working through it.

If comments arrive in several languages, preserve the original text alongside any translation. Ask a suitably fluent reviewer to check passages behind important findings, especially humour, understatement and descriptions of harm. A translation that turns "could hardly hear" into praise for quietness would change the sound decision. Record unresolved meanings rather than forcing every phrase into an English label. Use the same experience definitions across languages, while allowing people to express them differently.

Let one comment describe several experiences

Don't force every answer into one box. People experience an event as a connected visit, and a single reply may explain several parts of it. Assign each relevant label to the passage that supports it, while retaining the complete answer.

Take this invented answer from R017: "I couldn't see which entrance to use. Once I found it, getting my ticket scanned was quick. The first talk was worth coming for."

That supports a negative entry-location label, a positive scanning label and a positive programme label. Calling the entire comment "negative" loses the good scanning experience. Calling it "positive" hides the entrance problem. Neither summary tells the production lead what to preserve or change.

R042 writes, "The lunch line took up the break, so I got to the afternoon session late." Label meal collection and add a defined programme-access label for the reported missed session time. Don't label the food quality, because this answer says nothing about the meal itself. Don't label entry waiting just because the word "line" appears.

Record whether the passage describes praise, a problem, a suggestion or an unclear judgment. You can use those distinctions without creating an overall sentiment score. "Please keep the quiet room" is a request to retain something, and may be more useful than a generic positive label.

Mixed comments deserve a little extra care. "The speaker was engaging, but the material repeated last year" praises delivery while questioning usefulness for returning attendees. Keep both. Replacing the speaker would be a strange response if the actual issue was repeating introductory content.

For R017, a working record could say: response R017; question about what affected the experience; entry location negative; scanning positive; programme usefulness positive. Save each supporting phrase alongside its label. Count R017 once in the entry theme and retain both entry judgments. Add no claim about waiting time, because the reply gives none. This record lets the next reviewer check the classification without relying on your memory.

Avoid filling in missing motives. "I left early" does not mean the attendee disliked the programme. They may explain that their train changed, or they may give no reason. Record the stated experience and mark the reason unknown. You can investigate a pattern without inventing the explanation.

Agree how you will count before presenting percentages

Choose a counting rule that matches the claim. For a practical event review, I would usually count each identifiable respondent once within each theme, even if they mention that theme repeatedly. State whether that count combines several comment questions or covers only one.

In our simplified example, each of the 120 substantive answers comes from one distinct respondent to one open question. Suppose 30 mention entry and 24 mention sound. Some mention both. Those are counts of people mentioning each subject in this comment set, not separate slices of the audience.

A paragraph mentioning the entry queue three times still contributes one person to the entry count. If the same person describes entry in another question, don't count them again in an event-wide respondent total. You may still retain both passages because they explain different details.

Where you cannot reliably connect answers to the same respondent, say you're counting comments or submissions. Don't claim a unique-person count from an anonymous box that permits repeat entries. Don't try to identify people from their writing style to manufacture certainty.

The denominator, meaning the total you divide by, matters just as much. Thirty entry mentions out of 120 substantive answers is 25%. Out of 180 survey submissions, it is about 16.7%. Neither calculation establishes that the same share of the 600 attendees experienced an entry problem. Some people didn't reply, some left the box blank, and some discussed something else.

Also notice what the count at the top of that calculation says. We haven't yet separated favourable entry comments from complaints. "Thirty people mentioned entry" is accurate within the fictional dataset. "Thirty people complained about entry" would need a different count.

Write the denominator in words beside the finding. "Among the 120 substantive answers to the optional experience question, 30 mentioned entry" gives the reader more useful context than "Entry: 25%". You can include the percentage, but it shouldn't replace the explanation.

Theme percentages can total more than 100% because respondents can discuss several subjects. Don't force them to sum to a whole or divide them into a pie chart. Keep blank, explicit no-comment and substantive answers distinguishable in your method notes.

Pew Research Center's analysis of unanswered open questions shows that whether people answer can vary with the demands of the question and other factors. Its public-opinion research doesn't give an event response benchmark. It does reinforce why silence should remain missing information rather than being converted into satisfaction.

For the broader problem of who responded and what you can infer, see our guide to judging event survey sample size. More comments can reveal more experiences while still leaving important audience groups underrepresented.

Separate how often something appears from how serious it is

A frequency table can help you organise work. It cannot decide which problems are acceptable. A missing access route, a threatening interaction and a request for a different coffee supplier should not compete only on the number of mentions.

In the fictional conference, 20 respondents describe only entry problems, five describe mixed entry experiences and five describe only positive entry experiences. That makes 30 people mentioning entry, with 25 reporting a problem and ten mentioning something positive. The five mixed accounts belong in both judgments but only once in the theme total. Among the 24 sound mentions, 16 describe difficulty hearing and eight praise the sound in their area, with no mixed sound accounts in this example. One of the 25 respondents reporting an entry problem says the advertised step-free entrance was locked when they arrived. That report needs a specific review even if it never becomes a large category.

Record seriousness separately from frequency. Describe the consequence and the response needed in ordinary language. "Reported inability to use the advertised accessible route; operations lead to review with the venue today" is more useful than assigning a red score nobody has defined.

Don't present an unverified account as an established finding about a named worker or supplier. Preserve what was reported, route it to the responsible person, and investigate fairly. You can take a report seriously without claiming that every detail has already been confirmed.

For our example, the operations lead checks the route arrangements, venue handover and relevant staff records. If contact is possible and appropriate, the attendee can be invited to explain the experience through the agreed support channel. An anonymous response may limit follow-up, but it doesn't prevent the venue from checking the route and its opening procedure.

Keep this investigation separate from a general ranking of improvements. "Only one person mentioned it" isn't a reason to defer an access failure until after a popular menu change. Nor should an unresolved allegation be copied into a public sponsor report to demonstrate that the team is being thorough.

For routine service problems, seriousness still matters. Five comments describing a missed paid workshop may deserve more attention than a larger number asking for a different background playlist. Record the promised benefit, what people say happened and whether the problem prevented them receiving something they bought.

Avoid a scoring formula that makes responsibility disappear. Multiplying frequency by an impact rating can be a useful discussion aid if your team understands the assumptions, but a low total doesn't settle a safety, access or complaint decision. Name who is responsible for the next step and when it will happen.

Keep disagreement visible instead of averaging it away

You can receive accurate, opposing accounts of the same event. One person enjoyed the sound while another couldn't hear. Before deciding that the feedback is inconsistent, look at where and when each person experienced it.

Our fictional sound comments include 16 negative and eight positive accounts. Suppose the negative passages mainly name the rear seating area, while positive passages name seats nearer the stage. That pattern suggests a coverage problem to investigate. It doesn't prove the equipment was faulty or that every rear seat had the same experience.

Keep favourable accounts attached to their context too. They may tell you what worked and where. A report saying "sound was poor" could lead the production team to replace a setup that worked in much of the room. A more precise finding would describe the reported difficulty in the rear, the contrasting front-area accounts and the location details still missing.

Read the answers that challenge your preferred explanation. If somebody in the rear heard clearly, check whether they attended the same session. If the comments don't provide that information, retain the uncertainty. Don't move an inconvenient answer into "other" to make the pattern cleaner.

Ticket groups and attendance history can provide useful context when you have a legitimate, reliable connection. Returning conference attendees may describe familiar material as repetitive while first-time visitors find it useful. The programme decision could involve clearer session levels or different content, rather than simply replacing the tutor.

Use small groups carefully. If only a few people bought a package, a detailed breakdown can identify them and make minor numerical differences appear dramatic. Describe the limited evidence and investigate the experience. You don't need to publish every available cross-tabulation to make an operational decision.

Don't infer characteristics from names, writing style or complaints. A person mentioning a locked step-free route hasn't necessarily disclosed a disability, and you shouldn't invent a medical category for them. Analyse the route failure they described. Collect further personal information only when the specific follow-up needs it and your process supports it.

Review the labels before trusting the totals

Ask another person to label a varied selection without first seeing your choices. Include easy answers and the awkward ones, especially mixed experiences, unclear references and reports with serious consequences. Then compare the reasoning behind your differences.

The point isn't to prove that your labels are objective. It's to discover where the instructions leave too much room for interpretation. If one reader labels every queue as entry while another distinguishes meal collection, the counting problem starts with the definitions.

For a small event business, a colleague who understands the delivery can provide a useful challenge. They should still read what was written rather than explain away every complaint with something the production team remembers. Keep operational evidence in a separate column so it can be compared with the account fairly.

Where you disagree, return to the complete answer and the question it followed. Decide whether the rule needs changing, the answer needs several labels, or the meaning should remain uncertain. Write down the decision and apply it to other affected answers, including ones already completed.

If you're working alone, leave time between the first labelling pass and the review. Revisit a varied selection with the definitions in front of you, then check every high-consequence finding and every extract used in the report. This doesn't replace an independent review, but it is better than treating a first impression as a final classification.

Check totals against the source before presenting them. Count distinct respondent references for each theme, confirm which question supplied them, and inspect a few underlying passages. Look for duplicated imports, repeated submissions and labels that were renamed but never combined. Keep your decisions about suspected duplicates visible rather than deleting uncomfortable answers.

An identical phrase doesn't establish a duplicate. Two people can both write "Great event". Remove a response from the analysis only when you have a defensible reason under your stated rules, and retain a record of the exclusion. When identity is uncertain, qualify your count instead of guessing.

Use AI to assist the reading, with checks you can repeat

An AI tool can suggest labels, group similar passages or prepare a draft summary. The output is a proposed reading of the comments. You still need to check whether the proposed themes and counts are supported by the source material.

Before uploading anything, confirm that the service and its settings are approved for the data you hold. Consider the privacy notice you gave respondents, access permissions, retention and how the provider uses submitted text. Don't assume that deleting a name makes a complaint safe to upload. If the arrangement doesn't support the material, keep it out of that tool.

Use a permitted working dataset with only the context needed. Retain response references so every suggested label can point to an original passage. Ask the tool to return those references with its reasoning, and to mark ambiguous answers rather than complete their meaning.

A useful instruction for an approved tool is: "Apply these definitions to each response. Return the response reference, relevant text and proposed labels. Allow several labels. Do not infer causes or personal characteristics. Mark uncertain wording for human review. Keep serious reports visible regardless of frequency."

Try that on a small, varied set you've already read. Check whether it confuses ticket-purchase difficulty with entry at the venue, counts the same person twice, or treats a polite opening as evidence that the whole comment is positive. A fluent paragraph can conceal those mistakes.

Review all comments behind decisions with serious consequences, every unclear case and the evidence used in your final report. Check a varied selection of the remaining classifications too. If the check exposes a repeated error, revise the instructions and revisit the affected set. Don't extrapolate a reassuring check of easy praise comments to the complaints you haven't inspected.

Count labels from the verified records using a reproducible filter or calculation. Don't rely on a conversational answer to "How many people said this?" unless you've reconciled it with the source. Keep the tool version, instructions and review notes where practical, so another person can understand how the result was produced.

You may find that a spreadsheet and careful reading are quicker for a modest comment set. That's a reasonable outcome. Choose the method that gives you a defensible answer at the scale you have, rather than adding a tool because the export looks long.

Turn the finding into an investigation before approving spend

The next step is to distinguish the experience, your explanation and the proposed action. Those are different statements, and the evidence needed for each becomes stronger as you move towards spending money.

For entry, the fictional comment review finds 25 negative accounts among 120 substantive answers. The passages include difficulty finding the entrance and waiting after arrival. The team should report the separate subthemes once verified; it shouldn't treat every negative entry account as a staffing complaint.

The observation is that respondents described particular difficulties. A possible explanation is that unclear signs caused people to reach the wrong door. Another is that a short arrival surge exceeded the service arrangement. A proposed action might involve signs, staff or a different entrance. The comments haven't yet chosen among them.

The Government Digital Service's guide to analysing research sessions distinguishes observations, findings and actions. For an event business, that separation helps stop a manager's favourite solution being written into the report as though attendees had demonstrated it was necessary.

Ask the operations lead to compare the comments with the entrance plan, opening times, recorded incidents and observations from the door. Check what the records can actually establish. A ticket-scan timestamp may show when someone was scanned, but without an arrival observation it doesn't tell you how long they waited.

Suppose the investigation confirms that the entrance signs weren't visible from the main approach, while the available records don't establish the length of the queue. The team can approve a sign improvement and plan better observation next time. It doesn't have to claim that signs explain every complaint or promise that extra signs will remove all waiting.

Write a decision record that preserves this limit. "Improve approach signage; queue duration still unmeasured" is a useful result. "Feedback solved" is not. The remaining uncertainty should influence what you observe at the next event.

Our event debrief template can help you bring the relevant owners together. Send the evidence before the meeting, including the contrary comments. Don't use the meeting to make everyone read an unfiltered export for the first time.

Cost a small test and decide what success would look like

A finding becomes commercially useful when the proposed response has an owner, a cost and a way to check whether it helped. That doesn't mean every comment needs a financial return attached. It means routine improvement spending deserves a clear reason.

Here's a separate, invented costing for the entry decision. The organiser has a quote of $180 for revised signs, $140 for additional setup work and $80 for checking the approach before doors. The proposed test costs $400 in total. These are illustrative US-dollar costs, including assumed applicable purchase taxes, not market rates or supplier quotes you should expect.

For comparison, suppose an alternative extra staffing arrangement would cost $900 on the same assumed basis. Choosing the $400 option would mean spending $500 less than that alternative. It would not mean the event has saved $500 against its existing budget or earned any additional revenue. Neither option has yet been shown to fix the problem.

Don't convert 25 negative comments into 25 customers you will lose, then multiply by the ticket price to justify the project. You haven't measured their future purchases or the reasons they may return. If the business case depends on retention, state that as an uncertainty and track actual behaviour with appropriate records later.

For the sign test, decide what you will check before buying. The operations lead can walk the actual approach under expected conditions, observe whether arrivals find the entrance, record where people ask for directions and ask the same relevant experience question afterward. Keep queue monitoring separate so the cheaper change doesn't obscure another problem.

Allow for failed or inconclusive tests. If the event moves venue, changes the arrival route or receives far fewer arrivals, comparison becomes harder. Record those differences. A lower complaint count can reflect fewer replies or a different audience as well as a better entrance.

Where an access or safety investigation requires action, follow the responsible process rather than waiting for a commercial experiment. The small-test approach applies to suitable routine improvements. It doesn't turn essential arrangements into optional spending that must first prove a revenue benefit.

Use a record the next team can understand

You don't need an elaborate dashboard to preserve the reasoning. You need consistent fields and enough context for someone who didn't attend the review. Start with one record for each finding that might change a decision.

A reusable finding record can include:

• Event, date, question wording and collection period.

• Source set, response counts and the rule used for counting people or comments.

• Theme definition, relevant response references and favourable or conflicting accounts.

• What the comments support, what remains unclear and how serious the issue may be.

• Other evidence checked, the person responsible and the decision date.

• Approved action or further investigation, cost, delivery date and next-event checks.

For our fictional entry finding, the source set is 120 substantive answers from 180 survey submissions after an event attended by 600 people. Thirty respondents mention entry: 20 problem-only accounts, five mixed accounts and five positive-only accounts. That means 25 people describe a problem and ten describe something positive, with the five mixed respondents in both judgment counts. Those counts describe the comments, with each respondent counted once within the theme. They aren't an estimate of the event-wide problem rate.

The original response references lead to passages about entrance location, waiting and scanning. The location investigation supports revised signs. Waiting remains under review because arrival-to-scan duration wasn't recorded. The step-free entrance report, included in the entry count, follows its own access investigation, rather than being absorbed into the routine sign project.

The approved routine action is the $400 sign test, owned by the operations lead. The record names the pre-opening route check, observations during arrival and the follow-up question. It also preserves the $900 alternative as an option considered, without claiming either option has demonstrated a financial return.

You can include that finding in a wider post-event report, alongside actual revenue and costs. Keep the detailed working comments in a controlled location. The report needs the evidence and its limits; it doesn't need every identifiable sentence from the export.

Choose extracts that explain the finding fairly. A particularly angry or eloquent response isn't automatically representative. If you shorten a real comment, preserve its meaning and make omissions clear. If you paraphrase, label it as a summary. A marketing testimonial is a different use from an internal service review and needs its own appropriate arrangements.

Tell people what you changed, then check the next edition

After the review, tell attendees what you can confirm. "We've approved clearer signs on the approach and will check the route before doors" describes an action. "You spoke, we listened" leaves them guessing about whether anything happened.

Be equally clear about work still underway. You can say you're reviewing sound coverage in a particular room without promising new equipment. Don't publish details that identify a complainant, expose a worker to an unresolved allegation or imply that an individual case has been settled when it hasn't.

When you use Loopyah's attendee email tools, you can build the message and send yourself a test before it reaches the selected audience. Check the wording and recipient selection for the purpose of that send. Survey interpretation and individual complaint handling still need their own responsible process; don't assume the mailing tool has resolved them.

At the next edition, reopen the decision record before building a new survey. Check that the promised change happened, record the relevant operating conditions and keep the question comparable where possible. Review the new comments using the same definitions, noting any changes to the method.

If fewer respondents mention entrance difficulty, describe the new count with its own denominator and context. Compare observations as well as comments. If the result is mixed, keep the action under review rather than selecting a flattering quote and closing the file.

Before the next venue review, ask the operations lead to complete the entry finding record with the original comment references, the route-check result and the unresolved queue question. You can then discuss the sign test with the venue using the evidence you have, while keeping the access investigation and the missing arrival measurements on the agenda.