Privacy and preparation
Dataset missing attachment audit: when the ticket says see attachment
By SourceX Editorial · Updated
10 sections
Short answer
Audit in two passes. First find every ticket that refers to a file and record whether that file is present, intact and on the right record, or expired, redacted, deleted or never uploaded. Then grade whether the recorded resolution still makes sense without it. Exclude or relabel cases whose decisive evidence is gone, keep supporting gaps with a flag, and report counts by cohort in the delivery notes.
Checklist
Dataset missing attachment
Key takeaways
- A ticket that says see attached is a claim to verify, not proof a file exists.
- Record why a file is absent; expired links, redaction and deletion have different recovery paths.
- Never count a redacted placeholder or a zero-byte file as present.
- Grade impact by asking whether a reviewer could reach the resolution from the text alone.
- Check parent links and upload times, because readable files can still be the wrong evidence.
- Report gaps by cohort and narrow the scope where decisive gaps cluster.
How do you audit tickets whose required attachments are missing?
Audit in two passes: availability, then impact. First find every ticket whose text or metadata says a file should exist and check whether that file is present, intact and attached to the right record. Then grade how much the case depends on the missing file, because a lost screenshot of a settings page and a lost signed form are not the same defect.
At a glance
How do you audit tickets whose required attachments are missing
The output is a per-ticket worksheet and a cohort summary. Tickets whose decisive evidence is gone are excluded or relabeled, tickets missing only supporting files are kept with a flag, and the counts go into the delivery notes so the buyer's acceptance decision rests on measured facts instead of a sample that happened to open.
- Scope: fix the ticket population and date range and reconcile the ticket count to the source system.
- Detect: flag every ticket that references a file through metadata, inline image references or text.
- Check availability: record a state for each expected file and a hash for each file you hold.
- Grade impact: decide whether the case still makes sense without the file.
- Decide and report: apply the action for each grade and publish counts by cohort.
Which signals show that a ticket depends on a file?
Metadata is the strongest signal, text the weakest, and you need both. Helpdesk APIs attach file records to comments: Zendesk ticket comments carry attachment objects, and the Jira Cloud REST API returns attachment metadata separately from the file content. Email-sourced tickets add inline images, which Microsoft Graph marks with an isInline flag and a contentId, and which HTML bodies reference with cid: URLs defined in RFC 2392.
Text references catch what metadata misses, such as a customer who writes see attached when the upload failed or an agent who says the log you sent shows a timeout. They also produce false positives, so treat text-only hits as candidates that a reviewer confirms on a sample.
| Signal | Where it appears | Strength | False positive risk |
|---|---|---|---|
| Attachment record on a comment | Helpdesk API or export, per comment | Strong: the system recorded an upload | Low; but a record can exist while the file is gone |
| Inline image reference | HTML body with a cid: or embedded image tag; email attachment marked inline | Strong for email-sourced tickets | Medium: signatures and logos look like content |
| Phrases such as see attached, screenshot below, log attached | Ticket description and comments | Medium | High: customers say it when the upload failed |
| Agent reply that quotes or describes a file | Agent comments | Medium: shows the file mattered to the resolution | Low |
| Link to an external file host | Comment text | Medium | Medium: links often point to public docs, not evidence |
What availability states should the audit record for each expected file?
Record a state, not a yes or no. A file can be absent for several reasons, and each one has a different recovery path and a different meaning for the buyer. Several states are common in ordinary exports.
Zendesk's account data export, for example, lists attachments as links to the attachments API rather than including the files, and Zendesk warns that a stored content_url expires and then returns a 403 Forbidden error. A redacted Zendesk attachment is replaced by an empty file named redacted.txt and the redaction cannot be undone, and attachment deletion schedules can remove files from closed, archived tickets while the ticket itself remains. Each of those looks like a file to a naive script.
| State | How to detect | Recoverable? | How to record it |
|---|---|---|---|
| Present and intact | File downloaded, opens, size above zero, hash computed | Not needed | Hash, size, media type, parent comment ID |
| Link only, expired | Export holds a URL; fetching it later fails with an authorization error | Often, by re-requesting through the API while the account is active | URL, fetch time, error code, retry result |
| Redacted placeholder | File named redacted.txt or empty placeholder in place of the original | No; redaction is permanent in tools that document it as such | Placeholder name, comment ID; never count as present |
| Deleted by policy or user | Attachment flagged deleted, or a deletion schedule covers the ticket | Only from a backup taken before deletion | Deletion source and date if known |
| Inline image lost | cid: reference or inline flag with no matching body part | Sometimes, from the original mailbox | Content ID, message reference |
| Zero byte or corrupt | Size is zero, file will not parse, or type does not match extension | Rarely | Observed size and parse error |
| Never uploaded | Text says attached; no attachment record anywhere in the thread | No | Text snippet that triggered the flag |
The attachment availability and impact audit worksheet
One row per expected file, linked to its ticket and comment. The worksheet below lists the columns, allowed values and the rule each column enforces; it is version 0.1, dated 2026-10-18. Use a cryptographic hash such as SHA-256 from the Secure Hash Standard, FIPS 180-4, so a later delivery can prove that the file the buyer received is the one you audited.
At a glance
The attachment availability and impact audit worksheet
| Column | Allowed values or format | Rule |
|---|---|---|
| ticket_id, comment_id | Source system IDs | Both resolve in the source system; comment belongs to the ticket |
| reference_signal | metadata, inline, text, agent_quote, external_link | Text-only rows need reviewer confirmation on a sample |
| expected_file | File name or description from the reference | Free text kept short; no personal data copied from the file |
| availability_state | One of the seven states above | Redacted placeholders and zero-byte files are never present |
| sha256, size_bytes, media_type | Hex digest, integer, detected type | Required when state is present; detected type, not extension |
| uploaded_at vs resolved_at | Timestamps in UTC | Files uploaded after resolution are flagged, not used as evidence for it |
| impact_grade | decisive, supporting, incidental, unknown | Assigned by a reviewer who read the thread, not by a script |
| action | keep, keep_flagged, exclude, relabel, recover | Must match the default for the grade unless an exception note is filled |
| exception_note, reviewer_role | Short text, role name | Required whenever action differs from the default |
How do you grade how much a case depends on the missing file?
Ask one question per ticket: could a qualified reviewer reach the recorded resolution from the remaining text alone? If not, the file was decisive. If the text alone gets there but the file would add confirmation or detail, it was supporting. If the resolution never relied on it, it was incidental. Grade a sample twice with two reviewers and report how often they agree before trusting the grades at scale.
| Impact grade | Test | Illustrative example | Default action |
|---|---|---|---|
| Decisive | Resolution cannot be followed without the file | Agent says the error code in your log is the cause, and the log is gone | Exclude from evaluation; relabel or exclude from training |
| Supporting | Resolution follows from text; file confirms it | Customer describes the error in words and also attached a screenshot | Keep with a flag |
| Incidental | File played no part in the resolution | Email signature image or company logo | Keep; drop the reference |
| Unknown | Reviewer cannot tell from the thread | Short thread with see attached and a one-line fix | Keep out of evaluation until a second review |
What defect could change acceptance even if every file opens?
Readable files can still make a package unusable for its declared task. The most common defects are structural, and an availability check alone will pass them.
At a glance
What defect could change acceptance even if every file opens
Test for them on the same sample you graded. Join each file to its parent comment by ID rather than by file name, compare upload time with the ticket's resolution time, and group hashes across the whole sample to spot repeats. A buyer running acceptance tests will often do exactly this, so finding these defects first keeps a delivery from being rejected for a problem that a short script could have caught.
- Wrong parent: the file is attached to a different ticket or comment than the one that refers to it, often after a migration merged threads.
- Timing leak: the file was uploaded after the ticket was resolved, so it reflects the answer rather than the evidence available at the time.
- Placeholder counted as present: redacted.txt or a zero-byte file passes a does-it-exist check.
- Duplicate or recycled file: the same hash appears across unrelated customers, which usually means a template or a mis-linked upload.
- Cross-customer content: a file belongs to another account than the ticket's requester, which is both a quality defect and a privacy problem.
- Type mismatch: the extension says image but the content is a document or archive, so downstream processing silently skips it.
Worked example: auditing a helpdesk sample before delivery
This example is illustrative, with invented counts, and describes no real company. A software company draws a random sample of 400 closed tickets from three years of helpdesk history. Detection flags 130 tickets that each expect one file: 95 through attachment records, 22 through inline images in email-sourced tickets and 13 through text alone.
Availability checks find 88 files present and intact, 19 links that had expired in an old export but recovered through the API, 9 redacted placeholders, 8 deleted by a retention schedule on archived tickets and 6 never uploaded. Reviewers grade the 23 tickets still missing files: 7 decisive, 11 supporting, 3 incidental and 2 unknown.
The exception: 3 of the decisive tickets have an agent reply that transcribes the key lines of the missing error log. The reviewers regrade them as supporting with an exception note, because the evidence the resolution relied on survives in text. The cohort view then shows most deleted files sit in the oldest year, so the company narrows the offered range rather than flagging a whole year of tickets one by one.
What should the cohort summary in the delivery notes report?
Report rates by cohort, not one total, because missing files cluster. A migration year, a channel such as email, or a product line with a short retention schedule can hold most of the gaps while the rest of the data is clean.
At a glance
What should the cohort summary in the delivery notes report
Keep the worksheet itself with the delivery record. The summary tells a buyer how large the problem is; the worksheet lets them check any single ticket, and it gives the next refresh a baseline to compare against.
- Reference rate: share of tickets that expect at least one file, by cohort.
- Availability rate: share of expected files present and intact, by cohort and by state.
- Decisive-missing rate: share of tickets graded decisive with no recovered file, the figure most likely to move acceptance.
- Reviewer agreement on impact grades for the double-graded sample.
- Actions taken: counts kept, flagged, excluded and relabeled, reconciled to the scoped ticket total.
- Known causes: retention schedules, redaction, migrations, with dates where known.
Narrowing the scope when gaps cluster or recovery gets risky
Narrow the scope when decisive gaps concentrate in one cohort; it is cleaner to drop a year or a channel than to ship a cohort where many cases cannot be followed. Pause and involve IT when recovery means restoring old backups, since a restore can bring back records that were deleted on purpose, including files removed for privacy reasons.
Bring in privacy review when recovered or present files contain screenshots, scanned forms or spreadsheets from customers, because file content often carries personal data and third-party material that text redaction never touched. And get counsel to look at any file that came from a customer's own systems before it is offered outside the company.
If your ticket history is a licensing candidate
Running this audit on a sample is useful preparation, but a company does not need it to learn whether it qualifies; SourceX begins with a brief company assessment and asks for no data at that stage. Where file collections are very large, they can stay in the company's storage or ship on encrypted drives, since SourceX does not host them. An audit worksheet like the one above later helps shape the scope a company approves before anything is delivered.
Frequently asked questions
What is a missing attachment audit?
It is a check of every ticket that refers to a file, recording whether the file is present, intact and on the right record, and how much the case depends on it. The result is a per-ticket worksheet plus counts by cohort that a buyer can use in acceptance testing.
Is a redacted attachment the same as a missing one?
For training and evaluation purposes, yes. A placeholder such as an empty redacted.txt file carries none of the original evidence, so the audit records it as redacted and never counts it as present.
Can expired attachment links be fixed?
Often, while the source account is still active. Re-request the file through the system's API with valid credentials, download it immediately and store it with its parent ticket and comment IDs instead of keeping the temporary URL.
Should tickets with missing attachments be thrown out?
Only when the file was decisive for the resolution. Tickets missing supporting or incidental files usually stay in, with a flag, so the dataset does not lose the cases where customers described the problem well in text.
How large a sample is enough for the audit?
Large enough to estimate the decisive-missing rate per cohort with a stated margin. If one cohort shows a much higher rate, audit more of that cohort before deciding whether to narrow the scope.
Sources
- Zendesk's attachments API marks deleted attachments, and redacting a comment attachment replaces it with an empty redacted.txt file that cannot be undone. Source
- Zendesk's account data export includes attachment links to the attachments API rather than the files, and a stored content_url expires and returns a 403 Forbidden error. Source
- Zendesk can remove attachments through redaction and through attachment deletion schedules that apply to closed, archived tickets while the ticket remains. Source
- Microsoft Graph file attachments carry an isInline flag, a contentId and a size in bytes. Source
- RFC 2392 defines the cid: URL scheme that lets an HTML body reference another body part of the same message, such as an inline image. Source
- The Jira Cloud REST API returns attachment metadata from one endpoint and attachment content from another. Source
- FIPS 180-4 specifies secure hash algorithms whose digests are used to detect whether a message has changed since the digest was generated. Source
See if your company qualifies
A short company assessment. No data uploads are needed.