Systems and records
Event log status normalization: reconciling five meanings of closed
By SourceX Editorial · Updated
9 sections
Short answer
To reconcile different meanings of closed across business systems, map each raw status into a state dictionary that records three things separately: whether the record is terminal, what the outcome was, and who or what closed it. Take end times from the transition that ended the work, follow merges and duplicates, and reconcile counts per raw value back to each system's totals.
Record timeline
Event log status
Key takeaways
- Closed, Done, Resolved and Won record different facts in different tools, so a single canonical Closed corrupts durations and outcomes.
- Separate lifecycle, outcome and closure actor into three columns of the dictionary.
- Take the end time from the transition that ended the work, such as Solved, not from a later lock.
- Merged and duplicate records point to a surviving record and carry no outcome of their own.
- Audit coverage, source totals, reopenings and links, and report the share of unknown outcomes.
How do you reconcile different meanings of closed across business systems?
Do not map every closed value to one canonical Closed. Build a state dictionary that records, for each raw status in each system, three separate things: whether the work reached a terminal state, what the outcome was, and what caused the closure. Then audit the result: every raw value mapped, counts reconciled back to each system's own totals, and reopenings kept as events rather than overwritten.
The reason is that a status called Closed, Done, Resolved or Won answers a different question in each tool. In one system it means an agent finished; in another it means a timer locked the record days later; in a third it means the work was abandoned or merged into something else. Treating them as the same event produces case durations, completion rates and outcome labels that are wrong in ways nobody can see from the merged log.
This page gives a state dictionary format, a worked record that passes, a normalization that fails, and an audit checklist. Field names and the cross-system example are illustrative; the system behaviors cited are from each vendor's own documentation.
Five systems, five things a closed status records
Each system's documentation defines closure in its own terms, and the differences are not cosmetic. The table summarizes what the closed-looking value records in five common tools.
Zendesk is the clearest case of a status that is a lock rather than an outcome. Its help center says agents cannot set Closed by hand; solved tickets move to Closed through automation, by default 4 days after being solved and in any case no later than 28 days, and a reply to a closed ticket creates a follow-up ticket instead of reopening it. The moment work ended is the Solved transition, not the Closed one.
Jira separates the status from the reason. Atlassian's documentation says the Resolution field records why a work item moved to Done, with values such as Fixed, Won't Fix or Cannot Reproduce, and that Jira treats an item with no resolution as Unresolved even if its status is Done.
| System | Closed-looking values | What the value records | Trap for a merged log |
|---|---|---|---|
| Zendesk Support | Solved, Closed | Solved: agent finished. Closed: record locked by automation later | Using Closed as the end time adds days to every case |
| Jira | Done category plus Resolution | Status says finished; Resolution says why (Fixed, Won't Fix, Duplicate) | Done with no Resolution counts as Unresolved in Jira's own reports |
| GitHub issues | Closed with reason completed or not planned | Whether the work was done or deliberately dropped; can close on a linked merge | Not planned read as fixed |
| Dynamics 365 case | State Resolved or Canceled, status reason | Resolved covers Problem Solved and Information Provided; Canceled covers Merged | A merged case counted as an abandoned one, or as a separate resolution |
| Dynamics 365 opportunity | State Won or Lost, status reason | Closed means the deal ended; Lost includes Canceled and Out-Sold | Closed opportunities summed as wins |
Split closed into three questions before mapping anything
Every closed-looking value is answering up to three questions at once, so the dictionary keeps them in three columns. Lifecycle asks whether the record can still change. Outcome asks what happened to the work. Closure actor asks who or what made the change.
Microsoft's Dataverse reference for the case table shows why the split is needed. Its state field has three values, Active, Resolved and Canceled, and resolved or canceled cases are read-only unless reactivated. Underneath, the status reason field holds Problem Solved and Information Provided under Resolved, and Canceled and Merged under Canceled. State answers lifecycle; status reason answers outcome; neither tells you whether a person or a rule made the change.
Process mining standards already make part of this distinction. The XES standard's lifecycle extension records transitions such as start, complete, ate_abort for an aborted activity and pi_abort for an aborted process instance, in the lifecycle:transition attribute. A completion and an abort are different transitions, even if both leave the case finished.
- Lifecycle: open, terminal but reversible (can be reopened or reactivated), terminal and locked.
- Outcome: completed, completed without a change (information given, cannot reproduce), not done by decision, duplicate or merged, won, lost, unknown.
- Closure actor: named person, workflow rule or automation, linked event (a merge, a pull request, a parent record), unknown.
The state dictionary: one row per raw value
The dictionary has one row for each combination of system, object type, field and raw value, including values that no longer appear in the current configuration. That last point matters because workflows change: a status that existed for three years and was retired last quarter still sits in the history.
- source_system and object_type, for example Jira issue or Dynamics 365 opportunity.
- raw_field, naming exactly which field holds the value: status, statecode, statuscode, resolution or state_reason.
- raw_value and the system's own grouping where one exists, such as Jira's status category or the Dataverse state a status reason belongs to.
- canonical_lifecycle, canonical_outcome and closure_actor, using the three axes above.
- end_time_rule: which transition marks the end of the work for duration measures, for example the Solved transition rather than the later Closed one.
- reversible: whether the source allows reopening or reactivation, and how a reopening appears in the history.
- valid_from and valid_to: the dates the mapping applies, tied to workflow configuration changes.
- evidence: the documentation or configuration export the mapping relies on, and the person who confirmed it.
- confidence: confirmed by the workflow owner, inferred from data, or unresolved.
| System and raw value | Lifecycle | Outcome | Closure actor | End-time rule |
|---|---|---|---|---|
| Zendesk Solved | Terminal but reversible | Completed (agent judgment) | Person | This transition |
| Zendesk Closed | Terminal and locked | Inherits outcome from Solved | Automation | Not an end time |
| Jira Done, Resolution Duplicate | Terminal but reversible | Duplicate | Person or workflow rule | This transition |
| Jira Done, no Resolution | Terminal by status only | Unknown | Person | Flag for review |
| GitHub closed, not planned | Terminal but reversible | Not done by decision | Person | This transition |
| GitHub closed by merged pull request | Terminal but reversible | Completed | Linked event | Merge time |
| Dynamics 365 case Canceled, Merged | Terminal, reactivation possible | Duplicate or merged | Person | Not an end time for the surviving case |
| Dynamics 365 opportunity Lost, Out-Sold | Terminal, reactivation possible | Lost | Person | This transition |
Worked record: one customer problem across four systems (illustrative)
A valid normalization keeps every raw event, maps each one through the dictionary, and builds one episode whose end time and outcome come from the right transitions. The example below is invented to show the structure; the identifiers and durations do not describe any real company.
A customer reports an export failure. Support opens Zendesk ticket Z-881 and a Dynamics 365 case C-55 is created for the account team, then merged into an existing case C-40 for the same customer. Engineering logs Jira bug J-310, which is resolved as Duplicate of J-212. J-212 is fixed by a pull request that closes GitHub issue G-97 when merged into the default branch. The agent marks Z-881 Solved after the fix ships; 4 days later Zendesk closes it.
Normalized, the episode starts at the Zendesk ticket creation and ends at the Solved transition, not the Closed one. The outcome is Completed, supported by the merged pull request. C-55 is recorded as merged into C-40 rather than as a second resolution, and J-310 as a duplicate pointing to J-212, so the episode counts one fix, not three closures. GitHub documents that merging a linked pull request into the default branch closes the issue automatically, so the closure actor on G-97 is the linked event, not a person.
| Raw event (illustrative) | Day | Mapped lifecycle and outcome | Role in the episode |
|---|---|---|---|
| Z-881 created | 0 | Open | Episode start |
| C-55 Canceled, status reason Merged into C-40 | 1 | Terminal, duplicate or merged | Link to surviving case, no outcome |
| J-310 Done, Resolution Duplicate of J-212 | 2 | Terminal, duplicate | Link to the issue that carries the fix |
| G-97 closed by merged pull request | 6 | Terminal, completed, linked event | Evidence of the fix |
| Z-881 Solved | 7 | Terminal but reversible, completed | Episode end time |
| Z-881 Closed by automation | 11 | Locked, outcome inherited | Ignored for duration |
A failing normalization and what it does to the numbers
The common failure is a mapping table with one row per raw value that says Closed equals done. Applied to the same illustrative episode, it produces four closures instead of one fix, an episode that ends on day 11 instead of day 7, and a merged case that looks like a second resolved problem.
Across a full export the distortions compound. Durations stretch by the auto-close delay on every helpdesk ticket. Completion rates rise because not-planned issues, canceled cases and duplicates count as completed work. In sales data, Dynamics 365 lists Canceled and Out-Sold as status reasons under Lost, so a flat Closed mapping of opportunities mixes wins and losses into one bucket.
Jira shows the reverse failure. Because Jira treats a Done item with no Resolution as Unresolved, a log that trusts the resolution field undercounts finished work wherever a workflow never set it. Neither error appears as a missing value; both look like clean data.
- Symptom: median duration jumps by a fixed number of days after a helpdesk migration. Likely cause: end time taken from a lock transition.
- Symptom: completion rate far above what the team reports. Likely cause: not-planned, canceled and duplicate outcomes mapped to completed.
- Symptom: one problem appears as several finished cases. Likely cause: merges and duplicate links not followed.
- Symptom: a block of Done items with unknown outcome. Likely cause: a workflow that never set the resolution field.
When clocks, identifiers or missing transitions rewrite the episode
These three faults can change which transition counts as the end and which records belong to one episode, so check them before trusting any normalized state. Each one has a test that can be run on the raw export.
Clock differences between systems can reorder events that happen close together, such as a merge and the resolution of the surviving case. Record each system's timestamp time zone and precision, and treat events closer together than the known skew as unordered unless a link or sequence number settles the order.
Identifier mismatches split or merge episodes wrongly. A ticket that references a bug by a key typed in free text, or a case merged into another without a stored link, needs either a confirmed join or a label saying the link was inferred.
Unrecorded transitions are the hardest. If a system stores only the current status, a record that was reopened and closed again looks like it closed once. The OCEL 2.0 specification allows object attribute values to be recorded with timestamps, independent of events, which is the right shape for status histories; where the source kept only the latest value, mark the episode as having an incomplete history rather than inventing the missing steps.
The normalization audit to run before anyone uses the log
The audit proves that normalization lost nothing and invented nothing. Run it per system and per period, and keep its output with the dataset so a reviewer can repeat it.
| Check | How to run it | Pass rule | If it fails |
|---|---|---|---|
| Coverage | List distinct raw values per system and field in the export | Every value has a dictionary row valid for its date | Add the row with the workflow owner, or label unmapped |
| Source totals | Count records per raw value in the source system and in the normalized log | Counts match for every value and period | Find dropped or duplicated rows before going further |
| Outcome balance | Compare canonical outcome counts with the team's own reports | Differences explained by documented mapping choices | Review mappings for not planned, canceled and duplicate values |
| End-time rule | Compare durations using the rule against durations using the last transition | Gap matches the known lock or auto-close delay | Recheck which transition is marked as the end |
| Reopen handling | Count records with more than one terminal transition | Each reopening kept as its own event | Rebuild from history tables, or flag incomplete history |
| Link integrity | Follow merge and duplicate links to their surviving records | Every link resolves to a record in the export | Label broken links; do not count the record as its own outcome |
| Unknown share | Count episodes with outcome unknown or confidence unresolved | Share reported with the dataset | Report it; never fill unknowns with the most common value |
Status histories in a data licensing review
Status histories that keep raw values, transitions and links are worth more than exports that keep only a final status, because the outcome of each case can be checked rather than assumed. A state dictionary and audit output also answer the first questions a technical reviewer asks about such records.
Ticket, case and issue histories offered to AI developers go through SourceX's rights review, contracting, preparation and delivery once a company picks which records to include. Its suppliers are U.S., Canadian and EU companies that had at least 50 full-time employees at their peak, not counting contractors; companies in the UK, Australia, New Zealand, India, Singapore, South Africa, Latin America, the UAE and Saudi Arabia can be considered when at least 90% of the offered data is in English, as confirmed on the dataset itself. The first step is a short assessment, and no status export has to be uploaded for it.
Frequently asked questions
What is event log status normalization?
It is the step that maps each system's raw status values onto a shared set of states so events from different tools can be combined into one log. Done well, it keeps the raw value, records lifecycle, outcome and closure actor separately, and is checked against each system's own totals.
What is the difference between Solved and Closed in Zendesk?
Solved is set by an agent when the work is finished and can still change if the requester replies. Closed is applied by automation, by default 4 days after solving and no later than 28 days, and a reply then creates a follow-up ticket. For duration measures, the Solved transition marks when the work ended.
Should a duplicate or merged record count as completed?
No. It should point to the surviving record and take no outcome of its own, or one problem will be counted as several finished cases. Jira's Duplicate resolution and the Merged status reason on Dynamics 365 cases both mark this situation.
What should the state dictionary include?
One row per system, object type, field and raw value, with the canonical lifecycle, outcome and closure actor, the rule for which transition ends the work, whether the record can reopen, the dates the mapping applies and who confirmed it. Include retired values that still appear in history.
How do you know the normalized log is correct?
Reconcile record counts per raw value back to each source system for every period, confirm every raw value is mapped, and compare outcome counts with the team's own reports. Report the share of episodes whose outcome stayed unknown instead of filling it in.
Sources
- Zendesk's help center says agents cannot close tickets manually; solved tickets close by automation, by default after 4 days and no later than 28 days, and replies to closed tickets create follow-ups. Source
- Atlassian's documentation says Jira's Resolution field records why a work item moved to Done and that items without a resolution show as Unresolved. Source
- GitHub's documentation lists completed and not planned as the reasons a closed issue can carry. Source
- GitHub's documentation says merging a linked pull request into the default branch closes the linked issue automatically. Source
- Microsoft's Dataverse reference defines case states Active, Resolved and Canceled, with Merged as a status reason under Canceled. Source
- Microsoft's Dataverse reference defines opportunity states Open, Won and Lost, with Canceled and Out-Sold as status reasons under Lost. Source
- The XES standard's lifecycle extension defines transitions including start, complete, ate_abort and pi_abort. Source
- The OCEL 2.0 specification records time-stamped object attribute values independent of events and supports qualified object-to-object relationships. Source
See if your company qualifies
A short company assessment. No data uploads are needed.