Troubleshoot Journal Entries

Troubleshoot Journal Entries

A Journal entry can fail in several ways.

It may:

• never appear;
• appear in the wrong journal;
• use the wrong title;
• contain incomplete information;
• contain incorrect information;
• display poor Markdown formatting;
• appear more than once;
• arrive later than expected; or
• become difficult to find among similar entries.

Begin by identifying the exact symptom.

Then trace the entry back to the conversation, workflow, schedule, tool, or manual action that created it.

Do not repeat a write action until you know the first attempt did not complete.

Identify how the entry was created

First, determine where the entry came from.

It may have been created:

• manually in the Journal;
• by an AI model in Workbench;
• by a workflow in RunFlows;
• by a scheduled workflow;
• through a Gene-provided flow;
• through a Journal tool; or
• by another supported Journal action.

The source determines where you should look for errors.

Entry source	Where to investigate
Manual entry	Journal editor and save action
Workbench	Conversation, enabled tools, and Activity log
RunFlows	Workflow output, intermediate results, and tool activity
Scheduled workflow	Schedule Manager history, run status, and destination
Gene flow	Gene settings, required tools, and workflow output

Start with the visible symptom

Use the table below to choose the first check.

Symptom	First check
No entry appears	Whether the write action ran and which journal was selected
Duplicate entry	Run history, retries, overlapping schedules, and titles
Wrong title	Title instruction, date value, and workflow input
Incomplete content	Source input and the first processing step
Incorrect content	Model instruction, source facts, and intermediate output
Broken formatting	Markdown structure and content passed to the write step
Wrong journal	Destination setting in the tool or workflow
Late entry	Workflow runtime, schedule history, and tool delay
Old entry seems missing	Sorting, title changes, date, and user data location

Correct the first clear issue before changing other parts of the process.

If a manual entry does not save

Check that:

• the entry has a title;
• the editor contains content;
• the save action was completed;
• Feluda remained open during the save;
• no application error appeared;
• the entry list was refreshed; and
• the correct journal is open.

Try a short non-sensitive test entry.

For example:

# Journal Save Test

This is a temporary test entry.

Save it and confirm that it appears in the entry list.

If the test works, review the original title and content for unusual formatting or an interrupted save.

If an AI-written entry does not appear

In Workbench, check:

• whether the Journal tool was enabled;
• whether the selected model supports the required tool use;
• whether the instruction asked for an actual write action;
• whether the model only drafted the content in the conversation;
• whether the Activity log shows a Journal tool call;
• whether the tool call completed;
• whether an error was returned; and
• which journal or destination was used.

Use a direct instruction such as:

Use the enabled Journal tool to create the entry.
Do not only display the entry text in the conversation.

Then review the Activity log again.

If a workflow entry does not appear

Open the run in RunFlows.

Trace the workflow from its starting block to the Journal step.

Confirm that:

• the workflow received input;
• earlier processing steps completed;
• the correct branch ran;
• the Journal step was reached;
• the Journal tool was available;
• the correct destination was configured;
• the write action returned a result; and
• the final Output block did not hide an earlier failure.

Find the first step that returned an unexpected or empty result.

If a scheduled entry does not appear

Open Schedule Manager and check:

• whether the schedule was active;
• whether the planned run started;
• whether Feluda and the computer were available;
• whether required providers and local services were running;
• whether the workflow completed;
• whether the Journal step ran;
• whether the schedule used the expected input; and
• whether the entry was written under another title.

Separate a missed schedule from a failed Journal write.

If no run appears in recent history, the schedule did not start normally.

If the run appears but no entry exists, investigate the workflow and Journal action.

If Journal Monitor shows nothing

Confirm that:

• Journal Monitor is open;
• the correct journal is selected;
• Feluda remains open;
• the expected process is running;
• the workflow writes to the selected journal;
• the write step has been reached; and
• no application error is visible.

Test the monitor with a small manual entry in the same journal.

If the manual entry appears, the monitor is working and the problem is likely in the workflow, tool, or destination.

If the entry was written to the wrong journal

Pause any related schedule before continuing.

Review:

• the selected journal in the workflow;
• the Journal tool settings;
• Gene configuration;
• similarly named journals;
• destination values passed from earlier blocks;
• recent changes to the workflow; and
• whether the correct workflow version ran.

Correct the destination and test with non-sensitive information.

Confirm the result in the intended journal before resuming recurring runs.

If the title is wrong

A title problem may come from:

• a fixed title that was not updated;
• an incorrect date or reporting period;
• a missing value;
• an unclear AI instruction;
• the wrong workflow input;
• the wrong branch;
• a copied test title; or
• an older workflow version.

Keep title creation simple.

For example:

Weekly Project Review — 2026-W23

is more reliable than asking the model to create a long decorative title.

If the date in the title is wrong

Check:

• the schedule timezone;
• the computer date and time;
• daylight-saving changes;
• whether the title uses the run date or reporting period;
• whether the date was supplied manually;
• whether the model interpreted an ambiguous date; and
• whether the source uses another timezone.

Use an exact date format.

Prefer:

2026-06-08

instead of:

today

for recurring automated titles.

If recurring entries use the same title

A fixed title can make separate runs appear to be duplicates or overwrite expectations.

Add a unique value such as:

• date;
• week;
• month;
• project;
• region;
• reporting period; or
• run identifier.

For example:

Daily Operations Briefing — 2026-06-08

is clearer than:

Daily Operations Briefing

Test the title across more than one run.

If the entry content is incomplete

Review the content before the Journal write step.

Check whether:

• the original source was incomplete;
• the workflow received the full input;
• an earlier extraction step lost information;
• a model ignored part of the instruction;
• the input was too long;
• a tool returned only part of the data;
• the wrong branch ran; or
• a length limit removed important details.

Find the earliest point where the information became incomplete.

Do not begin by changing the Journal tool if it saved exactly what it received.

If the entry contains incorrect facts

Compare the entry with the original source.

Look for:

• changed names;
• incorrect dates;
• wrong amounts;
• invented deadlines;
• missing decisions;
• unsupported claims;
• suggestions presented as facts; and
• information placed under the wrong heading.

Improve the AI or extraction step that prepared the content.

Add rules such as:

Use only information from the source.
If a required detail is missing, write "Not provided."
Do not guess.

If suggestions are mixed with facts

Update the content structure.

Use separate sections:

## Confirmed Findings

[Facts supported by the source]

## Suggested Next Steps

[Recommendations for review]

Do not allow AI suggestions to appear under headings such as Confirmed Decisions unless a person approved them.

If missing information is hidden

Empty fields can make the reader think the workflow forgot them.

Use a visible value such as:

Not provided

Or:

Pending confirmation

Test examples where important information is intentionally absent.

The workflow should not invent the missing detail.

If Markdown formatting is broken

Check whether:

• headings begin with the correct number of hash signs;
• bullet lists use consistent markers;
• numbered lists are separated correctly;
• table rows contain the same number of columns;
• links use valid Markdown;
• code blocks are opened and closed;
• blank lines separate sections; and
• special characters were altered before the write step.

Review the exact content passed to the Journal action.

The write step may be working correctly while the prepared Markdown is malformed.

If a table displays incorrectly

Confirm that the table includes:

• a header row;
• a separator row;
• the same number of cells in each row; and
• simple cell content.

For example:

| Owner | Action | Deadline |
|---|---|---|
| Sam | Prepare files | Friday |
| Mia | Confirm interviews | Not provided |

Avoid very long paragraphs inside table cells.

Use headings and bullet lists when the content is too detailed for a table.

If links do not work

Check:

• the full destination address;
• the Markdown brackets and parentheses;
• whether spaces or line breaks damaged the link;
• whether the source still exists;
• whether access is required; and
• whether the link points to the intended source.

Use descriptive link text.

Do not place private credentials or sensitive access values in a Journal link.

If code blocks break the rest of the entry

Confirm that every code block has both an opening and closing fence.

For example:

```text
Input → LLM → Output
```

An unclosed code block can cause later Markdown to display incorrectly.

Keep code examples separate from normal paragraphs.

If duplicate entries appear

Check for:

• the same workflow running twice;
• repeated clicks in RunFlows;
• overlapping schedules;
• a manual and scheduled run;
• a retry after a timeout;
• two workflows writing to the same journal;
• duplicate schedules; and
• delayed tool confirmation.

Compare:

• entry titles;
• timestamps;
• workflow names;
• run history;
• schedule history; and
• content.

Pause related schedules before continuing when duplicates repeat.

Before deleting or changing a duplicate

Confirm whether the entries are truly identical.

They may represent:

• separate runs;
• different reporting periods;
• a corrected version;
• an intermediate and final entry;
• a manually reviewed copy; or
• two different sources.

Mark entries clearly when both need to remain.

For example:

Weekly Project Review — 2026-W23 — Corrected

If a timeout may have created the entry

Do not retry immediately.

First:

1. open the Journal;
2. search or browse for the expected title;
3. compare timestamps;
4. review tool activity;
5. check RunFlows or Schedule Manager history; and
6. retry only if the first write did not complete.

A delayed confirmation can make a successful write look like a failure.

If an entry appears late

Review:

• workflow runtime;
• provider response time;
• local model performance;
• tool delays;
• external service availability;
• schedule trigger time;
• maximum runtime; and
• whether another run used the same resources.

Compare the Journal timestamp with the run start and completion times.

A late entry may have been written by a slow but successful workflow.

If an old entry seems to have disappeared

Check:

• date sorting;
• whether newest or oldest entries are shown first;
• title changes;
• another journal;
• the entry date;
• whether it was marked or moved;
• whether Feluda is using the expected user data location; and
• whether the application interface changed after an update.

Feluda stores Journal data separately from the application files, so normal updates should not remove personal entries.

Avoid assuming the entry was deleted until you have checked sorting, journals, and data location.

If entries appear missing after an update

Confirm that:

• the update completed normally;
• Feluda is running under the same operating-system user;
• the expected data folder is being used;
• the Journal list has loaded fully;
• sorting has not changed;
• the entry exists in another journal; and
• the interface layout has not moved the relevant control.

Personal flows, journals, Genes, and settings are stored separately from the application files and are intended to remain available after updates.

Do not create replacement entries until you know the originals are missing.

If only some entries are visible

Check whether:

• a journal filter is active;
• a different journal is selected;
• the list is sorted unexpectedly;
• the application has not finished loading;
• titles changed;
• recent entries use another reporting period; or
• a workflow wrote to a different destination.

Compare Journal Monitor, RunFlows, and Schedule Manager information.

If the entry cannot be edited

Confirm whether:

• it is a manual entry;
• the edit control is available;
• the entry is still loading;
• another process is writing to it;
• the application needs to be refreshed;
• the workflow should create a corrected version instead; and
• the entry is intended to remain as a historical record.

For important AI-generated records, adding a correction section or new corrected entry may be clearer than silently replacing the original.

Preserve historical corrections

When an entry acts as a record, keep the original context visible.

Add:

## Update — 2026-06-10

The launch date was later confirmed as 20 June 2026.

Or create:

Weekly Project Review — 2026-W23 — Corrected

Mark the earlier entry as Superseded when appropriate.

If a corrected entry is also duplicated

Use distinct titles and explain the relationship.

For example:

**Status:** Superseded

Replaced by:
Weekly Project Review — 2026-W23 — Corrected

This prevents the corrected version from looking like an accidental duplicate.

If a Journal tool is unavailable

Check whether:

• the required capability is active;
• the related Gene is installed and synced;
• the selected model supports tool use;
• the tool is enabled for the conversation or workflow;
• required settings are complete;
• Feluda has been refreshed; and
• the tool appears in Workbench.

Test the smallest supported Journal action first.

If a permission error appears

Review whether the requested action is allowed.

Check:

• the selected destination;
• Journal access;
• file or user-data permissions;
• Gene permissions;
• connected service permissions; and
• whether the operation attempts to access an unsupported location.

Do not expand permissions automatically.

Confirm that the destination and action are correct first.

If the workflow says success but no entry exists

Treat the workflow message as one piece of evidence.

Also check:

• the tool result;
• the Journal destination;
• the actual Journal list;
• the entry title;
• the run timestamp;
• warnings;
• errors; and
• whether the model only drafted the content.

A model-generated success sentence does not prove that a write action completed.

If the entry exists but the workflow reports failure

The write may have completed before a later step failed.

Confirm:

• the entry title;
• the content;
• the timestamp;
• the later failed step;
• whether the workflow should be retried; and
• whether a retry would create a duplicate.

Fix the later step without repeating the Journal action when possible.

If recurring entries become inconsistent

Compare several recent entries.

Review:

• titles;
• section names;
• date formats;
• tables;
• source sections;
• missing-value rules;
• model selection;
• workflow version; and
• AI instructions.

A provider or model change can produce different formatting from the same prompt.

Test the recurring template in Workbench and RunFlows.

If reports become too long

Update the content-preparation instruction.

Consider:

• limiting the summary;
• including only the current reporting period;
• moving detailed sources to links or references;
• separating completed and open actions;
• removing repeated background information; and
• using concise tables.

Do not shorten the report by removing required facts.

If reports become too short

Check whether:

• the source is incomplete;
• the input was truncated;
• a length limit is too strict;
• an extraction step missed fields;
• the wrong source was used;
• the selected model ignored requirements; or
• a tool returned less information.

Compare the report with the source and the previous successful entry.

If entries contain sensitive information unexpectedly

Pause related workflows or schedules.

Review:

• the source input;
• AI instructions;
• extraction fields;
• enabled tools;
• the selected journal;
• whether personal details were necessary; and
• who can access the device.

Remove unnecessary sensitive information according to your normal data practices.

Never save API keys, passwords, access tokens, or other credentials in the Journal.

If the Journal becomes difficult to browse

Improve organisation by:

• using consistent title prefixes;
• including reporting periods;
• separating unrelated projects;
• using one purpose per entry;
• marking corrected and superseded entries;
• keeping recurring structures consistent;
• reviewing old test entries; and
• using separate journals when that improves clarity.

Do not rely on titles such as Report, Test, or Final.

Use a small test entry

When troubleshooting, use a simple non-sensitive entry.

For example:

Title:
Journal Test — 2026-06-08

Content:
# Journal Test

This is a temporary Journal troubleshooting entry.

Use the test to confirm:

• the destination;
• the write action;
• the title;
• Markdown rendering;
• Journal Monitor; and
• entry visibility.

Clearly mark or remove test content according to your normal Journal practices.

Change one thing at a time

Avoid changing the title, prompt, model, tool, workflow, schedule, and destination at the same time.

Change one item.

Repeat the same test.

Confirm whether the result improved.

This makes the cause easier to identify.

Re-test the complete path

After correcting the issue, test:

• the source input;
• the processing steps;
• the title;
• the Journal action;
• the destination;
• the final confirmation;
• Journal Monitor;
• duplicate handling; and
• scheduled execution when relevant.

A fix that works manually may still fail in the scheduled environment.

Know when to pause automation

Pause the workflow schedule when:

• entries are repeatedly missing;
• duplicates continue;
• the wrong journal is used;
• sensitive information is written incorrectly;
• formatting makes reports unusable;
• the source is unreliable;
• the provider or tool is unavailable; or
• no one can review the results.

Resume only after successful manual testing.

A practical troubleshooting routine

Use this process:

1. Identify how the entry was created.
2. Define the exact symptom.
3. Confirm the intended journal and title.
4. Review Workbench, RunFlows, or Schedule Manager activity.
5. Find the first failing or incorrect step.
6. Check the source information.
7. Review the prepared Markdown before the write.
8. Confirm the Journal tool and permissions.
9. Check the actual Journal before retrying.
10. Correct one issue.
11. Test with a small non-sensitive entry.
12. Review the saved result and Journal Monitor.
13. Check for duplicates.
14. Re-test scheduled execution when relevant.
15. Resume automation only when the result is dependable.

This approach helps you correct the cause without creating additional entries or losing the original context.