Loading…
Guide
A fishbone diagram puts an observable problem at the head and possible causes on the main bones and their branches. This tool records more than cause names: it keeps the parent cause, the evidence status, the supporting or refuting material, the next verification step and the owner, so the exported diagram connects to investigation work rather than stopping at a brainstorm.
Updated 2026-09-193 sources6 min read
A fishbone diagram puts an observable problem at the head and possible causes on the main bones and their branches. This tool records more than cause names: it keeps the parent cause, the evidence status, the supporting or refuting material, the next verification step and the owner, so the exported diagram connects to investigation work rather than stopping at a brainstorm.
The default six categories are people, machine, material, method, environment and measurement, and they can be renamed for a service, study or life setting. They are prompts that help you look for what is missing, not answers that have to be filled in. Every line on the diagram first represents an explanation still to be tested, not a proven cause.
Kaoru Ishikawa's quality management work introduced the cause-and-effect diagram systematically as a quality control tool. The method descriptions from ASQ and IHI also use it to identify and organise possible causes of a result. Its value is making explanations explicit and searching the different categories for what was missed; it does not discover a single root cause by virtue of being drawn. ASQ method description, IHI tool description
The main bones carry categories and the branches carry causes or more specific sub-causes. This implementation stores the structure with stable IDs and parent references; the canvas keeps the category cards and shows the parent reference on a sub-cause. A cross-category parent reference raises a check prompt so that the layout does not make it look as though the attribution has been settled. Circular references cannot exist and are rejected on import.
Compared with a mind map, a fishbone organises explanations around one observed result; compared with 5 Whys, it encourages looking across several categories rather than continuing down a single chain. The two can be combined: widen the candidates first, then pick the chain that needs an evidence check. Compared with formal fault tree analysis, this tool has no failure probabilities, logic gate calculation, safety integrity or certification conclusions.
Its limits come from that same open structure: participants tend to write more of the causes they are familiar with, treat an authority's statement as evidence, or stop at “someone was careless”. The categories and connections enforce no experimental design, so processes, information, resources, environment and counter-examples still have to be examined. Completing a fishbone cannot replace an investigation of a high-risk incident.
The “recovery fails after release” example states the problem as: a user opens an old draft, sees a blank page, and cannot tell whether there is no data or the recovery failed. The method category proposes insufficient cross-version recovery testing, with the sub-cause that old-format samples are missing; the machine category considers limited storage; the measurement category considers poor error messages. Each hypothesis connects to a check, for example replaying the recovery process with an archive from the previous release, rather than declaring that a particular developer caused it.
The “repeatedly postponed study tasks” example separates task breakdown, when the materials arrive, work and caregiving arrangements, and the feedback method. Comparing available time with the submitted artefact can change the explanation; counting study hours alone cannot show whether understanding is sufficient.
Problem: an old draft opens as a blank page
Category: method
Main cause r1: insufficient cross-version recovery testing
Sub-cause r2 → r1: old-version data samples are missing
Status: hypothesis to be tested
Verification: check historical release samples and replay the recovery process
Owner: test leadUp to 80 causes, 20 levels of parent-child structure and six renameable categories.
The default categories do not suit every industry and their names can be changed, but this page is not an arbitrary graphics editor.
Very long causes and problems enlarge the canvas and PNG may be scaled down; SVG keeps vector detail and JSON keeps the editable structure.
“Refuted by evidence” is a judgement the user records, and the tool does not verify the material automatically; a refutation should not be deleted just to keep the diagram tidy.
It cannot identify root causes, infer probabilities or verify causation automatically.
Medical, safety and legal matters should be handled by qualified people following the appropriate formal process.
Do not put a solution forward as a cause: “training is needed” does not say what capability gap was observed. Nor should a symptom be restated as an explanation, such as “it was late because it was not finished on time”. Propose a specific mechanism that material could support or refute, then decide the check.
Do not fill in unsupported causes just to balance the six categories. The “people” category can discuss support, handover and capability conditions, and should not become a column for blame. Keep other explanations and counter-examples, and record what result would remove support from the current hypothesis.
It does not. There may be confounders, parallel causes or a faulty measurement, so alternative explanations still have to be sought.
Because the main bones represent categories, and a cross-category reference easily misleads a reader of the diagram. The data is kept, and the user should check whether the category or the stated relationship needs adjusting.
Click undo and the parent and its children come back together; cross-node references also need checking. Use JSON for long-term backup rather than relying on the browser history stack.
Cause sorting, drawing and exporting happen in the browser, with no account sync. Autosave is not a backup and browser cleanup can delete it. A share link contains the whole unencrypted document and can be read by anyone holding it; above 8 KB, switch to a JSON file. Investigation records often involve personal or customer information, so de-identify first and never put raw sensitive material into a link. A failed import does not replace the current work.
Updated 2026-09-19
Organize possible causes and subcauses into editable categories with evidence and follow-up actions; export the diagram and investigation notes
Checking local recovery and shared data…
Local storage is not a backup and can be cleared by the browser. JSON restores all fields. Share links are not encrypted; do not include sensitive data.