Root Cause Hypothesis Log: Prediction and Falsification Template
Summary
A root cause hypothesis log needs 5 sections: observed failure, hypotheses, predictions and refutation conditions frozen before evidence, evidence for and against, and verdicts. Quality engineers, technical service staff and supplier quality teams use it on a chemical or manufacturing complaint whose cause is not yet confirmed and has more than one plausible explanation.
A hypothesis that cannot be refuted predicts nothing; Platt's strong inference asks for experiments whose outcomes each exclude one or more hypotheses (Platt, 1964).
Evidence consistent with every hypothesis may have no diagnostic value, so Analysis of Competing Hypotheses scores each item against all hypotheses at once (Heuer, 1999, chapter 8).
Confirmation bias is the seeking or interpreting of evidence in ways partial to a hypothesis in hand (Nickerson, 1998).
An unproven hypothesis has no evidence that it is correct; a disproved one has positive evidence that it is wrong (Heuer, 1999).
What is a root cause hypothesis log for, and who uses it?
A root cause hypothesis log is a working record that lists every candidate cause of one failure, fixes in advance what each cause predicts and what would rule it out, and then records how the evidence judged each one. Quality engineers at a chemical supplier, technical service engineers at a distributor and supplier quality engineers at a manufacturer use it while a complaint is open and its cause is not confirmed.
The log is the analysis sheet behind a corrective action report, not the report itself. An 8D or complaint report states the root cause and the actions; the hypothesis log shows why the other candidates were set aside and under which conditions. That record matters when the same symptom returns six months later and someone asks whether contamination was really excluded or only never checked.
Root cause analysis has a published process standard. IEC 62740:2015 (edition 1.0) "describes the basic principles of root cause analysis (RCA) and specifies the steps that a process for RCA should include", and it notes that RCA can be used for non-conformances in quality management systems as well as for failure analysis (IEC, 2015). The hypothesis log is one way to keep the hypothesis and evidence steps of such a process visible.
Use the log when a complaint has at least two credible causes in different areas, such as raw material, process and measurement. The editable Word template and Excel log are at the end of this page.
Which sections does the root cause hypothesis log have?
The root cause hypothesis log has 5 sections, filled in order, and the first three are complete before anyone retrieves the evidence. Each section answers one question.
Section 1, observed failure, records what was seen, where, when and how it was measured, and the comparison scope: the affected units, comparable unaffected units and the period. Each statement is marked observed (someone on the team saw or measured it) or reported (someone told the team). "The coating went bad" stays reported until restated as an observable, such as craters per panel.
Section 2, hypotheses, lists one row per candidate cause, each with a cause domain (raw material and lot, formulation, process conditions, equipment, contamination and carry-over, handling and storage, measurement error, the customer's own process change) and a mechanism chain from the cause to the observation.
Section 3, predictions and refutation conditions, states for each hypothesis what the records should show if it is true and what finding would rule it out, with a freeze date.
Section 4, evidence for and against, records each item of evidence and how it bears on every hypothesis, not only the favored one.
Section 5, verdicts, gives each hypothesis a verdict of confirmed, ruled out or open, an exclusion scope for each one ruled out, and the confirmation test with its owner for each one still open.
The Word template splits these five sections into ten numbered parts, followed by a sign-off checklist and the names of who prepared and reviewed the log. The table previews every part; the example values come from the illustrative cratering case later on this page.
Part | What goes in it | Units or format | Example value (illustrative) |
|---|---|---|---|
1 Observed failure and comparison scope | Case reference, product and lot numbers, site or line, log owner, date opened; each statement with how it was measured, observed or reported, and source; onset date and type, affected units, comparable unaffected units, comparison period and gaps | Value with unit; YYYY-MM-DD; observed or reported | Craters on lines A and B from 2026-10-01, line C unaffected |
2 Implicated components | Product and SDS revision date, component, CAS number, content and undisclosed share for each implicated product | CAS number; % w/w; % | Waterborne coating, SDS revised 2026-03-01, undisclosed share 4 % |
3 Change inventory and cause domains | Each change axis (lot, supplier and grade, set points, cleaning and maintenance, season and weather, source water, measurement method, personnel) marked changed, unchanged or unconfirmed with date and record; each cause domain with its hypothesis IDs or the reason none applies | C, U or ?; YYYY-MM-DD | Cleaning and maintenance: C, 2026-09-30, cleaner batch record |
4 Hypothesis ledger | One block per hypothesis: cause domain, failure family, mechanism chain, prediction, refutation condition, freeze date, added after freeze | Pick list; observable with unit; YYYY-MM-DD; yes or no | H2, compressor B oil carry-over, frozen 2026-10-05 |
5 Examination order and evidence plan | Position, hypothesis ID, records to fetch, holder, date requested; unconfirmed facts at framing and what would resolve each | Rank; record name; YYYY-MM-DD | 1, H2, compressor B maintenance log, customer maintenance team, 2026-10-05 |
6 Evidence per hypothesis | Each evidence item once with value, unit, source and date, observed or reported, and its judgment against every hypothesis | S, N or R per hypothesis | E2: H1 R, H2 S, H3 N, H4 N |
7 Verdict and exclusion scope | Verdict per hypothesis, exclusion scope for each one ruled out, key evidence IDs | Confirmed, ruled out or open; lots, dates, methods, units | H1 ruled out for lot 2 on lines A to C |
8 Mechanism and sufficiency of surviving causes | Surviving hypothesis IDs and whether the cause is compound, the mechanism chain with the principle at each link, the confirmed concentration or exposure and the value the mechanism requires, with its source | Value with unit; source | Left blank while H2 is open |
9 Overturning conditions and discrimination tests | Each condition that would change a verdict and its effect; each test with method, prediction per hypothesis, owner and days required | Value with unit; days | Nitrogen spray trial: H2 predicts no craters, customer process engineer, 7 days |
10 Limitations | Open hypotheses with how to confirm each and who holds the record; missing figures and their effect on confidence and exclusion scope | Text | H3 open: old cleaner batch not retained |
Which columns does a root cause hypothesis log need?
A root cause hypothesis log needs 14 columns per hypothesis row, and the four that carry the method are the prediction, the refutation condition, the freeze date and the exclusion scope. The table below previews every column of the hypothesis sheet and decides which hypotheses stay open and what would settle each one; the example values describe one illustrative cratering complaint on a waterborne coating and are not data from a real case.
Column | What goes in it | Format | Example value (illustrative) |
|---|---|---|---|
Hypothesis ID | Short unique label | Text, H1, H2 | H2 |
Cause domain | Where the cause would lie | Pick list of 8 domains | Contamination and carry-over |
Failure family | Kind of failure it would produce | Pick list | Contamination and carry-over |
Mechanism chain | Cause to observation in links | Text, one link per clause | Silicone in compressed air reaches wet film, lowers local surface tension, film pulls back into craters |
Prediction | What the records show if true | Observable with unit | Craters on every line fed by compressor B, none on line fed by compressor A |
Refutation condition | Finding that rules it out | Observable with unit | Craters at the same rate on panels sprayed with bottled nitrogen |
Freeze date | Date prediction and refutation were fixed | YYYY-MM-DD | 2026-10-05 |
Added after freeze | Whether evidence already seen suggested it | Y or N | N |
Record to examine first | First record to pull | Record name and holder | Compressor B maintenance log, customer maintenance team |
Supporting evidence | Items consistent with it | Evidence ID list | E3, E5 |
Counter-evidence | Items inconsistent with it | Evidence ID list | E4 |
Verdict | Confirmed, ruled out or open | Pick list | Open |
Exclusion scope | Conditions under which a ruling-out holds | Text with units and dates | Not applicable while open |
Confirmation test and owner | Test that would settle it, who runs it | Text, name or role, due date | Nitrogen spray trial, customer process engineer, 2026-10-12 |
The prediction and refutation condition columns carry the weight: a row with an empty refutation condition does not enter the log, because nothing could ever rule that hypothesis out. The supporting evidence and counter-evidence columns hold IDs that point to a separate evidence sheet, where each item has its own date, source and holder, so one lab result can be cited against several hypotheses without copying it. The Excel log's Evidence sheet has 8 columns.
Column | What goes in it | Format | Example value (illustrative) |
|---|---|---|---|
Evidence ID | Short unique label | Text, E1, E2 | E2 |
Evidence item | What the record shows, with value and unit | Text with value and unit | Lot 2 sprayed on line C: no craters on 20 panels |
Date of record | When the record was made | YYYY-MM-DD | 2026-10-07 |
Source and holder | The record and who holds it | Text | Customer line C log |
Observed or reported | Whether the team saw it or was told | O or R | O |
Retrieved on | When the team obtained the record | YYYY-MM-DD | 2026-10-07 |
Diagnostic | Whether the item is judged differently across the hypotheses | Y or N | Y |
Judgment per hypothesis | Supports, neutral or refutes for each hypothesis | H1:S, H2:N, H3:R | H1:R, H2:S, H3:N, H4:N |
The Excel log also adds pick lists for cause domain, failure family and verdict so that entries stay comparable across complaints.
How do you write a prediction and a refutation condition for a root cause hypothesis?
A prediction for a root cause hypothesis states an observable that the records would show if the cause were true, and a refutation condition states an observable that would show it is false; both name the record, the unit and the comparison. Write them from the mechanism chain, link by link. If raw material lot 2 caused the defect, the prediction is that the defect appears with lot 2 and not with lot 1 under the same process settings, and the refutation condition is that lot 1 retained material produces the defect at the same rate.
A good refutation condition is a finding someone can actually look for. "The material is fine" is not one. "Retained sample of the complained lot meets the release specification on retest and the customer's reference lot shows the same defect" is one, because each part names a record that exists or can be produced.
John Platt called this strong inference: devise alternative hypotheses, then devise a crucial experiment "with alternative possible outcomes, each of which will, as nearly as possible, exclude one or more of the hypotheses" (Platt, 1964). He warned that an explanation that excludes nothing "predicts everything, and therefore does not predict anything" (Platt, 1964). That is the test for each row: if no conceivable record could contradict the hypothesis, rewrite it more narrowly or drop it.
Keep predictions comparative where you can: affected against unaffected units, lots or lines separates causes better than a single reading.
Why freeze root cause hypotheses before looking at the evidence?
Root cause hypotheses are frozen before the evidence is retrieved so that the evidence tests them instead of shaping them, which keeps the first suspect from absorbing every later finding. Once a team has seen the batch record or a lab result, it is easy to write a prediction that the result already satisfies.
Confirmation bias, as Raymond Nickerson defined it in his review, "connotes the seeking or interpreting of evidence in ways that are partial to existing beliefs, expectations, or a hypothesis in hand" (Nickerson, 1998). In complaint work this shows up as reading an ambiguous lab result as support for the cause the team suspected on day one, while the same result fits two other causes equally well.
Research methods treat the same problem with preregistration. The Center for Open Science explains that "the same data cannot be used to generate and test a hypothesis, which can happen unintentionally and reduce the credibility of your results" (Center for Open Science, accessed 2026). A freeze date in the hypothesis log plays the same role on a smaller scale: it records which predictions existed before which records were opened.
Freezing does not forbid new hypotheses. When the evidence suggests a cause nobody listed, add it as a new row with its own freeze date and mark it Y in Added after freeze. The log then shows honestly which hypotheses were tested blind and which were suggested by the evidence they are judged against.
How do you fill in the root cause hypothesis log before and after the evidence?
Fill in the root cause hypothesis log in two passes: before the evidence, complete the observed failure, the hypotheses, the predictions and the refutation conditions; after the evidence, complete the evidence matrix and the verdicts. The steps run in this order.
Write the observed failure from measurements, and mark each item observed or reported.
Fix the comparison scope: affected units, comparable unaffected units and the period.
List what changed around onset: lot, supplier or grade, set points, cleaning, season, water, measurement method and staffing. Mark each changed, unchanged or unconfirmed.
Write at least one hypothesis per cause domain that fits the change list, or a reason why the domain holds none.
Write the mechanism chain, prediction and refutation condition for each, then enter the freeze date.
Name the first record to examine for each hypothesis and its holder.
Retrieve the records and enter each as an evidence item with its date and source.
Judge every evidence item against every hypothesis as supports, neutral or refutes.
Give each hypothesis a verdict and, where ruled out, its exclusion scope.
Step 8 follows Analysis of Competing Hypotheses, which uses a matrix with hypotheses on one axis and evidence on the other so that each item is weighed against every hypothesis (Heuer, 1999, chapter 8). Heuer notes that if an item of evidence "seems consistent with all the hypotheses, it may have no diagnostic value" (Heuer, 1999). Mark such items as not diagnostic.
How do you record a verdict and exclusion scope in a root cause hypothesis log?
A verdict in a root cause hypothesis log is one of confirmed, ruled out or open, and every ruled-out hypothesis carries an exclusion scope that states the conditions under which the ruling-out holds. Confirmed means the prediction was met and the refutation condition was looked for and not found. Ruled out means the refutation condition was found. Open means neither has happened yet.
Keep a hypothesis open when there is simply no evidence for it. Heuer separates the two cases: for an unproven hypothesis "there is no evidence that it is correct", while for a disproved one "there is positive evidence that it is wrong", and he advises that unproven hypotheses "should be kept alive until they can be disproved" (Heuer, 1999). A cause that nobody checked is open, not ruled out.
The exclusion scope stops a ruling-out from spreading beyond what the evidence covered. If retained samples of lots 1 to 3 met specification on retest, the scope reads "raw material ruled out for lots 1 to 3 at the release tests listed"; a later complaint on lot 4, or on a property the release tests do not cover, reopens the row. Write the scope with lot numbers, dates, test methods and units, so another engineer can tell whether a new case falls inside it.
When more than one hypothesis survives, keep each, with a confirmation test, an owner and a due date; a compound cause is valid when the evidence supports each part.
What does an illustrative root cause hypothesis log entry look like?
An illustrative root cause hypothesis log entry for a cratering complaint shows four hypotheses frozen on one date and judged against the same evidence. The case is invented to show the form; it describes no real customer, product or result.
A customer reports craters on panels coated with a waterborne coating on two of three spray lines from 2026-10-01 (illustrative). The team freezes four hypotheses on 2026-10-05 (illustrative). The table below shows how the same evidence decides between them.
Hypothesis (illustrative) | Prediction | Refutation condition | Verdict (illustrative) |
|---|---|---|---|
H1 Silicone contamination in coating lot 2 | Craters with lot 2 on all lines | Lot 2 sprayed on line C without craters | Ruled out for lot 2 on lines A to C, 2026-10-07 |
H2 Oil carry-over from compressor B | Craters only on lines fed by compressor B | Craters persist with bottled nitrogen | Open, nitrogen trial due 2026-10-12 |
H3 Change in substrate cleaner | Craters began with new cleaner batch | Craters on panels cleaned with old batch | Open, old batch not retained |
H4 Measurement change in crater count | Count rose when inspection light changed | Same count under old light | Ruled out, light unchanged since 2026-09 |
H1 is ruled out only within its scope: lot 2 on these three lines. H2 is the leading candidate because lines A and B share compressor B, but it stays open until the nitrogen trial runs. H3 stays open because the record that would refute it does not exist, which the log states instead of guessing.
The illustrative example is kept on a separate page of the Word template and on its own sheet in the Excel log, so the blank form stays clean.
When does a root cause hypothesis log not fit?
A root cause hypothesis log does not fit when the cause is already confirmed by an authoritative source, when the task is immediate safety response or containment, or when the failure has one obvious cause and one decisive check. In those cases the log adds work without changing the decision.
It also does not replace a standard method a customer requires. Some customers require their own corrective action format, such as an 8D report, and the AIAG CQI-20 Effective Problem Solving Guide (2nd edition) sets out its own stage structure with progress check questions (AIAG, accessed 2026). The hypothesis log feeds the root cause discipline of such a report; it does not stand in for the customer's form.
The log is weak when the comparison scope is missing. With no unaffected line, lot or period to compare against, predictions cannot be comparative, and many refutation conditions cannot be tested. Record that gap in section 1 and expect more hypotheses to stay open.
It is not a tool for assigning blame. IEC 62740:2015 states that its techniques "are not designed to assign responsibility or liability, which is outside the scope of this standard" (IEC, 2015). Keep commercial responsibility, credits and claims in a separate document, because a log read as a liability record invites people to argue with the evidence instead of recording it.
Finally, the template does not judge whether corrective actions worked. That belongs to verification in the corrective action report, against success criteria set before the action.
Root cause hypothesis log checklist
Use this root cause hypothesis log checklist before the evidence is opened and again before the verdicts are signed.
Before the evidence:
The observed failure is stated as measurements with units, location and time.
Every statement is marked observed or reported.
The comparison scope names affected and unaffected units and the period.
The change list covers lot, supplier, set points, cleaning, season, water, measurement and staffing.
Every cause domain has a hypothesis or a written reason why it has none.
Every hypothesis has a mechanism chain, a prediction and a refutation condition.
The freeze date is entered for every row.
Each hypothesis names the first record to examine and who holds it.
After the evidence:
Every evidence item is judged against every hypothesis.
Items consistent with all hypotheses are marked as not diagnostic.
Hypotheses added after the freeze are marked as added later.
Each verdict is confirmed, ruled out or open; nothing unchecked is ruled out.
Each ruled-out hypothesis has an exclusion scope with lots, dates and methods.
Each open hypothesis has a confirmation test, an owner and a due date.
Lubinpla's root cause analysis task works in the same order on your own documents and site history: it separates observations from reported statements and freezes hypotheses, predictions and refutation conditions before it looks at the evidence.