Problem solving / RCA

Evidence register

Evidence register is licensed CC BY 4.0. Attribution: Katafacts (katafacts.com).

1 · What it is

What it is

An evidence register is a real, checkable answer to 'do we actually know that, or are we just saying it' — a fixed set of claim categories, and for each one, the real evidence (or lack of it) actually backing it. 'Management by fact, not opinion' is the whole discipline: a claim with no evidence logged, or evidence that's never been re-confirmed, is named directly rather than quietly treated as settled. Evidence can decay — a root-cause claim gets superseded by a more careful later analysis, a customer proof point ages — so `verified` is something the practitioner re-confirms, never assumed true forever once entered.

2 · When to use it

When to use it — and when not to

Use it when

Specific claims are already circulating (a root cause, a fix that 'worked,' a customer who 'loves it') and you want to know honestly which ones are backed by real evidence.
A decision is about to be made that assumes a claim is settled fact, and you want to check that assumption before committing further resources to it.
You want to know which claim areas have no evidence logged at all yet, not just which claims feel confidently stated.

Not when

You don't have a real claim to check yet — this register verifies claims already in circulation, it doesn't generate new ones.
The claim is genuinely trivial or one-off, not something a real decision depends on — the overhead isn't worth it yet.

3 · How to fill it in

How to fill it in

Scope: What set of claims does this register cover?
Categories: List every claim category you want checked for real evidence, even ones you expect are still empty.
Evidence items: For each item: the category it backs, the evidence type, source, date, and whether it's actually been verified — your own facts, never guessed.
Narrative: Drafted from the register's own computed coverage.

4 · What good looks like

What good looks like

The example below checks three claims already circulating about Line 2's changeover procedure and its recent on-time recovery — including one, honestly marked unverified, that the attribution reality check already showed could just as easily be explained by a concurrent confound rather than the change everyone's crediting.

Same example, as a downloadable xlsx workbook.

Download .xlsx

Evidence register

Line 2 — evidence register

Checks whether specific operational claims already circulating about Line 2's new changeover procedure and its recent on-time recovery are actually backed by real evidence, or just being asserted.

Priya Nair, Line Supervisor · 2026-03-24

Scope

Checks whether specific operational claims already circulating about Line 2's new changeover procedure and its recent on-time recovery are actually backed by real evidence, or just being asserted.

Categories

Nozzle cleaning step risk
Changeover time baseline
Sequencing rule effectiveness
Queue board update disciplineNo evidence yet

Evidence items

Nozzle cleaning step risk

The nozzle cleaning step is the changeover procedure's highest-risk failure mode

documented-case · FMEA on the Line 2 changeover procedure — scored and ranked, not a gut-feel list · 2026-02-20

Changeover time baseline

Line 2's baseline changeover time is approximately 45 minutes

measured-data · Line 2 booth changeover log (currently a paper form, transcribed weekly) · 2026-03-10

Sequencing rule effectiveness

Unverified

The new small-batch sequencing rule caused the on-time shipment rate to recover

anecdotal · Informal — whoever presents the finding at the follow-up review · 2026-03-17

Narrative

Two claims are well-evidenced: the nozzle cleaning step's risk ranking traces to a real, scored FMEA rather than a gut feel, and the 45-minute changeover baseline traces to measured log data. The third claim — that the sequencing rule itself caused the on-time recovery — is honestly marked unverified, because the attribution reality check already ran on this exact recovery and found the credit swings entirely depending on which model you use: a concurrent seasonal dip in order volume could just as easily explain it. Treating that claim as confirmed here would quietly undo the honesty that reality check already established. A fourth category, queue board update discipline, has no evidence logged at all yet — a real gap worth closing before assuming that FMEA-flagged risk is under control.

5 · Common mistakes

Common mistakes

Marking a claim verified because it sounds right or everyone already believes it.
A confident-sounding claim with no real basis is worse than an honest gap — it lets a decision get made on an assumption nobody actually checked.
Treating a claim as permanently verified once confirmed, with no re-check.
Evidence decays — a customer who once proved a claim can churn, a root cause can be superseded by a later, more careful analysis — verified status should reflect the last real check, not a one-time stamp.
Leaving a category out of the register because you already suspect it has no evidence.
An unlisted category can't be flagged as a gap — naming it and letting it come up empty is exactly how this tool surfaces the gap plainly instead of hiding it by omission.

6 · What it connects to

What it connects to

upstream

FMEA
A documented, scored FMEA is real evidence for a risk-ranking claim — a stronger source than an unverified assertion, and this register is where that distinction actually gets recorded.
Attribution reality check
When an attribution reality check finds a claim's credit swings sharply across models, that's exactly the signal this register should record as unverified rather than letting the claim stand as settled.

downstream

7 · Where AI helps

Where AI helps

Judgement — stays yours

Deciding whether something counts as real evidence or is still just an assertion
Deciding which coverage gap to close first

Analysis — AI helps

Tightening a claim's wording from rough notes into a specific, checkable statement
Drafting the narrative from the register's own computed coverage

Drudgery — automated

Identifying which categories have no evidence logged
Identifying which logged items are still unverified
Exporting to xlsx in the house format

9 · Rate this kata