Skip to content

Evidence and confidence

How Wires Uncrossed states what it knows.

This is the policy behind "big claims made calmly and immediately backed with structure or evidence" in the brand voice. It applies to anything a client reads: assessment findings, narrative documents, proposals, case studies, articles, readout decks, and the sentence you say out loud in the room.

It exists because WUE's primary differentiator is being diagnostic-led — the independent outside view that finds what leadership could not see. An unevidenced assertion delivered confidently is indistinguishable from the opinion they could have got internally for free. The evidence is the product.

The rule

Cite the claim, or soften the claim.

Every assertion is either backed by something the reader can go and check, or its language drops to match what is actually known. Never the third option — a firm claim with nothing behind it.

The ladder

Match the language to the backing. This is the whole policy in one table.

Backing Register Example
A measurement, or a document you can name Direct. State it. "Lead time is 19 hours, of which 14 is review wait."
Several independent, consistent observations Attributed. Name the source, not just the conclusion. "Three of the four teams described the same review queue unprompted."
One observation, or a reasonable inference from measurements Hedged and owned. Say whose reading it is. "On what we have seen so far, this reads as a review bottleneck rather than a build one."
A pattern from prior engagements, not from this one Marked as pattern. Never let experience pass as evidence. "In organisations at this stage we usually find X. We have not confirmed that here."
Nothing yet Not a finding. Say what would settle it. "We cannot yet tell whether this is capacity or prioritisation. Two weeks of queue data would answer it."

Downgrade willingly. A hedged finding that survives scrutiny is worth more than a firm one that gets picked apart in the readout, because the second one costs the credibility of everything beside it.

State the assumptions

Where a number or a claim required interpretation, say so beside the claim, not in a caveats appendix nobody reads. Assumptions that go unstated get inherited as facts by whoever reads the document next.

Worth stating explicitly:

  • A rescaled or converted reading — "the survey was 1–5, rescaled to 0–10".
  • A unit read from an ambiguous source — "assumes the dashboard reports MTTR in minutes".
  • A value that was missing and estimated — and that it was estimated.
  • Scope you did not cover — "this covers the two platform teams; the mobile group was not assessed".
  • Whose account something rests on — "as described by the on-call rota owner".
  • Recency — "the maturity self-assessment is from 2024 and has not been revisited".

Where two sources disagree, say they disagree. Reconciling them silently throws away a finding: the disagreement is often more informative than either reading.

Leave room to be wrong

Write findings so that new information revises them rather than embarrassing them.

  • Date the reading. "This is our read at the end of week one" ages honestly; "the constraint is X" does not.
  • Name what would change your mind, and say it before anyone asks. That is what makes the finding falsifiable, and a falsifiable finding is a credible one.
  • When new evidence arrives, change the position and say that you have. "We had this as a tooling constraint; the queue data shows it is a prioritisation one." Being seen to update on evidence is the diagnostic working. Defending a superseded finding is the one thing that makes an independent assessment worthless.
  • Never revise silently. A finding that quietly changes between two documents reads as either carelessness or spin.

This is not hedging everything. Confidence where it is earned is exactly the calm authority the brand voice asks for. The discipline is that the confidence tracks the evidence in both directions.

Guardrails

USE

  • "The measurement shows…" / "We observed…" / "Three teams described…"
  • "On the evidence we have…" / "Our reading is…"
  • "We have not confirmed this yet" / "This would be settled by…"
  • "We assumed X; if that is wrong, Y changes"
  • "This was our earlier reading. The new data changes it to…"
  • "Two sources disagree here, and that is itself worth attention"

AVOID

  • "Clearly" / "obviously" / "it is well known that" — all four do the work evidence should be doing
  • "Best practice says…" — an appeal to authority with no author
  • Bare industry benchmarks as proof of a local claim: "elite performers deploy daily" says nothing about this organisation
  • A generic SDO claim with no reading behind it — including a bare "elite performers deploy daily" benchmark, whether or not someone still calls the four measures DORA
  • "The problem is…" where "our reading is…" is honest
  • Percentages with no denominator, and any figure whose source you could not name if asked
  • Passive attribution — "it has been suggested that" — which hides who said it
  • Stacking vivid symptoms in front of one weak measure so the entry reads well-evidenced

Never

  • Present a pattern from another engagement as a finding from this one.
  • Quantify an impression. An invented number is worse than an honest adjective, because it survives into someone's business case.
  • Bury an assumption in a footnote and state the conclusion firmly in the body.
  • Round a hedge away between the draft and the client version.

In a HoEN narrative specifically

The narrative format puts the discipline in the structure: the bold thesis is the claim, ### Symptoms is what it looks like, and ### Signals & measures is what backs it.

  • An entry with no ### Signals & measures is a claim with no evidence. Hedge the thesis or demote the entry.
  • Symptoms are not evidence. They are what a client would recognise; measures are what you would show if they disagreed. Do not let three vivid symptoms stand in for a measure.
  • Put each interpretation in ### Signals & measures as its own bullet, so it is read with the evidence rather than after it.
  • ## tbc is the honest home for something real but unplaced. Use it instead of attaching a thin claim to the nearest need.
  • Where the ranking disagrees with the model's own position, the narrative says why. An unexplained up 7 is an assertion.

See hoen-data-assist for the working format and its validator.