A supplier that reports how many of its prototypes were stopped has converted an admission into a filter. Buyers who intend to make a decision find that credible. Buyers who want a demonstration that succeeds go elsewhere.
The second effect is the commercially useful one, and it is the reason the disclosure is not simply a cost.
The number has to be defined to do any work
Failure-rate honesty means reporting a precisely defined outcome over a disclosed set of initiatives with appropriate follow-up and uncertainty.
Four things get called failure: an unmet hypothesis, a harmful incident, a closed investment, and a failed transfer. Those answer different questions, and substituting one for another produces a number that cannot be compared with anything.
The portfolio record starts at a declared entry event — approved experiment start, say — with a stable identifier, the original hypothesis, the criteria, the owner, and the planned review. Abandoned work, renames, and scope changes are preserved. And the report states whether it is a census of that portfolio or a sample, with the exclusions and unavailable records named.
At the cutoff, evidence status is classified separately from investment and operating status, and inconclusive results, running studies, and unknown outcomes stay visible. Counts come before percentages.
The arithmetic that shows why
Take 20 started initiatives at a cutoff: eight met the declared criterion, four did not, three ended inconclusively, five are ongoing.
Four of 20 is 20 percent unmet. Among the 12 with a met-or-unmet classification, four of 12 is 33 percent unmet. Eight initiatives have no resolved classification at all.
Both percentages are correct and they answer different questions. A showcase of the eight met cases displays eight successes out of eight selected cases and describes nothing about the cohort.
None of those fractions is a shutdown rate, a rate of harmful incidents, or a prediction of eventual outcomes.
Two counting rules keep it honest. Related variants are counted under a declared unit, rather than treating every successful retry as a new initiative while retaining one failed predecessor. And ongoing projects have not had the same opportunity to reveal failure, so a rate among completed work carries that conditioning.
Survival needs the same discipline
The same problems appear in twelve-month survival reporting, and the missing-outcome bound is the useful instrument.
Take 20 transfers whose twelve-month anniversaries have passed: 12 confirmed operating, four confirmed not operating, four unknown. Confirmed operation is 60 percent. Allowing the unknown cases to fall either way gives a logical range of 60 to 80 percent. The complete-case figure of 12 of 16 — 75 percent — must not be presented as the observed proportion for all 20.
That range is a missing-outcome bound rather than a confidence interval, and it is more honest than either endpoint alone.
Two refinements matter. Non-response after the anniversary is a missing outcome, not observed success. And a planned retirement because the need ended counts as non-operation in a snapshot without being an adverse investment outcome — so closure reasons are retained rather than collapsed.
Duration reporting divides the same way. Four operate engagements starting together, where one transfers at 60 days, one at 90, one retires at 45, and one is still running at day 100, produce a median of 75 days among observed transfers and 60 days among all observed endings. The still-running engagement contributes 100 observed days and no endpoint. Retirement is not transfer.
What the disclosure buys
The mechanism runs through buyer selection rather than through reputation.
A buyer choosing between a supplier reporting a 33 percent unmet rate with its denominators and a supplier reporting nothing has information about the first and none about the second. If the buyer’s actual objective is a decision, the first is the one whose evidence they can use.
The published record also constrains the supplier’s own behaviour, which is the second-order effect. A firm that reports stopped work has a standing reason to stop work honestly, because the alternative is a portfolio record that will not survive contact with the next transfer.
The framing has to stay accurate to be useful. A long-running pilot with a current question, authorised exposure, and a funded operating owner is not a failure because it is long. Duration alone does not qualify anything, and a framing that treats all pilots or all large procurements as failures is making a claim about prevalence that nothing here supports.
The rule
What stays fixed is that a rate carries its definition, its denominator, and its missingness. What changes is which outcome the rate describes, and the same portfolio produces several defensible numbers.
Not to be confused with
A benchmark. No industry rate is established, and none of these figures is observed evidence. They are arithmetic demonstrating what the denominators do.
An accusation. A framing that names an unresolved responsibility at the point of increased commitment is testable. One that names an enemy is positioning, and it should be tested with the intended audience before it is published.