When the Result Stayed Negative

P65.owning-failure.03 · Audience: guest, it-ml, language-pro · Prerequisites: The Failure Story That Builds Trust

Post-Mortem Communication — Blameless but Accountable Raising an Ethical Objection — Concrete, Not Preachy

LLM grading (this page):Real gradingOff — no LLM calls

A data scientist spends a quarter building the model everyone believed in. The controlled experiment comes back: it does not beat the baseline. Not "needs tuning" — does not beat it, on a fair test she designed herself. Across town, a language school ran a semester-long conversation-club format the whole staff was excited about; attendance data and end-of-term assessments say it moved nothing. Neither of these is an incident — nothing broke, nobody erred, there is no post-mortem timeline with a guilty fork in it. And yet both people now face the hardest communication in professional life: standing in front of the people who funded a quarter of work and saying "the answer is no, and I recommend we stop" — without spin, without ceremony of self-blame, and without letting the room conclude the quarter was wasted. That skill has a structure, and this module teaches it.

🗺️ So far you have built: the ownership craft of this pillar: the failure story with your mechanism named (module 01), and — if you've taken it — the post-mortem discipline of module 02: facts a reader can check, blameless but accountable. Both dealt with things that went wrong. This module deals with something different: work done well that returned the answer nobody wanted.

In this module you will: the negative result as a finding — what a 'failed' quarter actually bought and how to price it; the kill decision — making the call to stop, and why the person who ran the work must be the one to recommend it; the salvage inventory — what transfers, what stops, said concretely; and the no-happy-ending telling: reporting a negative result without spin, and without the self-flagellation that misreads a finding as a fault. The ladder closes it: write the kill recommendation for an initiative you are given, then tell a negative result of your own.

Step 1 / 6 — A negative result is a finding

Distinguish two things that feel identical at 6 p.m. on results day. A failure is when the work went wrong: the experiment was flawed, the rollout broke, a fork was missed — the material of modules 01 and 02. A negative result is when the work went right and the answer was no: the fair test was run, and the model does not beat the baseline; the format was piloted properly, and it does not move learning. The distinction is not consolation — it changes what you owe. A failure owes an ownership statement about the mechanism. A negative result owes a defense of the method and a clear statement of the finding: "we now know this doesn't work, and here is why that knowledge is solid."

ⓘ Concept: Failure = the work went wrong; negative result = the work went right and the answer is no

rerun test: same process, same answer? → finding (defend the method, state the result) · answer depends on avoidable flaws? → failure (name the mechanism)

Why it matters — Conflating the two corrupts the telling in both directions. Treat a negative result as a failure and you will perform contrition for having discovered something true — which teaches your organization that honest tests are punished, and the next quarter's experiment will be designed to be unfalsifiable. Treat a genuine failure as 'just a negative result' and you have found a new flavor of blame-laundering. The test question: if a competent stranger reran your process, would they get the same answer? If yes, you have a finding — the quarter converted a plausible, expensive belief into knowledge, which is precisely what quarters are for; the belief was going to be spent on either way, and the experiment was the cheapest way to spend it. If no — if the answer is 'no' because the test was weak — then you are in module 01's territory, and the mechanism needs naming.

✅ Key takeaways

Separate failure (the work went wrong — owes an ownership statement about the mechanism) from negative result (the work went right and the answer is no — owes a defense of the method and a plain statement of the finding). The rerun test decides: same process, same answer means you have a finding.
Kill decisions drift by default — one more tuning round at a time. Define the stop-line (ideally before results), and when it's crossed, be the one to recommend stopping: from the person who ran the work, the kill is unanswerable and converts you from advocate to scientist.
Half the recommendation is the salvage inventory: what transfers (named, auditable — the harness, the dataset, the closed question) and what stops (named, actioned). Vague salvage reads as spin; a missing stop-list is how dead projects keep breathing.
Report it with neither spin nor sackcloth: both refuse to let the result mean what it means. Lead with the finding, defend the method, own only what was genuinely owed (a stop-line set too late — that part is module-01 material), and let the no stay a no.
As a story, the negative result needs method more than mechanism, and an ending that upgraded the deciding ('we set stop-lines before starting now') rather than redeeming the outcome. 'That no is still saving us money' is a good ending; it doesn't need to be happy.

Next: P65 complete — the owning-failure track closes the pillar

The quarter that answered "no" was not the wasted one — the wasted quarter is the one that answered nothing, or answered "maybe" forever. Model or method, curriculum or format, campaign or clause: every field pays for its knowledge in exactly this currency, and the professionals who can carry a clean no into a room without flinching are the reason their organizations stop paying twice.

Ask the mentor about this module

Ask a question about this content. The mentor explains and grounds its answer in what you are studying; asking is recorded as a learning signal, not a grade.

Images, PDF or text. Kept on this device only.

Keeping your files on this device

Keep a copy of the files I attach, in this browser

Off by default. The mentor always gets your file; this only decides whether your own copy stays here. Copies live in this browser only - they do not follow you to another device, and clearing site data removes them.

Ctrl/Cmd + Enter to send

🎓 Practice ladder

2 graded rungs · ~22 min

Two mentor-graded craft rungs. Rung 1 gives you a three-month initiative whose results came back flat and asks for the upward kill recommendation — finding, method, salvage, stop-list. Rung 2 is a negative-result story of your own, told without spin and without self-flagellation; a low-stakes example is always an accepted substitute. The mentor grades the telling, never the outcome.

Rung 1 — Recommend the kill (explorer)

Loading exercise…

Rung 2 — Your negative result, told straight (practitioner)

Loading exercise…

Try it yourself

A scratch console for this page's ideas — ungraded, nothing you run here is recorded.

Scratch console

A scratch console with the scientific stack (pandas, numpy, scikit-learn). Runs on the server — no network, resource-limited and measured.

Output appears here.

My notes on this module

Loading your notes...

Where next?

Continue

Raising an Ethical Objection — Concrete, Not Preachy →

Ethical Voice · P66

Go up a level