"Improved onboarding" is not a win. Here's the version that is.

The bullet that says nothing

Somewhere in a self-review right now, someone has written "Improved onboarding." They did the work. They mean it. And it will do almost nothing for them, because to the person reading it in a calibration room, that line is invisible.

"Improved onboarding" is a category, not an accomplishment. It describes an area you touched, not a result you produced. The reader can't tell what you actually did, whether it worked, or whether it mattered — so they round it down to "did some onboarding stuff" and move on to the next candidate whose bullet gave them something to hold onto.

“In a room where your work is being compared against other people's, a vague bullet doesn't just underperform — it hands your impact to whoever wrote theirs more concretely.”

The good news is that fixing this is mechanical. You don't need to have done more work. You need to say the work you did in a form that survives a skeptical read. Here's the recipe we use, and it works on any bullet you've got.

Start with the skeleton: situation, action, result

The STAR structure gets recommended so often it's become background noise, but the reason it works is specific: it forces the three things a reader needs and a vague bullet omits.

Situation — why this mattered. What was wrong, at risk, or worth doing. This is what makes the reader care.

Action — what you specifically did. Not "we," not the team — the part that was yours, in enough detail that it's clearly a decision or a skill, not a job description.

Result — what changed because of it. This is where wins live or die.

Most weak bullets are missing the situation and the result, and blur the action into a team effort. "Improved onboarding" is all three failures at once.

The line that makes or breaks it: the result

The result is where you turn an activity into evidence, and the sharpest tool for that is one honest number. A number does three things a verb can't: it makes the claim concrete ("cut" versus "reduced by 38%"), it makes it comparable (the reader can weigh it against other work), and it makes it memorable — the person championing you in the room has something specific to repeat.

Watch what a number does to the onboarding bullet. Before: "Improved onboarding." After: "Rewrote the new-hire environment setup after three hires in a row lost their first day to broken tooling — the next two shipped a PR on day one."

No percentage in that one, and it's still ten times stronger, because the result is concrete and checkable. When you do have a hard metric, use it. Before: "Improved API performance." After: "Cut checkout-API p95 latency from 850ms to 520ms by adding a read-through cache and killing an N+1 query — removed the top cause of timeouts during peak traffic."

Same amount of underlying work in both the before and the after. The difference is entirely in how it's stated.

What if you don't have a number?

This is the honest part, and it's where a lot of AI-assisted writing goes wrong. When there's no metric, the temptation — and the thing a "make me sound impressive" AI will happily do — is to invent one. Don't. A fabricated number is the single most dangerous thing you can put in a review, because it's the one a skip-level will ask about, and "I estimated" is not where you want that conversation to go.

When you don't have a number, you have three honest moves, in order.

Go get it. More often than you'd think, the data exists and you just haven't looked — the dashboard, the incident count, the cycle-time graph, the before/after in the PR description. Five minutes of digging turns a vague bullet into a real one.

Bound it honestly. If you can't get the exact figure, a direction or a range is still real: "roughly halved," "from most weeks to almost never," "three of the four flaky tests." Say what you can defend, not more.

Describe the concrete outcome. If there's genuinely no metric, fall back to a specific, observable change — "the next two hires shipped on day one," "the runbook is now what on-call reaches for first." Concrete beats quantified-but-fake every time.

The checklist a good mentor runs on your draft

Once you've got a draft, run it through the same short list a sharp reviewer would. These are the questions that catch the most common weaknesses.

• Is there a measurable outcome? If not, add a number, a range, or a concrete result. No result, no win.
• Is it clear what you did? Hunt for "we" and "helped," and replace them with the specific thing that was yours.
• Does it lead with a strong verb? "Was responsible for" and "worked on" bury the action — start with the thing you did: cut, shipped, unblocked, rewrote, led.
• Does it answer "so what"? If a reader could respond "and did that matter?", the result line isn't doing its job yet.

Run those four questions on any bullet and it gets better every time. It's not a talent; it's a pass.

Where the tool comes in

This checklist is exactly what we built the "strengthen a win" feature in Cookd to run automatically — you hand it a rough win, and it gives you back a sharper one-line summary, a tightened STAR rewrite, and a short list of what's missing: no measurable outcome, passive phrasing, unclear ownership. And, true to a rule we're strict about, when your note has no number in it, it doesn't invent one — it asks you for it.

But the reason we're telling you the checklist instead of just the feature is that the checklist is the real gift. It's the same pass a good mentor would make on your draft, and you can run it by hand tonight.

Pull up your last self-review or your resume. Find three bullets that read like "improved onboarding." Run the four questions on each. That half hour will do more for how your work lands than any amount of rewriting it under pressure the night before.