How I turned “is this job worth my time?” into an explainable rubric, a pipeline that runs every weekday morning, and a brief I can read in 30 seconds.
This page covers design and build. How accurate the scores are is the next study — see the end.
About 20 new postings arrived every weekday (up to 87 on a Monday), many of them reposts. Triaging them by hand cost about an hour a day, all to answer one question: which two or three roles get a tailored application today?
Each stage does one job. Nothing downstream edits what’s upstream: the digest and brief are rebuilt from the tracker every time.
Design rule:
Select any stage to see what it does and the rule behind it.
My criteria are a written rubric. Must-haves and red flags reject a role outright; everything else starts from a base-fit judgment and moves by named amounts.
Real postings from the tracker; employers generalized.
I was the user, so I iterated daily. Every feature exists because an earlier version made me read too much or misread a good role.
Wednesday · October 21 2026
Morning Brief5
“Small, steady steps beat the occasional sprint.”
A quick networking task marked Immediate in your job-search task tracker.
A deep-effort task marked Immediate in the tracker.
See the full rundown — 21 postings evaluated, 19 passed over, 3 pre-screened out on title ⏭️ — in the rendered digest.
Pursue, 7/10 — a core SQL, experimentation and statistics match with a 2+ year bar, but over 100 people have already applied1 Salary +0 · match alone ≈ 7/103, in today’s job digest.
Maybe, 6/10 — mixed-methods research that names a cognitive psychology PhD as a plus, but it was reposted two days ago. Salary +0 · match alone ≈ 6/10, in today’s job digest.
One new issue since the last run, with a section on how research teams are hiring this quarter, in the newsletter digest.
Each role gets its strongest positive, then its biggest catch, in one sentence. Added after early briefs listed only the negatives, so 7/10 roles read like rejections.
🔄 reposted · 🆔 relisted under a new ID · ⚠️ page didn’t fully load. Reposts always sort to the bottom of their section, because they rarely convert.
A chip shows how many points pay added or took away, so a strong match is never hidden by a pay rule — and I can see the job match on its own.
The title opens the posting; “in today’s job digest” jumps straight to that role’s full card, with every field and factor.
One unattended run quietly redesigned the page, and the regression lasted five days before I caught it. The layout is now a written contract: no change unless I ask for it in words.
Mockup with placeholder content, rebuilt with the real brief’s styles.
Every role in the brief links to its card in the day’s full digest, where everything behind the score is laid out — and where the roles the brief leaves out still get a line.
21 postings evaluated: 1 Pursue, 1 Maybe, 19 Pass. 2 previously-evaluated listings reappeared and were skipped.
These listings reappeared in today’s alerts. They were evaluated on an earlier date and were not re-scored.
Pay, years bar, location and recency, plus matches and misses against my resume, each citing the experience behind it. Every Pursue and Maybe gets the full card.
The same factor chips as the waterfall above, ending in the total.
Listings I’ve already seen keep their original score. A ⭐ flags a returning Pursue, the most actionable line on the page.
Each rejection shows its deciding reason and chips, so I can audit it in seconds. ⏭️ marks roles skipped on title alone.
“Tailor” starts a tailored-resume request for that role, pre-filled.
Mockup with placeholder content, rebuilt with the real digest’s styles.
In week one, 42% of postings came back without a description, yet the scores still looked complete. I listed the possible explanations and tested each one.
Already full-size; resizing changed nothing.
Forced it to “visible” — still nothing.
The same postings loaded instantly, back to back, from another source.
Confirmed in network traffic. A server-rendered source returned all 22 test postings.
The tracker doubles as usage data. It showed that most of Monday’s pile was reposts.
Jul 22 – Oct 21. Hover or tap a bar for counts.
Next I’ll test how good its judgments are:
Counts come from the system’s own tracker, Jul 22 – Oct 21, 2026. Employer names in examples are generalized, the brief shown is a mockup with placeholder content, and pay thresholds are omitted.