The job-search intelligence system · Operation Hired
The method behind the numbers
This is the worksheet behind the receipts. Operation Hired is not a demo I spun up to look busy: it is the system I actually built and ran to manage a real job search, under real load, with real consequences when it got something wrong. So I kept the receipts. Here is exactly what each number means, when it was true, and how it is checked.
The numbers, defined
Every posting pulled into the local database is scored by deterministic code before I ever see it. This is the running total as of July 2026; it grows each time the pipeline runs.
How it is checked A COUNT over the scored rows in the database. Each score can be reproduced from the same source row.
In June 2026 I corrected the scoring model and re-ran every row then in the database, 5,905 of them, through the corrected model in a single batch, after backing up first. Every row re-scored and stored without a single failure.
How it is checked Re-run the batch against the backup, then diff the row counts and read the error log. Zero failed rows.
Tests cover the scoring edges, the geo logic, and the fraud heuristics. They are green before anything ships, so a change that breaks a rule is caught before it touches data.
How it is checked The test runner in the project. The count is what the last run reported, as of July 2026.
Scoring is deterministic code, weighted rules over a local database, with no model call. The ten-thousandth posting costs the same as the first: nothing.
How it is checked Read the scoring code: no network client, no API call. The cost is structural, not a rounded estimate.
The only model spend is drafting an application I have chosen to pursue, never across the whole pool. This is the mean of 76 logged Claude Opus calls, about 3,100 input and 1,060 output tokens each.
How it is checked The logged token counts for those 76 calls, priced at the published per-token rate. Read straight off the usage field.
What "0 re-score errors" does not mean
Zero errors is not zero bugs. It means the re-score batch processed all 5,905 rows without a single failure, backed up first so it was reversible. It does not mean the scoring model is perfect or that every score is the final word on a job. The score is a heuristic; what I stand behind is that it is deterministic, reproducible from the source row, and reversible. A number I could not explain or reproduce would not be on this page.
Why the $0 is real, not rounding
Scoring never calls a model. It is weighted rules running over a local SQLite
database, so a posting scores for zero tokens whether it is the first or the fifteen-thousandth.
The only place money is spent is drafting an application, and only for a job I have approved,
which is where the ~$0.04 comes from. That figure is measured from real logged
calls, not a projection, which is why it carries a tilde: real usage varies call to call.
One honest approximation
The fraud heuristics cut the flagged-noise in the board by roughly 80% in early runs. I keep that one as an approximate operational figure, not a precise metric, because the denominator shifts as listings appear and expire. When a number is an estimate, I say so and round it down; when it is exact, I give it exactly. The difference is the whole point of the page.
Can you see it?
The tool runs on a real job search, so the database holds personal data and the code is not public. What I can do is walk you through it live: the scoring rules, the test run, the re-score log, and the token usage behind the per-draft figure. The evidence is real and it is on hand; it is just not something I will paste on the open web with my own applications inside it.
What this proves, and what it does not
This does not prove your business will see these numbers. Yours would be different, and the specific figures here are point-in-time, dated on purpose. What transfers is the method: measure your own work, define every number so it means one thing, keep the receipts, and make the result reproducible and reversible. That is how I would build and measure a system inside your business, and it is why I point to this one.