What to do this morning, and what it costs not to.

The same desk the search runs on, pointed at the demo workbook — a fictional register of forty-eight openings, run like a CRM: every row carries a sales-cycle status, a stage clock, the people in the loop and a dated next step. Every number on this page is computed when it loads, from the rows, events and runs held below — nothing is typed in to look decisive.

The register:

What to do next

computed priority · four sorts

Priority = rows unblocked ÷ √minutes, ×1.6 if dated, ×1.3 if delay has a cost. Follow-ups are generated from the stage clock, not written by hand.

ActionPriorityUnblocksMinutesAge · cost of delay

The register, measured

The funnel, with conversion

rows that reached each point of the sales cycle, and the rate from the point before

Variant results

vanilla control against the full package · response rate per arm

Channel yield, not channel share

rows per source, and how many reached the top two tiers

Rows by tier

computed from each row's ratings and the three guards

Geography gate

Remote or Denver passes · Gap is a flag for a human · Unknown cannot be gated

The opportunities — click one for its timeline

status · stage clock · who you are talking to · next step

Employer	Role	Tier	Score	Status	Days in status	Last touch	Talking to	Next step	Due	Geography	Source

Pick a row

Every send, reply, call and note is a dated event on the row — the activity timeline a CRM keeps, held in the register rather than in anyone's inbox.

The sales cycle — what each status means

status is the latest stage actually held · a booked interview is a next step

Status	What it means	Enters when	Who sets it	Forecast weight	Rows now

Tier is fit — how good the opportunity is, computed from the rubric. Status is where the deal stands. They answer different questions, and neither is typed into the other. Forecast weights are the operator's settings, not learned.

Machinery — the unattended runs

last twenty-four hours

When	Agent	What it did	Rows written	Refused	Outcome

Refusals are logged as loudly as writes. A run that fails says so; a run that finds nothing says so — success and silence never look the same.

Tracker Bridge — every write, every refusal

the demo workbook's Bridge Log, line for line

When	Batch	Line	Op	Tab	Key	Column	Status	Detail

The bridge reads the newest batch document, applies it, logs every line — including what it refused and why — and renames the document so it is never applied twice. It never writes a score, a tier or a human ruling.

How a row is scored

the rubric is published before the first row arrives

The operating contract

two lanes · one arbiter · two gates

The heavy lift

Research, sourcing, scoring, the register, every file and its version, the ledger of its own faults. Reads a file before it says anything about the file.

The second model

Language and tone, and a standing job to find what the first one got wrong. Never restates a figure or a name it did not copy from the source.

The arbiter

A human rules every disagreement between the two, as a two-column choice with the strongest case for each side. Neither model settles one alone.

Never decides which opportunities to pursueNever sends anythingNever scores a company without a published rubricNever writes a protected column