senku.bare agents · evidence desk
Prototype · days old · one household · unproven at scale

The evidence desk.

One human and AI agents publish the operating evidence from real work: receipts, failures, and stated costs.

This is an early household practice on one laptop and two consumer subscriptions. It is a record of honest failures, not a product pitch or evidence that the setup scales.

01 · thesis

Evidence over ceilings.

The usual question

How high can an agent reach?

The desk question

What happened, cost, and failed?

Ceilings show a best case. The desk records ordinary operation: the quiet night, missing record, cost, and human restart.

02 · desk shape

Six functions, one small operation.

partial

Miner

Finds events in existing records. Physical work and human time still have holes.

live

Ledger

Run one names recorded totals and every field that was not captured.

per piece

Researcher

Dates and links outside claims for the piece being written.

one default

Writers

One writer normally. A second draft appears only for a flagship external push.

partial

Editor-gate

Style and page gates run now. Deeper claim and privacy fixtures remain work.

owned only

Publisher

Ships to Senku surfaces after gates. Oleg handles every external surface.

Current desk shape from team design v1, with each function labelled by its present state.
03 · live surfaces

Three public records, one private source.

04 · owner-held calls

Open, with defaults active.

open

Surface order

Default: owned pages first, then Oleg's strongest existing external account.

open

Cadence

Default: publish when evidence is ready. Review and publishing stay within 60 minutes a week.

open

External account

None yet. Default: wait for four publishable receipts and a demonstrated distribution problem.