Overlay 5 of 5 · loop
Pro AI Task Engine Loop with 10 subagents
The same grid again, at full detail. x.Project carries the
five build steps a–j; y.Agent carries the five apply steps
k–t. Each Dev step pairs with its Ops counterpart — a with
k, b with l, and so on — and
z.Results holds the pair.
Every pair is then checked by a SOW gate belonging to
i, not to the builder: Market, Rates, Docs and Book in Sell; Input, Steps, Output and
Test in Ops; then Pay, then the Ledger. Each returns PASS / PARTIAL / NO-GO with a
grounding sub-score, and the band decision surfaces as GO · NO-GO. j and
t are the two monitors, which is what makes the loop re-entrant rather than a
pipeline that ends.
