How is this different from an audit?
An audit answers where you are. Monitoring answers whether that is moving, and in which direction — which only means something if the second reading was taken the same way as the first. The work is less about running more measurements than about refusing to let the conditions drift between them.
What exactly stays fixed?
Five things are sealed at the first run and held for every run after it. If any one of them changes, the comparison is reported as a new baseline rather than a delta.
- Query set versionThe approved panel is version-sealed. Adding a question opens a new series instead of extending the old one.
- Measured surfacesThe same fifteen surfaces, named exactly as they were named at t0 — an interface renamed by its vendor is recorded, not silently merged.
- Repetition countSeven runs per cell, every time. Reading a query once and reading it seven times produce different numbers, so the count is part of the contract.
- Judging rulesWhat counts as a citation, what counts as a mention, and how an ambiguous answer is resolved — written down before the run, not after seeing it.
- Sealing momentEach run closes with a hash receipt: interface, model, capture time and settings per cell. A later reading is compared against a record that cannot be quietly edited.
The full protocol behind these five, including why seven runs and how the intervals are computed, is published on the methodology page.
Monthly spot read, or quarterly full grid?
Two rhythms, one grid. The monthly read is narrow enough to catch movement early; the quarterly read is wide enough to settle the quarter’s verdict.
A quarter, as it is scheduled
Month 1
Baseline sealed
The panel is approved and version-sealed, the first run closes with its receipt, and that record becomes the t0 every later reading is compared against.
Month 2
Spot read
The same panel, run again on the same rules — month-over-month movement, new and dropped sources, and a written note on what to watch next.
Month 3
Full grid and quarter close
A full-width run with every cell status shown, joined to the two spot reads on one panel, closing with a decision deck and the complete change-card ledger.
Observation lane, or execution lane?
Some teams want the reading and will execute it themselves. Others want the reading and the moves in one loop. The two are sold as separate lanes so the scope stays legible.
Observation lane
Competitive Signal — the reading, monthly
A fresh competitive read every month, settled by a quarter-end full grid, with change cards for new, dropped and worsened items. Your team executes. Competitive Signal is an observation program and is excluded from the audit credit — the credit applies to the sprints, Growth Operations and Category Leadership.
Execution lane
Growth Operations and Category Leadership
Measurement, a written decision memo and executed moves with completion receipts, in one monthly loop that settles into a quarterly delta. Both lanes are eligible for the audit credit.
What the execution lane ships week by week is on the programs page; the monthly and annual prices are on the pricing page.
How often should you re-measure?
Match the cadence of your decisions rather than a fixed calendar. AI answers move day to day — our own repeated runs and published external experiments observe the same thing — so a single audit is a coordinate and a change verdict needs repeated waves on the same protocol. For quarterly budget decisions, a monthly wave on the same grid is usually what makes the quarterly number defensible.
And a baseline cannot be created retroactively: no one can put your query panel to last quarter’s engines. The earlier the first run, the longer the series you own.
What do we need to start a series?
A sealed baseline on the same domain, taken within the thirty days before the program starts, plus the approved query panel. If a recent paid audit already sits on that domain, it becomes the t0 and the series starts from it.