Staleness is a calibrated property of the data model.

Most platforms treat data freshness as a UX concern. The dashboard shows a refresh date and trusts the reader to do the rest. The platform treats freshness as a property of the fact lifecycle. Every fact decays on a calibrated clock. When the clock runs out, the fact ages out of the executive view. The dashboard surfaces the decay to the operator. The Refresh tier converts the decay into a commercial event.

Staleness is the recurring-revenue spine of the platform. The first engagement captures the corpus. The Refresh tier keeps it alive. The decay function is what determines when the operator needs to spend again.

The decay function is not a global age-out timer. It is a per-fact function of four signals.

Salience. The corroboration count across independent conversations. A pain point named by one out of one hundred advisors decays fast. A pain point named by forty decays slowly. The corroboration count is a first-class property on the fact.

Recency pressure. A fact unrefreshed for several cycles loses retrieval weight even if no contradicting evidence has landed. Workforce knowledge that nobody has confirmed in six months is not the same as workforce knowledge that landed last week.

Conflicting evidence. A new fact that contradicts an existing one accelerates the decay of the existing fact. The old fact is either invalidated or split into a segment-scoped variant where it still applies. The mechanism is documented on The Segment Vector.

Domain half-life. The per-dimension baseline. Brand perception decays in weeks. Training gaps decay only when the training catalog or product changes. The half-life table lives on Bi-Temporal Validity.

The dashboard surfaces the decay where the operator looks. A header reads “thirty-seven percent of your brand-perception facts have aged out of validity in the last sixty days.” A per-dimension panel breaks the number down by dimension and by segment, so the operator can see whether the staleness is concentrated in a slice (one region, one tier, one customer cohort) or spread across the corpus.

The number is computed against the bi-temporal model and the segment vector. A fact whose invalid_at has fired in the window is counted as aged out. A fact whose decay score has crossed the retrieval threshold is counted as aged out. A fact that has been split into a tighter segment scope is counted as aged out for the segments it no longer covers. The number is not a marketing illustration. It is the same number the brand-partner API uses to decide what to return.

The Refresh tier is the commercial vehicle that converts the staleness clock into a billable event. Pricing is additive on top of the original engagement: six thousand dollars per additional one hundred conversations, re-piped through the same pipeline, re-enriched against the same schema, joined to the same corpus.

A new Refresh round resets the recency pressure on the dimensions the round refreshes. It does not retire the historical record. The new conversations land in the corpus with their own learned_at, their own valid_at, and their own segment vectors. The bi-temporal model holds both the new state and the historical state. The dashboard shows the operator the new state and lets them replay the historical state at will.

Ninety days after handover, the operator can opt into a Refresh cycle. That commit is the validation of the recurring-revenue thesis. The Refresh tier exists because the methodology surfaces the need for it on a clock the operator can see.

The Pulse tier is a single engagement at two hundred and fifty conversations, with no committed cadence. It exists to prove the methodology on a single-market footprint. The Macro tier is one thousand conversations at eighty-five thousand dollars, with a quarterly Refresh cadence as the default assumption. The Custom enterprise tier (for deployments at scale) ships with a weekly refresh cadence built into the methodology. The cadence is what justifies the infrastructure pricing.

A weekly cadence on a fifty-thousand-conversation corpus keeps the brand-perception map fresh enough to underwrite a brand-partner SLA. A quarterly cadence on a thousand-conversation corpus keeps the dashboard relevant without overspending on capture. The cadence is calibrated against the half-life of the dimensions the client cares about most. A pharmacy deployment can hold a longer cadence on regulatory and compliance facts. A beauty deployment cannot hold a longer cadence on brand perception.

A one-shot research deliverable is a photograph of the workforce on a single date. It tells the operator what was true on that date. The continuous corpus with a refresh cadence is a video. It tells the operator what is moving and which dimensions are moving the fastest. A signal that shows up in one percent of conversations is statistical noise in any single round. Across fifty thousand conversations refreshed weekly, the same one percent is five hundred events per week and a stable trend line.

The pricing power of the layer follows from this. A report sells for the cost of a consulting deck. A continuously refreshed intelligence corpus sells for the value of the monetization stream it enables. At scale, the model targets a small share of a multi-billion-dollar adjacent market for year-one annualised revenue. That number is unreachable from any pricing model built around tools, seats, or reports. It only works because the corpus is infrastructure, and infrastructure has a cadence the operator pays to maintain.