gossip

How the map is computed

The studio's Decode is a map of a scope's own audience, computed from its connected accounts and the corpus window rather than written by an analyst. This is the computation stage by stage, the two rungs its output carries, and where it stands against what the field ships.

Specified, not built

01 . What goes in

A scope. One container a tenant owns, holding the accounts they connect, the slice of the open corpus they point at, and anything they upload. The map is computed over the scope and nothing outside it.

The corpus half is the house's own window: rooms, lexicon bands and segments, already at the publishable floor of k ≥ 20 before a scope can read them. The connected half is a measured network, a follower or interaction graph the tenant has the standing to connect. The distinction matters for everything below, because a judged reading of public rooms and a measured graph are different inputs and admit different operations (the house method is the first kind; this page is the second).

02 . The computation

Every stage is arithmetic. Nothing here is a person's reading, which is what lets a tenant re-run it and get the same map, and what lets the map state a failure mode instead of an opinion.

fig 1 . the pipeline, from a scope to a run

connected accountsa measured graph, held in the scope
the corpus windowrooms, bands, segments, at k ≥ 20
the graphaccounts and their ties, one adjacency
the cutcommunity detection, one named method
clusterseach at or over the floor, or not reported
profilinglexicon bands, platform time, sources, canon
personas + quantrule sets, edges, the agent vector
the runspec'd, not built

Filled dots are inputs. The spine is the map; the run hangs off its end because it is a second instrument and it is not built. The uploads enter at profiling, not at the graph: they describe a cluster, they do not draw one.

The graph is the adjacency the connected accounts give. The cut partitions it. Each cluster is then profiled against the corpus window: which lexicon bands it over-indexes on, where its members spend platform time, which sources it cites, which canon it reads. The profile is what turns a cluster into a persona with rule sets, and the edges between clusters, weighted by the ties that cross them, are the map's tension and affinity edges. The quant block on each persona is the agent vector the run reads: adoption threshold, channel susceptibilities, amplification, backlash amplification, rule weights.

03 . The cut, and its failure mode

A cut that a user cannot inspect is a black box wearing the instrument's name. So the workspace runs one named method, prints its threshold, and shows what changes when the threshold moves.

The candidates are the three families the field publishes rules for, and each fails in a way that can be stated: a co-commenting embedding merges small rooms into their neighbours as the window shortens; Jaccard over user overlap splits one community into two whenever it has two venues; a co-membership threshold returns no cluster at all below it, which is a result rather than an error. Which one the workspace runs is S8, a call the record holds open. The page will name it here the day it is ruled, and the surface will carry the threshold as a control with its effect visible.

A rule that can be re-run is worth more than a definition that can be argued with.

The field's alternative, found across the sweep of community-mapping practice, is a boundary that is judged by an analyst or bought as a licensed archetype grid. Neither can state a failure mode, because neither is an operation.

04 . What a persona is

A persona is a cluster with three things attached, and the schema is the house's audience-decode schema, unchanged by the studio.

fig 2 . one persona, its parts

PartWhat it holdsWhere it comes from
rule setsThe named rules the cluster's behaviour follows, and the tensions between them. Named, so a reader can disagree with one by name. profiling against the corpus window
edgesAffinity and tension to every other persona, signed and weighted. The map's topology; later, the run's. the ties that cross clusters
quantThe agent vector: threshold, susceptibilities, amplification, backlash, rule weights. With its own provenance block. analyst priors until a backtest exists

The third column is the point of the next section. Two of the three parts are computed from the scope; the third is not, and its provenance says so.

05 . Two rungs on one receipt

Computing a persona from a real graph moves its structure up the ladder. It does not move its agent vector. The map and its vector hold different rungs, and the receipt shows both.

fig 3 . the measurement ladder, and where each part files

directionalthe quant vector: priors, confidence low, calibration pending
thincounted, under a 14-day baseline
measuredthe structure: clusters and edges from a held graph
calibratedsurvived a backtest with a held-out episode

The underline marks the two rungs a computed map files at today. No map this house can produce files its vector above directional, and the surface says so or the copy law is broken.

This is the single most likely place for the surface to overclaim, which is why it is a rule of the receipt rather than a note in a method page: a map whose structure is measured looks calibrated, and it is not. The calibration loop that would move the vector is Hindsight's backtest, and Hindsight is spec'd, not built.

06 . Against the field

The standing directive weighs every component the house builds against what the field ships. Three columns of the market bear on this one; every vendor figure below is self-reported and none has been independently audited.

fig 4 . the nearest answers, and the distinction

The fieldWhat it doesWhat this does instead
the simulation tierArtificial Societies, Aaru, Synthetic Users, Subconscious.ai: networks of AI personas, from 300 to 5,000, that predict a response. None reads meaning; none hands the read to the people it describes. Personas carry named rules and tensions, computed from the scope's own graph and profiled against a corpus the tenant can inspect. The agent vector is labelled a prior.
community mappingA boundary judged by an analyst, or bought as a licensed archetype grid. The field publishes methods rather than definitions. One named cut with a stated failure mode and a movable threshold, on the surface.
validationTwo public accountability surfaces in the landscape, one of which contradicts its own homepage. Nobody prints a provenance rung on a persona. Two rungs on every receipt, structure and vector separately, and the ladder printed beside them.

Every disclosed venture dollar in the category sits in the first row. The house's bet on named rules over predicted response is a stance, stated as one.

07 . What is not built, and what it waits on

The computation above is a specification. The workspace serves on a seeded scope; the computed map it would run is not built, and the calls that gate it are the record's, not this page's.

The run this map feeds is a second instrument with its own method page, how the run runs, and its own state. The house's Decode over public rooms, which has issued readings, is described at thegossip.io/decode/method/; the two share a schema and nothing else.