Previous value: -"Exact topic that must materialize as a new distribution, alongside instruction, when the user explicitly asks to add one. Without research evidence it is added as an assumption and flagged as such."New value: +"Exact topic that must materialize as a new distribution, alongside instruction, when the user explicitly asks to add one. Without research evidence it is added as an assumption and flagged as such. Other charts and targets stay; an existing one that states the same personal fact (for example personal and household income whose brackets cannot be compared) stays as a chart, and the requested one is allocated in its place."

Previous value: -"Optional exact active distribution selection from the grounding review. Supplying this uses selection mode and applies the user-authored composition directly, including exact shares: to set a chart to e.g. 70% Female / 30% Male, send the complete active set with that chart's `pct` values changed. They become the user's targets beside the evidence (shown as \"Set by you\"); no research runs and the members are re-allocated to them."New value: +"Optional exact active distribution selection from the grounding review. Supplying this uses selection mode and applies the user-authored composition directly, including exact shares: to set a chart to e.g. 70% Female / 30% Male, send the complete active set with that chart's `pct` values changed. They become the user's targets beside the evidence (shown as \"Set by you\"); no research runs and the members are re-allocated to them. Send each chart's stored `name`, `label` and `sourceIdx` from the Audience's stored grounding (structuredContent audience.grounding.distributions); the Audience's saved evidence is reattached to every chart and value it already has (names shown in the text output may be translated; use the stored ones). A chart or value it does not have yet (send `sourceIdx: -1`) is added as the user's target."

Previous value: -"The change the user wants, folded into the replayed original brief. Its reach decides how members change: a composition edit (\"make men 40%\", \"add an income distribution\") re-allocates only members whose segment changed; a trait change (\"now they are all vegan\", \"SMB owners instead of enterprise buyers\") evolves members in place (same name and portrait, rewritten profile, knowledge retrained); a population pivot (\"now Gen-Z students instead of retirees\") rebuilds every member in place with a new persona, portrait and knowledge; an additive instruction (\"and now add 10 African consumers\") adds new members within the plan limits and leaves the others untouched; a stated total for the same people (\"make it 10\", \"reduce to 10 people\") keeps the members who best hold the Audience's composition (and any member named or said to keep, \"keep the CEO\") and removes the rest from this Audience only (their Minds stay in the library, nobody is retrained), and asking for a copy or version (\"a 10-person version\") leaves this Audience as it is and makes a copy holding those members (the completed result's `resize` names who was kept and left out); anything else in the same instruction (\"make it 10 and add an income chart\") is applied right after the resize. Mind ids never change and no Mind is deleted. Use this, not query, to adjust the Audience."New value: +"The change the user wants, folded into the replayed original brief. Its reach decides how members change: a composition edit (\"make men 40%\", \"add an income distribution\") re-allocates only members whose segment changed; a trait change (\"now they are all vegan\", \"SMB owners instead of enterprise buyers\") evolves members in place (same name and portrait, rewritten profile, knowledge retrained); a population pivot (\"now Gen-Z students instead of retirees\") rebuilds every member in place with a new persona, portrait and knowledge; an additive instruction (\"and now add 10 African consumers\") adds new members within the plan limits and leaves the others untouched; a stated total for the same people (\"make it 10\", \"reduce to 10 people\") keeps the members who best hold the Audience's composition (and any member named or said to keep, \"keep Anna\", \"keep the CEO\"; a member named to remove, \"drop Anna\", always leaves; a name several members carry decides nothing and the result's notes say so) and removes the rest from this Audience only (their Minds stay in the library, nobody is retrained), and asking for a copy or version (\"a 10-person version\") leaves this Audience as it is and makes a copy holding those members (the completed result's `resize` names who was kept and left out); anything else in the same instruction (\"make it 10 and add an income chart\", \"20 more, and make women 60%\") is applied right after, once, while wording that only preserves (\"do not change their facts\") starts nothing. The parts are read by a model in any language and checked against the instruction's own words; an instruction that cannot be read in full changes nothing and is retried automatically (`AUDIENCE_EDIT_UNREAD` while queued). Mind ids never change and no Mind is deleted. Use this, not query, to adjust the Audience."

Previous value: -"Recalibration rewrites member profiles and retrains knowledge on the Mind itself, so members that also belong to other Audiences change there too. The API refuses with 409 AUDIENCE_MINDS_SHARED (listing the Minds and the other Audiences) unless this is true. Set it only after the user explicitly accepts that reach."New value: +"Recalibration rewrites member profiles and retrains knowledge on the Mind itself. Members that also belong to other Audiences are copied into this Audience first, so the other Audiences stay unchanged. Set this only when the user explicitly wants those members rewritten in place, in every Audience they belong to."

Previous value: -"Go deeper, in the background: research what the Audience still lacks — every standard dimension its evidence reports missing or assumed, plus `topics` — and add it; every chart the Audience has stays, and stronger evidence replaces weaker only for the same question or dimension. Go deeper runs in the background and this tool returns at once; get_audience_creation_progress with the audienceId follows research.phase (researching, then researched). Not combinable with distributions, query or requestedDistribution. Refused with 409 AUDIENCE_RESEARCH_RUNNING while the Audience's background research is still running (it goes deeper on its own and updates the Audience when done)."New value: +"Go deeper, in the background: research what the Audience still lacks — every standard dimension its evidence reports missing or assumed, plus `topics` — and add it; every chart the Audience has stays, and stronger evidence replaces weaker only for the same question or dimension. Go deeper runs in the background and this tool returns at once; get_audience_creation_progress with the audienceId follows research.phase (researching, then researched). Not combinable with distributions, query or requestedDistribution. Asked while the Audience's background research is still running, it is queued behind that research and starts when it finishes."

Previous value: -200New value: +4096

Previous value: -200New value: +4096

Previous value: -200New value: +4096

Previous value: -"The change the user wants, folded into the replayed original brief. Its reach decides how members change: a composition edit (\"make men 40%\", \"add an income distribution\") re-allocates only members whose segment changed; a trait change (\"now they are all vegan\", \"SMB owners instead of enterprise buyers\") evolves members in place (same name and portrait, rewritten profile, knowledge retrained); a population pivot (\"now Gen-Z students instead of retirees\") rebuilds every member in place with a new persona, portrait and knowledge; an additive instruction (\"and now add 10 African consumers\") adds new members within the plan limits and leaves the others untouched. Mind ids never change and no Mind is deleted. Use this, not query, to adjust the Audience."New value: +"The change the user wants, folded into the replayed original brief. Its reach decides how members change: a composition edit (\"make men 40%\", \"add an income distribution\") re-allocates only members whose segment changed; a trait change (\"now they are all vegan\", \"SMB owners instead of enterprise buyers\") evolves members in place (same name and portrait, rewritten profile, knowledge retrained); a population pivot (\"now Gen-Z students instead of retirees\") rebuilds every member in place with a new persona, portrait and knowledge; an additive instruction (\"and now add 10 African consumers\") adds new members within the plan limits and leaves the others untouched; a stated total for the same people (\"make it 10\", \"reduce to 10 people\") keeps the members who best hold the Audience's composition (and any member named or said to keep, \"keep the CEO\") and removes the rest from this Audience only (their Minds stay in the library, nobody is retrained), and asking for a copy or version (\"a 10-person version\") leaves this Audience as it is and makes a copy holding those members (the completed result's `resize` names who was kept and left out); anything else in the same instruction (\"make it 10 and add an income chart\") is applied right after the resize. Mind ids never change and no Mind is deleted. Use this, not query, to adjust the Audience."

Previous value: -"Optional exact active distribution selection from the grounding review. Supplying this uses selection mode and applies the user-authored composition directly."New value: +"Optional exact active distribution selection from the grounding review. Supplying this uses selection mode and applies the user-authored composition directly, including exact shares: to set a chart to e.g. 70% Female / 30% Male, send the complete active set with that chart's `pct` values changed. They become the user's targets beside the evidence (shown as \"Set by you\"); no research runs and the members are re-allocated to them."