Update to AdAgnt
Snapshot Nov 16, 2026 · 18:12 UTC · version 1.0.2
Collection source: downloaded plugin package. These snapshots do not have a confirmed matching collection source. Differences in file lists alone do not establish changes to the package.
Instructions updated for adagnt-revenue
Full description — Now says: Set the revenue assumptions AdAgnt uses (provider label and average order value), then read per-campaign ROAS estimates with scale / keep / fix / kill...
Observed in instructions or declared skills. Runtime behavior has not been tested.
Up to 3 plugins free · no card required.
All fields and original values.
Full description
Sep 30, 2026 · 23:11 UTC
Connect real revenue sources (GA4, Shopify, Stripe, Klaviyo) and judge campaigns by true ROAS — revenue divided by spend — instead of platform-reported conversion value. Use when the user asks what their ads actually earn, wants to conne...
Nov 16, 2026 · 18:12 UTC
Set the revenue assumptions AdAgnt uses (provider label and average order value), then read per-campaign ROAS estimates with scale / keep / fix / kill verdicts. Revenue is modelled as conversions × average order value, never pulled from ...
Skill instructions
Sep 30, 2026 · 23:11 UTC
Connect real revenue sources (GA4, Shopify, Stripe, Klaviyo) and judge campaigns by true ROAS — revenue divided by spend — instead of platform-reported conversion value. Use when the user asks what their ads actually earn, wants to conne...
Nov 16, 2026 · 18:12 UTC
Set the revenue assumptions AdAgnt uses (provider label and average order value), then read per-campaign ROAS estimates with scale / keep / fix / kill verdicts. Revenue is modelled as conversions × average order value, never pulled from ...
Compare saved observations
Download comparison JSONSKILL.md text changes
− removed · + added
--- before +++ after @@ -1,47 +1,49 @@ --- name: adagnt-revenue -description: Connect real revenue sources (GA4, Shopify, Stripe, Klaviyo) and judge campaigns by true ROAS — revenue divided by spend — instead of platform-reported conversion value. Use when the user asks what their ads actually earn, wants to connect a store or analytics source, or questions the numbers their ad platforms report. +description: Set the revenue assumptions AdAgnt uses (provider label and average order value), then read per-campaign ROAS estimates with scale / keep / fix / kill verdicts. Revenue is modelled as conversions × average order value, never pulled from GA4, Shopify, Stripe or Klaviyo. Use when the user asks what their ads earn, wants revenue-based verdicts, or questions platform-reported conversion value. --- -# AdAgnt Revenue & True ROAS +# AdAgnt Revenue Estimates & ROAS Verdicts -Ad platforms grade their own homework: the conversion value Google or Meta reports is their claim about what they drove. This skill wires in the ground truth — actual revenue from the user's store, payment processor, or analytics — and judges every campaign against it. +What this skill does, and does not do. AdAgnt does not connect to GA4, Shopify, Stripe or Klaviyo and ingests no orders from them; no such connector exists. The revenue tools take two assumptions — which provider the user sells through (stored as a label) and an average order value — and model revenue per campaign as the platform's reported conversions × that order value. Every revenue figure in this skill is an estimate built on the ad platforms' own conversion counts. Every result carries a `revenue_basis` field saying so; repeat it to the user in plain words each time you show a number. Never call these figures real, measured, verified or provider-confirmed revenue, and never say a store or analytics account is "connected". -## Step 1 — Connect a revenue source +Why it still earns its place: conversion *value* is what each platform claims it drove, and it is often the weakest number in the report. Modelling revenue from conversion counts and one order value the user vouches for gives a consistent yardstick across platforms. The verdicts rank campaigns against each other; they are not an accounting of profit. -Check what's already wired up with `list_revenue_sources`. If nothing is connected (or the user mentions a new source), onboard one with `connect_revenue_source`: +## Step 1 — Set the revenue assumptions -- `source` — one of `ga4`, `shopify`, `stripe`, `klaviyo`. Pick whichever is closest to the money: Stripe or Shopify if they charge there, GA4 if e-commerce tracking is solid, Klaviyo for email-attributed flows. -- `property_id` — the store/property identifier. Optional in sandbox; ask for it on live accounts. -- `average_order_value` — sandbox only: calibrates the simulated revenue stream. Ask for a realistic figure rather than defaulting, so sandbox numbers look like the user's business. +Check what is saved with `list_revenue_sources`. If nothing is set, or the user wants different assumptions, save them with `set_revenue_assumptions`: -Connecting is a write, but a safe one — it creates a data connection, not an ad-account change. Still confirm the source and identifier with the user before calling, and verify afterward with `list_revenue_sources` (check sync status, not just presence). +- `source` — one of `ga4`, `shopify`, `stripe`, `klaviyo`: the provider the user sells through. It is kept as a label on the assumptions only; nothing is connected to it or pulled from it. Say that before you call. +- `property_id` — optional store or property identifier, stored next to the label for the user's reference. +- `average_order_value` — the number that drives every estimate (default 120). Ask for a realistic figure for the user's business rather than defaulting; a wrong order value scales every verdict with it. -One source is enough to start. If the user connects several, be explicit that revenue is joined from all of them — don't double-count a Shopify order that also shows up in Stripe; ask which source is authoritative for order revenue. +Saving the assumptions is a write, but to AdAgnt's own records, not to any ad account or provider. Confirm the provider label and order value with the user first. The call backfills a 90-day modelled revenue baseline from every connected platform's campaign conversions and returns `revenue_basis` and a message saying the figures are modelled; pass that on. Verify with `list_revenue_sources` afterward. -## Step 2 — Read true ROAS +One set of assumptions is enough. If the user saves several, the modelled events are joined from all of them, so make sure the order values describe different parts of the business rather than the same orders twice. -`get_true_roas` returns per-campaign revenue ÷ spend from the connected sources, ranked worst-to-best, each with a verdict. What the verdicts mean and what to do: +## Step 2 — Read the ROAS estimates -- **kill** — spend with no revenue to show for it over the window. Recommend pausing; the burden of proof is on the campaign now. Check one thing first: is the sales cycle longer than the date range? A 30-day window will condemn a 60-day B2B cycle unfairly. -- **fix** — real revenue, bad ratio. Something specific is broken: targeting, landing page, bid strategy, or creative. Diagnose before touching budget — `adagnt-performance` has the anomaly tools. -- **keep** — earning its budget. Leave it alone. Not every campaign needs an intervention this week. -- **scale** — high true ROAS with headroom. Propose a budget increase in steps (20–30% at a time), not a doubling — efficiency usually falls as spend rises. +`get_roas_estimates` returns, per campaign, modelled revenue ÷ spend, ranked worst-to-best, each with a verdict. Thresholds: kill below 0.8, fix from 0.8, keep from 1.5, scale from 3.0. The tool name says "true"; the numbers are modelled, so present them as estimates. What the verdicts mean and what to do: + +- **kill** — spend with no modelled revenue to show for it over the window, which means few or no reported conversions. Recommend pausing; the burden of proof is on the campaign now. Check one thing first: is the sales cycle longer than the date range? A 30-day window will condemn a 60-day B2B cycle unfairly. +- **fix** — conversions are coming in, but the ratio is bad at the user's order value. Something specific is usually broken: targeting, landing page, bid strategy or creative. Diagnose before touching budget — `adagnt-performance` has the anomaly tools. +- **keep** — earning its budget on the estimate. Leave it alone. Not every campaign needs an intervention this week. +- **scale** — high estimated ROAS with headroom. Propose a budget increase in steps (20–30% at a time), not a doubling — efficiency usually falls as spend rises. Set `min_spend` so tiny campaigns don't pollute the ranking (default 10 is fine; raise it on large accounts). Use the same `date_range` the user's targets are stated in. Present the ranking worst-first, exactly as returned — the user's biggest losses are the lede, not the wins. -## Step 3 — Explain the attribution gap +## Step 3 — Explain the gap between modelled revenue and platform value -`get_revenue_attribution` returns, per campaign and platform, both the platform-reported conversion value and the actual attributed revenue — and the gap between them. When the user asks why the numbers disagree (they will), the honest explanation: +`get_revenue_attribution` returns, per campaign and platform, the platform-reported conversion value, the modelled revenue (conversions × order value), spend, both ROAS figures and the gap between them. When the user asks why the numbers disagree, the honest explanation: -- **Platforms over-claim.** Each platform attributes any conversion it touched, so the same order is claimed by Google *and* Meta. Summing platform-reported value counts money twice; connected revenue counts it once. -- **View-through and window inflation.** A platform may claim a sale because someone saw an ad a week before buying through another door. Revenue-side attribution is stricter. -- **Under-reporting happens too.** Blocked tracking, iOS privacy limits, and offline sales make platforms miss revenue they genuinely drove. A gap in either direction is information. +- **Different inputs, same conversions.** Platform conversion value is the value each platform assigns to the conversions it claims; the modelled figure applies the user's one order value to the same conversion counts. A gap means the platform's per-conversion value differs from the user's order value. It does not mean one side has seen the orders. +- **Platforms over-claim across each other.** Each platform attributes any conversion it touched, so one order can be counted by Google *and* Meta. The modelled figure inherits this: it is built on each platform's conversions, so it does not de-duplicate across platforms either. Say so before summing anything. +- **Under-reporting happens too.** Blocked tracking, iOS privacy limits and offline sales make platforms miss conversions, and the model misses them for the same reason, because it starts from the same counts. -The rule of thumb to give the user: **platform numbers for optimizing within a platform, true ROAS for deciding between platforms and for judging total spend.** A campaign is never "actually profitable" on platform-reported value alone. +The rule of thumb to give the user: **platform-reported value for optimizing within a platform, the modelled estimate for a consistent comparison across platforms, and neither for a profit claim.** A campaign is never "actually profitable" on either figure alone; that needs order data AdAgnt does not have. ## Step 4 — Act on it -Verdicts are recommendations, not actions. Any pause, budget change, or reallocation that follows goes through the normal write-tool discipline: exact payload shown, explicit yes, one call, verify with a read tool. For cross-platform budget moves based on true ROAS, `optimize_cross_platform_budget` files a plan to the approval queue by default — keep it that way. Record what was decided with `update_strategy` (`append_learning`) so next month's review starts from this one's conclusions. +Verdicts are recommendations, not actions. Any pause, budget change or reallocation that follows goes through the normal write-tool discipline: exact payload shown, explicit yes, one call, verify with a read tool. For cross-platform budget moves based on the ROAS estimates, `optimize_cross_platform_budget` files a plan to the approval queue by default — keep it that way. Record what was decided, and the order value it rested on, with `update_strategy` (`append_learning`) so next month's review starts from this one's conclusions.
Full technical diff · 2 changed fields
changed /description
"Connect real revenue sources (GA4, Shopify, Stripe, Klaviyo) and judge campaigns by true ROAS — revenue divided by spend — instead of platform-reported conversion value. Use when the user asks what their ads actually earn, wants to connect a store or analytics source, or questions the numbers their ad platforms report."
"Set the revenue assumptions AdAgnt uses (provider label and average order value), then read per-campaign ROAS estimates with scale / keep / fix / kill verdicts. Revenue is modelled as conversions × average order value, never pulled from GA4, Shopify, Stripe or Klaviyo. Use when the user asks what their ads earn, wants revenue-based verdicts, or questions platform-reported conversion value."
changed /skill_md_contents
"---\nname: adagnt-revenue\ndescription: Connect real revenue sources (GA4, Shopify, Stripe, Klaviyo) and judge campaigns by true ROAS — revenue divided by spend — instead of platform-reported conversion value. Use when the user asks what their ads actually earn, wants to connect a store or analytics source, or questions the numbers their ad platforms report.\n---\n\n# AdAgnt Revenue & True ROAS\n\nAd platforms grade their own homework: the conversion value Google or Meta reports is their claim about what they drove. This skill wires in the ground truth — actual revenue from the user's store, payment processor, or analytics — and judges every campaign against it.\n\n## Step 1 — Connect a revenue source\n\nCheck what's already wired up with `list_revenue_sources`. If nothing is connected (or the user mentions a new source), onboard one with `connect_revenue_source`:\n\n- `source` — one of `ga4`, `shopify`, `stripe`, `klaviyo`. Pick whichever is closest to the money: Stripe or Shopify if they charge there, GA4 if e-commerce tracking is solid, Klaviyo for email-attributed flows.\n- `property_id` — the store/property identifier. Optional in sandbox; ask for it on live accounts.\n- `average_order_value` — sandbox only: calibrates the simulated revenue stream. Ask for a realistic figure rather than defaulting, so sandbox numbers look like the user's business.\n\nConnecting is a write, but a safe one — it creates a data connection, not an ad-account change. Still confirm the source and identifier with the user before calling, and verify afterward with `list_revenue_sources` (check sync status, not just presence).\n\nOne source is enough to start. If the user connects several, be explicit that revenue is joined from all of them — don't double-count a Shopify order that also shows up in Stripe; ask which source is authoritative for order revenue.\n\n## Step 2 — Read true ROAS\n\n`get_true_roas` returns per-campaign revenue ÷ spend from the connected sources, ranked worst-to-best, each with a verdict. What the verdicts mean and what to do:\n\n- **kill** — spend with no revenue to show for it over the window. Recommend pausing; the burden of proof is on the campaign now. Check one thing first: is the sales cycle longer than the date range? A 30-day window will condemn a 60-day B2B cycle unfairly.\n- **fix** — real revenue, bad ratio. Something specific is broken: targeting, landing page, bid strategy, or creative. Diagnose before touching budget — `adagnt-performance` has the anomaly tools.\n- **keep** — earning its budget. Leave it alone. Not every campaign needs an intervention this week.\n- **scale** — high true ROAS with headroom. Propose a budget increase in steps (20–30% at a time), not a doubling — efficiency usually falls as spend rises.\n\nSet `min_spend` so tiny campaigns don't pollute the ranking (default 10 is fine; raise it on large accounts). Use the same `date_range` the user's targets are stated in.\n\nPresent the ranking worst-first, exactly as returned — the user's biggest losses are the lede, not the wins.\n\n## Step 3 — Explain the attribution gap\n\n`get_revenue_attribution` returns, per campaign and platform, both the platform-reported conversion value and the actual attributed revenue — and the gap between them. When the user asks why the numbers disagree (they will), the honest explanation:\n\n- **Platforms over-claim.** Each platform attributes any conversion it touched, so the same order is claimed by Google *and* Meta. Summing platform-reported value counts money twice; connected revenue counts it once.\n- **View-through and window inflation.** A platform may claim a sale because someone saw an ad a week before buying through another door. Revenue-side attribution is stricter.\n- **Under-reporting happens too.** Blocked tracking, iOS privacy limits, and offline sales make platforms miss revenue they genuinely drove. A gap in either direction is information.\n\nThe rule of thumb to give the user: **platform numbers for optimizing within a platform, true ROAS for deciding between platforms and for judging total spend.** A campaign is never \"actually profitable\" on platform-reported value alone.\n\n## Step 4 — Act on it\n\nVerdicts are recommendations, not actions. Any pause, budget change, or reallocation that follows goes through the normal write-tool discipline: exact payload shown, explicit yes, one call, verify with a read tool. For cross-platform budget moves based on true ROAS, `optimize_cross_platform_budget` files a plan to the approval queue by default — keep it that way. Record what was decided with `update_strategy` (`append_learning`) so next month's review starts from this one's conclusions.\n"
"---\nname: adagnt-revenue\ndescription: Set the revenue assumptions AdAgnt uses (provider label and average order value), then read per-campaign ROAS estimates with scale / keep / fix / kill verdicts. Revenue is modelled as conversions × average order value, never pulled from GA4, Shopify, Stripe or Klaviyo. Use when the user asks what their ads earn, wants revenue-based verdicts, or questions platform-reported conversion value.\n---\n\n# AdAgnt Revenue Estimates & ROAS Verdicts\n\nWhat this skill does, and does not do. AdAgnt does not connect to GA4, Shopify, Stripe or Klaviyo and ingests no orders from them; no such connector exists. The revenue tools take two assumptions — which provider the user sells through (stored as a label) and an average order value — and model revenue per campaign as the platform's reported conversions × that order value. Every revenue figure in this skill is an estimate built on the ad platforms' own conversion counts. Every result carries a `revenue_basis` field saying so; repeat it to the user in plain words each time you show a number. Never call these figures real, measured, verified or provider-confirmed revenue, and never say a store or analytics account is \"connected\".\n\nWhy it still earns its place: conversion *value* is what each platform claims it drove, and it is often the weakest number in the report. Modelling revenue from conversion counts and one order value the user vouches for gives a consistent yardstick across platforms. The verdicts rank campaigns against each other; they are not an accounting of profit.\n\n## Step 1 — Set the revenue assumptions\n\nCheck what is saved with `list_revenue_sources`. If nothing is set, or the user wants different assumptions, save them with `set_revenue_assumptions`:\n\n- `source` — one of `ga4`, `shopify`, `stripe`, `klaviyo`: the provider the user sells through. It is kept as a label on the assumptions only; nothing is connected to it or pulled from it. Say that before you call.\n- `property_id` — optional store or property identifier, stored next to the label for the user's reference.\n- `average_order_value` — the number that drives every estimate (default 120). Ask for a realistic figure for the user's business rather than defaulting; a wrong order value scales every verdict with it.\n\nSaving the assumptions is a write, but to AdAgnt's own records, not to any ad account or provider. Confirm the provider label and order value with the user first. The call backfills a 90-day modelled revenue baseline from every connected platform's campaign conversions and returns `revenue_basis` and a message saying the figures are modelled; pass that on. Verify with `list_revenue_sources` afterward.\n\nOne set of assumptions is enough. If the user saves several, the modelled events are joined from all of them, so make sure the order values describe different parts of the business rather than the same orders twice.\n\n## Step 2 — Read the ROAS estimates\n\n`get_roas_estimates` returns, per campaign, modelled revenue ÷ spend, ranked worst-to-best, each with a verdict. Thresholds: kill below 0.8, fix from 0.8, keep from 1.5, scale from 3.0. The tool name says \"true\"; the numbers are modelled, so present them as estimates. What the verdicts mean and what to do:\n\n- **kill** — spend with no modelled revenue to show for it over the window, which means few or no reported conversions. Recommend pausing; the burden of proof is on the campaign now. Check one thing first: is the sales cycle longer than the date range? A 30-day window will condemn a 60-day B2B cycle unfairly.\n- **fix** — conversions are coming in, but the ratio is bad at the user's order value. Something specific is usually broken: targeting, landing page, bid strategy or creative. Diagnose before touching budget — `adagnt-performance` has the anomaly tools.\n- **keep** — earning its budget on the estimate. Leave it alone. Not every campaign needs an intervention this week.\n- **scale** — high estimated ROAS with headroom. Propose a budget increase in steps (20–30% at a time), not a doubling — efficiency usually falls as spend rises.\n\nSet `min_spend` so tiny campaigns don't pollute the ranking (default 10 is fine; raise it on large accounts). Use the same `date_range` the user's targets are stated in.\n\nPresent the ranking worst-first, exactly as returned — the user's biggest losses are the lede, not the wins.\n\n## Step 3 — Explain the gap between modelled revenue and platform value\n\n`get_revenue_attribution` returns, per campaign and platform, the platform-reported conversion value, the modelled revenue (conversions × order value), spend, both ROAS figures and the gap between them. When the user asks why the numbers disagree, the honest explanation:\n\n- **Different inputs, same conversions.** Platform conversion value is the value each platform assigns to the conversions it claims; the modelled figure applies the user's one order value to the same conversion counts. A gap means the platform's per-conversion value differs from the user's order value. It does not mean one side has seen the orders.\n- **Platforms over-claim across each other.** Each platform attributes any conversion it touched, so one order can be counted by Google *and* Meta. The modelled figure inherits this: it is built on each platform's conversions, so it does not de-duplicate across platforms either. Say so before summing anything.\n- **Under-reporting happens too.** Blocked tracking, iOS privacy limits and offline sales make platforms miss conversions, and the model misses them for the same reason, because it starts from the same counts.\n\nThe rule of thumb to give the user: **platform-reported value for optimizing within a platform, the modelled estimate for a consistent comparison across platforms, and neither for a profit claim.** A campaign is never \"actually profitable\" on either figure alone; that needs order data AdAgnt does not have.\n\n## Step 4 — Act on it\n\nVerdicts are recommendations, not actions. Any pause, budget change or reallocation that follows goes through the normal write-tool discipline: exact payload shown, explicit yes, one call, verify with a read tool. For cross-platform budget moves based on the ROAS estimates, `optimize_cross_platform_budget` files a plan to the approval queue by default — keep it that way. Record what was decided, and the order value it rested on, with `update_strategy` (`append_learning`) so next month's review starts from this one's conclusions.\n"
Full snapshot data
{
"description": "Set the revenue assumptions AdAgnt uses (provider label and average order value), then read per-campaign ROAS estimates with scale / keep / fix / kill verdicts. Revenue is modelled as conversions × average order value, never pulled from GA4, Shopify, Stripe or Klaviyo. Use when the user asks what their ads earn, wants revenue-based verdicts, or questions platform-reported conversion value.",
"included_files": [],
"name": "adagnt-revenue",
"skill_md_contents": "---\nname: adagnt-revenue\ndescription: Set the revenue assumptions AdAgnt uses (provider label and average order value), then read per-campaign ROAS estimates with scale / keep / fix / kill verdicts. Revenue is modelled as conversions × average order value, never pulled from GA4, Shopify, Stripe or Klaviyo. Use when the user asks what their ads earn, wants revenue-based verdicts, or questions platform-reported conversion value.\n---\n\n# AdAgnt Revenue Estimates & ROAS Verdicts\n\nWhat this skill does, and does not do. AdAgnt does not connect to GA4, Shopify, Stripe or Klaviyo and ingests no orders from them; no such connector exists. The revenue tools take two assumptions — which provider the user sells through (stored as a label) and an average order value — and model revenue per campaign as the platform's reported conversions × that order value. Every revenue figure in this skill is an estimate built on the ad platforms' own conversion counts. Every result carries a `revenue_basis` field saying so; repeat it to the user in plain words each time you show a number. Never call these figures real, measured, verified or provider-confirmed revenue, and never say a store or analytics account is \"connected\".\n\nWhy it still earns its place: conversion *value* is what each platform claims it drove, and it is often the weakest number in the report. Modelling revenue from conversion counts and one order value the user vouches for gives a consistent yardstick across platforms. The verdicts rank campaigns against each other; they are not an accounting of profit.\n\n## Step 1 — Set the revenue assumptions\n\nCheck what is saved with `list_revenue_sources`. If nothing is set, or the user wants different assumptions, save them with `set_revenue_assumptions`:\n\n- `source` — one of `ga4`, `shopify`, `stripe`, `klaviyo`: the provider the user sells through. It is kept as a label on the assumptions only; nothing is connected to it or pulled from it. Say that before you call.\n- `property_id` — optional store or property identifier, stored next to the label for the user's reference.\n- `average_order_value` — the number that drives every estimate (default 120). Ask for a realistic figure for the user's business rather than defaulting; a wrong order value scales every verdict with it.\n\nSaving the assumptions is a write, but to AdAgnt's own records, not to any ad account or provider. Confirm the provider label and order value with the user first. The call backfills a 90-day modelled revenue baseline from every connected platform's campaign conversions and returns `revenue_basis` and a message saying the figures are modelled; pass that on. Verify with `list_revenue_sources` afterward.\n\nOne set of assumptions is enough. If the user saves several, the modelled events are joined from all of them, so make sure the order values describe different parts of the business rather than the same orders twice.\n\n## Step 2 — Read the ROAS estimates\n\n`get_roas_estimates` returns, per campaign, modelled revenue ÷ spend, ranked worst-to-best, each with a verdict. Thresholds: kill below 0.8, fix from 0.8, keep from 1.5, scale from 3.0. The tool name says \"true\"; the numbers are modelled, so present them as estimates. What the verdicts mean and what to do:\n\n- **kill** — spend with no modelled revenue to show for it over the window, which means few or no reported conversions. Recommend pausing; the burden of proof is on the campaign now. Check one thing first: is the sales cycle longer than the date range? A 30-day window will condemn a 60-day B2B cycle unfairly.\n- **fix** — conversions are coming in, but the ratio is bad at the user's order value. Something specific is usually broken: targeting, landing page, bid strategy or creative. Diagnose before touching budget — `adagnt-performance` has the anomaly tools.\n- **keep** — earning its budget on the estimate. Leave it alone. Not every campaign needs an intervention this week.\n- **scale** — high estimated ROAS with headroom. Propose a budget increase in steps (20–30% at a time), not a doubling — efficiency usually falls as spend rises.\n\nSet `min_spend` so tiny campaigns don't pollute the ranking (default 10 is fine; raise it on large accounts). Use the same `date_range` the user's targets are stated in.\n\nPresent the ranking worst-first, exactly as returned — the user's biggest losses are the lede, not the wins.\n\n## Step 3 — Explain the gap between modelled revenue and platform value\n\n`get_revenue_attribution` returns, per campaign and platform, the platform-reported conversion value, the modelled revenue (conversions × order value), spend, both ROAS figures and the gap between them. When the user asks why the numbers disagree, the honest explanation:\n\n- **Different inputs, same conversions.** Platform conversion value is the value each platform assigns to the conversions it claims; the modelled figure applies the user's one order value to the same conversion counts. A gap means the platform's per-conversion value differs from the user's order value. It does not mean one side has seen the orders.\n- **Platforms over-claim across each other.** Each platform attributes any conversion it touched, so one order can be counted by Google *and* Meta. The modelled figure inherits this: it is built on each platform's conversions, so it does not de-duplicate across platforms either. Say so before summing anything.\n- **Under-reporting happens too.** Blocked tracking, iOS privacy limits and offline sales make platforms miss conversions, and the model misses them for the same reason, because it starts from the same counts.\n\nThe rule of thumb to give the user: **platform-reported value for optimizing within a platform, the modelled estimate for a consistent comparison across platforms, and neither for a profit claim.** A campaign is never \"actually profitable\" on either figure alone; that needs order data AdAgnt does not have.\n\n## Step 4 — Act on it\n\nVerdicts are recommendations, not actions. Any pause, budget change or reallocation that follows goes through the normal write-tool discipline: exact payload shown, explicit yes, one call, verify with a read tool. For cross-platform budget moves based on the ROAS estimates, `optimize_cross_platform_budget` files a plan to the approval queue by default — keep it that way. Record what was decided, and the order value it rested on, with `update_strategy` (`append_learning`) so next month's review starts from this one's conclusions.\n"
}SHA-256 of public snapshot: 02d38a582be3ccd64adf98a672d5c912504984c3dc61c2fcc91eb91dd82b9fe9