feat: add Compound Marketing operating loop

.gitignore

@@ -1,2 +1,4 @@
 .DS_Store
 *.swp
+__pycache__/
+*.py[cod]

ARCHITECTURE.md

@@ -91,6 +91,18 @@ All positioning skills load:
 
 ## Strategy
 
+    strategy/compound-marketing
+      - foundation/marketing-os
+      - foundation/marketing-os/references/golden-path.md
+      - foundation/marketing-os/references/collaboration.md
+      - foundation/marketing-os/references/knowledge-boundaries.md
+      - foundation/marketing-os/references/learning-loop.md (when evidence may compound)
+      - strategy/program-brief, marketing/gtm, launches/gtm-plan, or strategy/one-pager
+        (selected dynamically after strategy approval)
+      - craft/editing and marketing-science/measurement (during closeout)
+      Carries one governed decision record from context through strategy, market action,
+      and learning. Optional pod guidance is subordinate to this path.
+
     strategy/compound-brand
       - foundation/marketing-os
       - marketing-science/research

CONCEPTS.md

@@ -1,5 +1,43 @@
 # Marketing OS Concepts
 
+## Compound Marketing
+
+### Compound Marketing
+The operating method in which every project produces effective marketing now and
+improves the system that produces the next project. Its governed loop is context,
+strategy, market, and memory.
+
+### Decision record
+The canonical ledger in a run's `run.json` for approved strategy, source authority,
+protected language, open questions, and rejected directions. Markdown is a generated
+human projection. Downstream work inherits the ledger's digest and stable IDs instead of
+reconstructing strategy from source material.
+
+### Governed run
+A project-local, single-writer record with one lifecycle, one routed output, one artifact,
+and explicit human gates. Closed and abandoned runs are terminal; related work uses a
+linked successor.
+
+### Handoff contract
+The machine-readable sidecar that gives a specialist the exact run, decision-record
+digest, governed IDs, unresolved questions, artifact owner, and record paths it must
+preserve.
+
+### Artifact receipt
+A machine-readable record connecting an accepted artifact to its run, decision-record
+digest, preserved decision IDs, content revision or hash, and named human verifier. The
+checker validates consistency; it does not authenticate the verifier.
+
+### Readiness
+The monotonic evidence states returned by the checker: structural readiness for coherent
+governance, operational readiness for a content-bound accepted run, and comparison
+readiness for a predeclared follow-up. A single run is never proof-ready.
+
+### Compounding proof
+Evidence that a comparable second real run inherited prior decisions and required less
+human correction without losing quality. Structural checks and retrospective cases are
+useful baselines, but do not establish this claim.
+
 ## Art-direction production
 
 ### Locked content manifest

README.md

@@ -1,29 +1,33 @@
 # Marketing OS
 
-Every's AI-native marketing operating system. It gives @EVERY and other agents a
-shared point of view, durable brand context, and repeatable workflows for running
-marketing work.
+Every's AI-native marketing operating system. Its core method is **Compound Marketing**:
+every project produces effective work now and improves the system that produces the
+next project.
 
 Marketing creates demand and cultural gravity. Growth converts that demand. Marketing OS
 connects the two through a measurable operating loop while protecting creative ambition,
 taste, and brand coherence.
 
 ## How It Works
 
-`foundation → research and audiences → positioning and voice → strategy → craft and launches → measurement → compounding`
+`context -> strategy -> market -> memory -> stronger next run`
 
-Every skill loads `foundation/marketing-os` as its root. Research and audience evidence
-inform positioning. Strategy turns that position into decisions and plans. Craft and
-launch skills produce the work. Measurement connects Marketing-sourced demand to
-subscriber quality and business economics. Completed work improves the system only when
-the learning transfers beyond the project that produced it.
+`strategy/compound-marketing` is the plain-language entry point. It initializes or
+resumes one governed, project-local run; routes its approved strategy into one selected
+program, one-pager, or GTM artifact; and closes the run through content-bound human
+receipts and measured results. Completed work improves the system only when the learning
+transfers beyond the project that produced it.
+
+Every specialist skill still loads `foundation/marketing-os` as its root. Research,
+audiences, positioning, voice, craft, launches, and measurement remain available when
+the active stage requires them.
 
 The repository map is:
 
 1. **foundation/** — The Marketing OS, the operating system underlying all skills
 2. **brand-voice/** — Voice guides for Every's master brand and each sub-brand (Cora, Spiral, Monologue, Plus One, Proof, Sparkle)
 3. **positioning/** — Positioning frameworks for Every master and each sub-brand
-4. **strategy/** — Higher-order strategy skills (messaging architecture, program briefs, one-pagers) that compose voice, positioning, and measurement
+4. **strategy/** — Compound Marketing plus higher-order strategy skills that compose voice, positioning, programs, GTM, and measurement
 5. **craft/** — Execution-level skills for art direction, copywriting, editing, naming, and channel-specific output (launch emails, LinkedIn posts, X posts, website copy, press comms)
 6. **launches/** — Orchestration skills for three launch tiers (improvement, feature, new product) that load strategy and all channel craft skills
 7. **marketing-science/** — Research, archetyping, brand equity, and measurement skills that inform positioning, programs, and decisions
@@ -43,8 +47,8 @@ The repository has eight skill families:
    collaboration rules, and native-document editing protocol
 2. **brand-voice/** — Voice systems for Every and its products
 3. **positioning/** — Durable, ruled master-brand and product positioning
-4. **strategy/** — Compound Brand, messaging architecture, measurable program briefs,
-   and executive one-pagers
+4. **strategy/** — Compound Marketing, Compound Brand, messaging architecture,
+   measurable program briefs, and executive one-pagers
 5. **craft/** — Verbal identity, art direction, copywriting, editing, naming, PR,
    social, email, web, and long-form execution
 6. **launches/** — The canonical launch process, launch calendar, brief, GTM strategy,
@@ -58,8 +62,9 @@ Optional historical cases live in `docs/cases/`. Skills never load them by defau
 
 ## What Belongs in the OS
 
-Marketing OS stores reusable methods: how to execute a launch, GTM plan, campaign, email
-series, channel strategy, measurement plan, or post-project review.
+Marketing OS stores reusable methods: how to turn context into strategy, execute a
+launch, build a GTM plan or recurring program, measure the result, and compound the
+learning.
 
 It does not store active roadmaps, launch briefs, campaign concepts, current rosters,
 temporary claims, prices, budgets, embargoes, or project calendars. Durable positioning
@@ -72,7 +77,7 @@ case.
 
 ## Current State
 
-As of July 24, 2026, Marketing OS contains 60 skills and 83 operating references.
+As of Aug. 19, 2026, Marketing OS contains 62 skills and 95 operating references.
 
 Ready for structured use:
 
@@ -88,18 +93,48 @@ Ready for structured use:
   Score, authority, and Consulting demand
 - Compounding from Douglas's edits and from campaign results without importing project
   details into unrelated work
+- A Compound Marketing golden path with canonical V2 run records, deterministic resume,
+  shared specialist handoffs, content-bound artifact receipts, lifecycle checks, and an
+  explicit proof standard
 
 Still in development:
 
 - Product-specific positioning and voice where source documents have not been ruled
 - Several execution-channel skills that remain explicit scaffolding
 - More forward tests and comparative evaluations
 - Live data integrations for repeatable quarterly reporting
+- A prospective comparable second run and independent teammate rerun required for a
+  publishable compounding claim
+
+The Brand Book and Thesis GTM cases are real retrospective baselines. They pass the
+structural record checks and intentionally report `proof_ready: false`; neither recorded
+enough prospective evidence to prove compounding.
 
 Files marked `## To do` are scaffolding, not authoritative product or channel guidance.
 
 ## Changelog
 
+### Aug. 21, 2026: Artifact Proof Gate
+
+- Added blank-project initialization, deterministic discovery and resume, legal lifecycle
+  transitions, canonical JSON, and generated human projections
+- Standardized one machine-readable handoff across program briefs, one-pagers, GTM master
+  plans, and execution-ready GTM plans
+- Required a content-bound, human-signed receipt before operational or comparison
+  readiness; the checker validates consistency without pretending to authenticate humans
+- Preserved schema V1 as legacy structural evidence and reserved proof eligibility for
+  prospective comparable operational runs by different named operators
+
+### Aug. 19, 2026: Compound Marketing V1
+
+- Added the governing Compound Marketing strategy and canonical golden path
+- Added one entry skill for strategy, recurring programs, GTM, and learning closeout
+- Added shared decision and run records that preserve human rulings across stages
+- Added signed artifact receipts and deterministic validation for decision fidelity and
+  comparable-run proof
+- Added honest Brand Book and Thesis GTM baselines without upgrading them into claims
+- Kept design, asset production, and agent-count claims outside V1
+
 ### July 24, 2026: Reusable Operating Layer
 
 - Added Marketing-sourced free subscribers, attribution, cohort quality, CAC, LTV:CAC,

STRATEGY.md

@@ -0,0 +1,97 @@
+---
+name: Compound Marketing
+last_updated: 2026-08-21
+---
+
+# Compound Marketing Strategy
+
+## Purpose
+
+Marketing work produces deliverables but discards much of the reasoning, context,
+correction, and evidence that made them good. Every new project starts too close to
+zero.
+
+## Positioning
+
+Every project must do two jobs: produce effective marketing now and improve the system
+that produces the next project.
+
+Compound Marketing preserves the decisions that matter, keeps them under human control,
+and adds only the learning that survives real use.
+
+## Users
+
+**Primary:** A marketing leader working with AI. They are hiring Compound Marketing to
+turn scattered company knowledge into strategy and GTM without surrendering authorship,
+judgment, or final taste.
+
+**First reuse audience:** Another marketer on the same team. They are hiring it to
+inherit the strategy and operating knowledge without requiring the original author to
+reconstruct the system.
+
+## Boundaries
+
+- V1 covers strategy, recurring marketing programs, GTM planning, evaluation, and
+  durable learning.
+- Design, art direction, asset production, and channel automation are deferred.
+- Multi-agent pods are optional infrastructure. Agent count and "Level 7" language are
+  not evidence that the system works.
+- Synthetic examples and structural tests may catch regressions, but cannot prove that
+  the marketing improved.
+- Project facts, temporary preferences, and isolated results do not become durable
+  doctrine automatically.
+
+_Resist a change when:_ It adds machinery without improving the core loop or making the
+system's effect easier to prove.
+
+## Key metrics
+
+- **Human review and integration burden:** Minutes spent correcting, reconciling, and
+  approving each accepted strategy or GTM artifact; recorded in the run record.
+- **First-pass acceptance:** Percentage of artifacts accepted without substantial
+  strategic rewriting.
+- **Critical defect rate:** Unsupported facts, source conflicts, diluted protected
+  language, or changed decisions reaching human review.
+- **Inherited-decision reuse:** Previously captured decisions or rules reused and
+  accepted without being reconstructed.
+- **Independent rerun success:** A different named operator completes the workflow from
+  its governed inputs without the baseline operator rebuilding the brief or system.
+
+The first publishable proof requires a comparable second run with at least 20% less
+correction or review time and no regression in acceptance or critical defects. The
+checker establishes eligibility under recorded evidence; a human separately authorizes
+the exact external claim.
+
+## Tracks
+
+### Context to Strategy
+
+Resolve company truth, customer evidence, prior decisions, protected language, and open
+questions into a governing strategy.
+
+_Why it serves the approach:_ The system cannot compound what it cannot remember or
+distinguish by authority.
+
+### Strategy to Market
+
+Turn the approved strategy into a GTM plan for a launch or a program brief for ongoing
+marketing. Preserve the audience, idea, proof, action, ownership, and decision gate.
+
+_Why it serves the approach:_ Strategy becomes valuable when it governs coherent action
+without being diluted during translation.
+
+### Market to Memory
+
+Compare drafts with human edits, evaluate outcomes against predeclared measures,
+classify the learning, and update the narrowest durable source.
+
+_Why it serves the approach:_ Each completed project leaves the next project with better
+judgment rather than more undifferentiated context.
+
+## Brand
+
+**One-liner:** Compound Marketing turns every marketing project into better work now and
+a better system for the next one.
+
+**Key message:** Preserve the decisions that mattered, put them to work, and use the
+result to improve the next run.

craft/editing/references/compound-from-edits.md

@@ -31,6 +31,9 @@ Human edits are evidence about the system, not cleanup to be absorbed silently.
 8. Update the narrowest durable source only when the edit passes the promotion rule.
    Otherwise keep the learning with the project.
 9. Treat the human-edited version as canonical for that project.
+10. When the project has a Compound Marketing run record, record the correction class,
+    affected decision IDs, durable destination or project-only status, and human review
+    time before closeout.
 
 ## Protect canonical language
 
@@ -66,3 +69,4 @@ Return:
 - Why each improved or redirected the work.
 - The durable rule or file to change.
 - Any project-specific context to keep local.
+- The run-record fields changed, when Compound Marketing is active.

docs/cases/compound-marketing-brand-book.md

@@ -0,0 +1,43 @@
+# Compound Marketing Case: Every Brand Book
+
+## Why this case exists
+
+The final Brand Book pass is the clearest observed example of human corrections becoming
+durable strategy. It is a retrospective baseline, not proof that Compound Marketing has
+already improved a comparable second run.
+
+## What happened
+
+The working deck began close to complete, but several important layers had been collapsed
+or rewritten during revision. Douglas repeatedly restored strategic distinctions:
+
+- Every is a media and technology company; the Frontier Team is the applied AI lab.
+- Learning is the customer promise, not another business unit.
+- Ideas, software, and education form the subscription system; community connects it.
+- Customer, brand, and commercial roles must remain distinct across audiences.
+- Aaker work, protected tonalities, and exact ruled language must survive compression.
+- Layout pressure is not a reason to invent more strategic categories.
+
+Those decisions now live in Every's positioning, audience, voice, and editing canon.
+
+## What it established
+
+- Human edits can reveal category, business-model, and framework errors rather than mere
+  copy preferences.
+- One approved decision must propagate across the complete artifact and every dependent
+  source.
+- The system needs a decision record before drafting so strategy does not disappear into
+  line edits.
+
+## Evidence limits
+
+The run did not record review minutes prospectively, did not begin with a governed
+decision record, and did not proceed into market execution. It establishes a real
+baseline and a set of durable corrections. It cannot support a compounding claim by
+itself.
+
+## Next comparable run
+
+The next company or product brand-strategy project should open a prospective run record,
+inherit the ruled distinctions, and measure review burden, substantial corrections, and
+accepted inherited decisions.

docs/cases/compound-marketing-thesis-gtm.md

@@ -0,0 +1,42 @@
+# Compound Marketing Case: Thesis GTM
+
+## Why this case exists
+
+The Thesis: 2027 planning process is the clearest observed GTM baseline. It shows how a
+real launch improved after strategy, source authority, protected language, and open
+questions were made explicit.
+
+## What happened
+
+The first workflow scored poorly in retrospective review because it:
+
+- Allowed logistics and metrics to crowd out the strategy.
+- Used unsupported dates and people-to-topic connections.
+- Replaced user-selected campaign language.
+- Turned a social brief into an operating manual.
+- Treated open pricing and programming questions as decided.
+
+The accepted work restored the After Automation story, the primary application action,
+and the Provoke and Promote campaign jobs. It separated the master GTM plan from the
+social brief, preserved open questions, and added source and scoped-edit safeguards.
+
+## What it established
+
+- Strategy and audience behavior must precede the calendar.
+- A campaign needs a mechanism and causal launch backbone.
+- Current facts, protected language, and open questions require different handling.
+- Scoped revisions should preserve untouched sections exactly.
+
+These rules now live in `marketing/gtm` and its regression cases.
+
+## Evidence limits
+
+The baseline was reconstructed after the fact. Review time was not measured
+prospectively, and the recorded score is partly human judgment rather than a reproducible
+end-to-end comparison. This is a real operational case, but not yet compounding proof.
+
+## Next comparable run
+
+The next substantial GTM project should open a prospective run record before drafting,
+inherit the Thesis-derived rules, and measure review minutes, first-pass acceptance,
+critical defects, and accepted inherited decisions.