engraphis-dashboard to see it live
-**Grounded, not guessed.** Memory with receipts. Local by default. [Explore the proof gallery](https://github.com/Coding-Dev-Tools/engraphis/tree/main/docs/advertising) or [read the campaign guide](https://github.com/Coding-Dev-Tools/engraphis/blob/main/docs/advertising/campaign.md).
+**Grounded, not guessed.** Memory with receipts. Local by default.
---
diff --git a/docs/advertising/campaign.md b/docs/advertising/campaign.md
deleted file mode 100644
index 9c0015d1..00000000
--- a/docs/advertising/campaign.md
+++ /dev/null
@@ -1,186 +0,0 @@
-# Engraphis proof-first campaign
-
-Engraphis should be marketed as a memory system that can show its work.
-
-> Grounded, not guessed. Memory with receipts. Local by default.
-
-This guide turns existing demos, benchmark artifacts, and product surfaces into a repeatable campaign. It does not introduce new product claims.
-
-## Message architecture
-
-### The flagship promise
-
-**Engraphis carries project knowledge forward without treating old information as current truth.**
-
-Use this promise when the audience needs the whole story: local storage, scoped recall, temporal validity, provenance, and context packing.
-
-### Four proof stories
-
-| Story | Hook | Existing proof | Primary action |
-| --- | --- | --- | --- |
-| Local | No account. No API key. Still remembers. | Dashboard quickstart and local-first README copy | Install free |
-| Grounded | No support, no answer. | Grounded recall and evidence-backed behavior fixture | Inspect grounded recall |
-| Temporal | Facts change. History stays visible. | Continuity demo, invalidation, supersession, and timeline | Watch the continuity demo |
-| Efficient | Fewer tokens, better evidence. | Registered `offline-chunking` and `offline-performance` fixtures | Read the benchmark |
-
-Keep one story per asset. Do not combine every feature into one graphic.
-
-## Four-week distribution sequence
-
-### Week 1: Local memory
-
-Lead with the relief of not starting from zero and the control of keeping memory on the machine.
-
-- Short screen clip: open the local dashboard and show the graph, provenance, and receipts.
-- Static post: “Your agent’s memory can stay on your machine.”
-- Setup post: `pip install "engraphis[mcp]"`, `engraphis-init`, and the Smart MCP command.
-- README placement: link the proof gallery directly below the existing knowledge graph image.
-
-### Week 2: Grounded recall
-
-Lead with a cited answer and an explicit abstain. The contrast is more memorable than another generic retrieval diagram.
-
-- Carousel: “Cited answer” beside “No support, no answer.”
-- Short clip: ask one supported question, then one off-topic question.
-- Technical post: explain why the grounded gate uses absolute support instead of the normalized recall score.
-- CTA: “Try grounded recall.”
-
-### Week 3: Temporal memory
-
-Lead with one changing repository decision. Show the old fact, the new fact, and the reason the old record remains queryable.
-
-- Timeline graphic: old validity window closes, new fact becomes current.
-- Continuity reel: use the existing 56-second demo artifact.
-- Blog post: “Engraphis does not just remember. It remembers what changed.”
-- CTA: “Inspect the why and timeline.”
-
-### Week 4: Context economy
-
-Lead with the smallest useful evidence, not a vague claim about speed or cost.
-
-- Stat card: `740.3 -> 214.3` tokens with Recall@5 `1.000` in the registered fixture.
-- Evidence card: `162.2 -> 42.4` tokens to the smallest evidence-holding memory.
-- Technical post: clarify that the compact payload proxy is separate from chunking and must not be added to it.
-- CTA: “Read the benchmark definitions.”
-
-## Reusable post hooks
-
-1. Stop replaying the whole chat.
-2. Memory with receipts.
-3. No support, no answer.
-4. Facts change. History stays visible.
-5. Some answers live in the graph, not the note.
-6. Your agent does not need more history. It needs the right evidence.
-7. Local memory should not require a trust fall.
-8. Find the symbol. Explain the decision.
-9. Nine Smart MCP tools first. Discover advanced actions only when needed.
-10. Bring your memory stack. Publish one immutable run.
-
-## Public benchmark challenge
-
-### Campaign idea
-
-Invite memory-tool builders to run a fixed, public-safe fixture and publish the artifact digest, command, configuration, and result summary.
-
-### Public copy
-
-> Bring your memory stack. Publish one immutable run.
->
-> Use the locked fixture, keep the comparison boundary explicit, and share the result without raw questions, answers, prompts, or private records.
-
-### Launch requirements
-
-- Anchor every published number to an evidence ID in `BENCHMARKS.md`.
-- Use the public runbook in `docs/PUBLIC_BENCHMARK_RUNBOOK.md` as the execution contract.
-- Publish the whole-input and source-file digests required by the runbook.
-- Keep raw questions, answers, prompts, context, and per-record content fingerprints out of public artifacts.
-- Present the challenge as a reproducibility standard, not as a self-selected victory lap.
-
-### Embed-ready result format
-
-```text
-Memory benchmark
-Stack: Local-first memory for agents
-Memory with receipts. Keep project history local, retrieve evidence, and carry the right context into every session.
- -Every headline below maps to a registered fixture, an existing demo, or a visible local product surface.
-Structure-aware chunks returned the relevant passage instead of the whole document, with Recall@5 held at 1.000 in the registered fixture.
- Open context efficiency evidence -Use a living fact, not a feature list, to show what makes Engraphis different from a chat replay or a flat note store.
-A correction closes the old validity window. It does not erase the record.
-Show one subject moving from an earlier decision to a current one. The audience sees the old fact, the new fact, and the evidence chain that explains the transition.
- Open the continuity demo -Make the repository graph the bridge between a symbol, the related path, and the memory that explains why the system looks that way.
- Open the graph quickstart -The code graph is best-effort. The explanation remains memory-backed and inspectable.
-Keep the first action concrete. Different audiences can enter through the local dashboard, Smart MCP, or the Python library.
-
pip install "engraphis[mcp]"
-engraphis-init
-codex mcp add engraphis -- engraphis-mcpGive coding agents scoped recall, durable memory, and compact context through the zero-configuration Smart gateway.
- Connect the gateway -Inspect memories, graph relationships, provenance, receipts, and temporal changes locally.
- Open the dashboard path -Start with the offline library when you need a memory engine without a hosted dependency.
- Use the library -A four-week sequence turns continuity, grounded recall, temporal memory, and context economy into one repeatable story.
- Open campaign guide -Trust is part of the product story. These boundaries keep the gallery useful to technical readers.
-The 740.3 to 214.3 comparison is retrieved memory content in a deterministic fixture. It is not provider billing, latency, or a universal customer outcome.
-The continuity demo is generated from a real in-memory MemoryService run. Any sample fallback is labeled and must not be presented as customer evidence.
-Symbol and file search, memory-backed explanation, path, impact, and export are distinct actions. The gallery keeps those terms separate.
-