Test Twin Forge live, rewrite build pages, fix citations and fictitious names

.gitignore

@@ -8,3 +8,6 @@ docs/.vitepress/cache/
 # OS / editor
 .DS_Store
 Thumbs.db
+
+# generated by scripts/pack-downloads.mjs
+docs/public/downloads/

Allfiles/digital-twin-starter/AGENTS.md

@@ -6,7 +6,7 @@ This repo is the Advanced Microsoft hackathon starter for a **digital twin**: a
 
 The twin is not a chatbot persona. It is a small set of markdown files that capture how a person decides, writes, overcommits, protects boundaries, and resolves tradeoffs. Agents must read the spec, cite the governing rule, and admit gaps instead of inventing preferences.
 
-Use the synthetic example name **Marcus Webb** in examples. Do not introduce real personal data.
+Use the synthetic example name **Avery Washington** in examples. Do not introduce real personal data.
 
 ## File layout
 
@@ -17,7 +17,7 @@ Use the synthetic example name **Marcus Webb** in examples. Do not introduce rea
 - `mcp_server.py` — MCP thin tools plus TODO thick-tool stubs.
 - `twinlib.py` — Copilot CLI large-prompt helper.
 - `run.ps1` — local council runner.
-- `test/` — sealed forced-choice comparison harness.
+- `test/` — the sealed-answer harness. `take_test.py` records the human's gut answers to 15 dilemmas BEFORE the build; `compare.py` runs the twin over the same 15 at the end and diffs them, naming the soul.md field behind each mismatch. No score is produced, by design.
 
 ## Council output contracts
 

Allfiles/digital-twin-starter/README.md

@@ -4,6 +4,28 @@ Build a portable **digital twin**: a compact spec for how someone works, plus ag
 
 This starter gives you the plumbing. You write the spec and the council agents.
 
+## The test harness — start here
+
+**Before you write anything**, seal your answers:
+
+```powershell
+python test/take_test.py
+```
+
+Fifteen forced-choice work dilemmas. Answer fast — what you *actually* did last time, not what you'd like to have done. It flags anything that took over 25 seconds, because a deliberated answer is an aspirational one.
+
+That writes `test/sealed-answers.md`. Don't open it again until the end.
+
+**At the end of the session**, have the twin answer the same fifteen cold:
+
+```powershell
+python test/compare.py
+```
+
+You get a line-by-line diff, and every mismatch names the `soul.md` field that dilemma was probing — so a miss points straight at the rule to fix.
+
+There is deliberately **no score and no leaderboard**. Mismatches are the output. The useful ones are where the twin picked the sensible, well-adjusted answer and you wouldn't have.
+
 ## Quickstart
 
 ```powershell

Allfiles/digital-twin-starter/digital-twin.code-workspace

@@ -0,0 +1,14 @@
+{
+  "folders": [{ "path": "." }],
+  "settings": {
+    "chat.agent.enabled": true,
+    "chat.useCustomizationsInParentRepositories": true,
+    "files.exclude": {
+      "**/__pycache__": true,
+      "**/.twin-scratch": true
+    }
+  },
+  "extensions": {
+    "recommendations": ["github.copilot", "github.copilot-chat", "ms-python.python"]
+  }
+}

Allfiles/digital-twin-starter/digital-twin/SKILL.md

@@ -1,7 +1,7 @@
 ---
 name: digital-twin
 description: |
-  Use for trigger phrases like "what would Marcus do", "draft in my voice",
+  Use for trigger phrases like "what would Avery do", "draft in my voice",
   "run the digital twin", "decide like me", or "check my boundary rules".
 ---
 

Allfiles/digital-twin-starter/digital-twin/references/soul.template.md

@@ -4,7 +4,7 @@
 
 ## Identity
 
-<!-- Role, scope, current season of work, what success means. Example subject for workshops: Marcus Webb, synthetic marketing manager. -->
+<!-- Role, scope, current season of work, what success means. Example subject for workshops: Avery Washington, synthetic marketing manager. -->
 
 ## Priority stack
 

Allfiles/persona-pack/README.md

@@ -1,26 +1,26 @@
 ---
-title: Marcus Webb Persona Pack
+title: Avery Washington Persona Pack
 ---
 
-# Marcus Webb Persona Pack
+# Avery Washington Persona Pack
 
 Entirely synthetic. Any resemblance to real people is coincidental.
 
 This pack gives hackathon teams a realistic substitute data set for building a digital twin when they cannot or do not want to use their own Microsoft 365 data.
 
-## Who Marcus is
+## Who Avery is
 
-Marcus Webb is a Marketing Manager at Northwind Traders, a fictional B2B software company. He owns product marketing for the Atlas platform launch while also juggling demand generation, sales enablement, an external agency relationship, and urgent requests from Sales.
+Avery Washington is a Marketing Manager at Northwind Traders, a fictional B2B software company. They own product marketing for the Lantern platform launch while also juggling demand generation, sales enablement, an external agency relationship, and urgent requests from Sales.
 
 ## How to use this pack
 
-Open these files as if they were Marcus's exported work context:
+Open these files as if they were Avery's exported work context:
 
-1. Start with `context/goals-and-role.md` to see what Marcus says about his role and priorities.
-2. Read `inbox/sent-mail.md` as the voice corpus. This is the best source for how Marcus writes.
-3. Read `calendar/30-days.md` to see how his time is actually spent.
+1. Start with `context/goals-and-role.md` to see what Avery says about their role and priorities.
+2. Read `inbox/sent-mail.md` as the voice corpus. This is the best source for how Avery writes.
+3. Read `calendar/30-days.md` to see how their time is actually spent.
 4. Use `context/org-chart.md` to understand stakeholder dynamics.
-5. Use `inbox/received-mail.md` to understand the pressure around him.
+5. Use `inbox/received-mail.md` to understand the pressure they're under.
 6. Only open `calendar/summary-stats.md` if you are facilitating or stuck; it gives away the patterns.
 
-What Marcus says and what his data shows deliberately do not always match. Finding those gaps is the point.
+What Avery says and what their data shows deliberately do not always match. Finding those gaps is the point.