docs(donna): align plan with shipped product (dashboard live past MVP, in-family failover, working-today refresh)

donna/index.html

@@ -190,7 +190,7 @@
   <h2>NEWT's first AI employee.</h2>
   <p><b>Mission:</b> make agentic AI usable by real businesses by attaching a named, accountable human to every outcome.</p>
   <p><b>What it is:</b> Donna is an AI executive assistant personified on Dawn Gamble — a real senior EA of fifteen years at Fibernetics — and operated under her review. She runs calendars, inboxes, travel, meeting prep and follow-ups. Nothing she produces reaches another human until a person has read it. Donna ships under <b>NEWT</b>, the business division of Fibernetics, a Canadian carrier running business voice and network since 2011.</p>
-  <p><b>Stage:</b> live early beta. Two marketing sites are deployed with a working application pipeline; pricing is published; the first five supervised seats are open; the reviewer dashboard is spec'd from the operating desk Dawn already runs.</p>
+  <p><b>Stage:</b> live early beta. Two marketing sites are deployed with a working application pipeline; pricing is published; the first five supervised seats are open; the reviewer dashboard — <a href="https://donna-desk.mosaic.site">Donna Desk</a> — is live and built out well past MVP, encoding the operating desk Dawn already runs.</p>
   <div class="cards">
     <div class="card"><div class="lab">Origin</div><h4>Born from a working desk</h4><p>The blueprint is Abby — the agent Zoe built that automates a real controller's finance reporting, document pipelines and website upkeep, cutting hours of manual work to minutes. Donna generalises that pattern to the executive desk.</p></div>
     <div class="card"><div class="lab">Positioning</div><h4>AI-enhanced human</h4><p>Clients trust a person, not a model. The visible face is a trusted EA; the AI turbocharges her output. Marketing never says "replace your EA" — it says one assistant now covers three desks.</p></div>
@@ -224,12 +224,12 @@
   <h3>Architecture</h3>
   <ul>
     <li><b>Claude (Anthropic) + Mosaic</b> — frontier reasoning plus instant deploy of servers, sites and apps; the same stack shipped both beta sites and the live application pipeline in days, not months.</li>
-    <li><b>Model failover</b> — automatic fallback to a second provider or a local model, because an assistant that goes down with its vendor isn't an assistant.</li>
+    <li><b>Model failover</b> — automatic in-family fallback is live today (Opus &rarr; Sonnet on transient failure; a weaker model degrades polish, never safety — everything still passes review). Second-provider and local fallback land in the funded plan.</li>
     <li><b>Data residency</b> — customer content stays in the Fibernetics environment in Canada; scoped to connected mailboxes; not used to train general models; exported and removed on exit.</li>
     <li><b>No new portal for customers</b> — Donna works inside the email, calendar, phone and finance tools the customer already runs. The dashboard is for the reviewer.</li>
   </ul>
   <div class="callout"><div class="lab">Working today</div>
-    <p>Live: both product sites, the six-question qualification quiz, published pricing, and a verified application pipeline (form → database → notification email, tagged by seat). Also live: the <a href="https://donna-desk.mosaic.site">Desk dashboard MVP</a> &mdash; per-executive buckets, an AI meeting-notes pipeline (raw notes &rarr; summary &rarr; suggested tasks/events, all held in a review queue for explicit approval), AI contact extraction, and the beta-application pipeline surfaced as a workable CRM. Not yet live (and said so publicly): Finance, Support and Voice agents.</p>
+    <p>Live: both product sites, the six-question qualification quiz, published pricing, a verified application pipeline (form &rarr; database &rarr; notification email, tagged by seat) &mdash; and the <a href="https://donna-desk.mosaic.site">Desk dashboard</a>, well past MVP: risk-ordered review queue (high-stakes replies surface first, replies hard-locked to human review), AI triage of pasted or forwarded mail with voice-matched reply drafts and per-executive voice profiles, meeting-notes and contact extraction with source-quote receipts, follow-up chasing, a travel desk that assembles itineraries and preps re-books, expenses to claim-ready CSV, real calendar sync with protected focus blocks, the &ldquo;Donna noticed&rdquo; anticipation ledger, one-tap escalation to the principal, morning briefing emails, full activity record with compliance export, one-file data export, and universal ingest (Gmail/Outlook connectors, drag-anything drop, a watched desktop folder). Applications double as a workable CRM with Donna drafting the promised two-business-day reply for approval. Not yet live (and said so publicly): Finance, Support and Voice agents; deep OAuth write-back to Google/Microsoft calendars is in the funded plan.</p>
   </div>
 </section>
 
@@ -362,7 +362,7 @@
   <ul>
     <li><b>Cost structure:</b> reviewers (the dominant variable cost, ~50% of supervised-seat revenue at maturity), AI compute (~5–10%), platform engineering (fixed, small — the stack is deliberately lean), sales via existing channel (low CAC by design).</li>
     <li><b>Break-even (illustrative):</b> ~35–40 supervised seats carries the early team; the platform seat's 85% margin pulls blended margin up as mix shifts toward software — which is the whole strategic design.</li>
-    <li><b>Funding need:</b> <b>$1.5M CAD for 18 months</b> (internal venture allocation or seed): Desk dashboard build, 6 reviewers hired and trained, SOC 2 Type I, WhatsApp + voice channels, VAR partner program, and marketing that stays in the honesty register.</li>
+    <li><b>Funding need:</b> <b>$1.5M CAD for 18 months</b> (internal venture allocation or seed): Desk dashboard deepening (OAuth write-back, channels), 6 reviewers hired and trained, SOC 2 Type I, WhatsApp + voice channels, VAR partner program, and marketing that stays in the honesty register.</li>
     <li><b>Use of funds:</b> ~45% engineering, ~30% desk staffing &amp; training, ~15% compliance/security, ~10% GTM.</li>
   </ul>
 </section>
@@ -372,7 +372,7 @@
   <div class="kicker">13 · Roadmap</div>
   <h2>Three horizons.</h2>
   <div class="hz">
-    <div class="h"><div class="n">Horizon 1 · now → 6 mo</div><h4>Prove the desk</h4><p>Fill 5 beta seats from live intake. Ship the Desk dashboard MVP (buckets, review queue, insights). Convert beta → paid with an explicit yes. Hire reviewers 2–3. SOC 2 program start. VAR pilot with 2 resellers.</p></div>
+    <div class="h"><div class="n">Horizon 1 · now → 6 mo</div><h4>Prove the desk</h4><p>Fill 5 beta seats from live intake. Deepen the shipped Desk dashboard (Google/Microsoft OAuth write-back, iMessage/SMS channel). Convert beta → paid with an explicit yes. Hire reviewers 2–3. SOC 2 program start. VAR pilot with 2 resellers.</p></div>
     <div class="h"><div class="n">Horizon 2 · 6–18 mo</div><h4>Scale the two engines</h4><p>EA platform seat GA with self-serve pricing. FIN agent live (invoice chasing, expense coding — reviewed before anything posts). WhatsApp channel for international. 100+ supervised seats, 500+ platform seats.</p></div>
     <div class="h"><div class="n">Horizon 3 · 18 mo +</div><h4>Own the category</h4><p>VOX voice agent on NEWT's own telephony. Donna-certified reviewer program → a marketplace where any certified EA sells supervised capacity on our rails, and we take the platform cut. The trust layer for agentic work.</p></div>
   </div>
@@ -388,7 +388,7 @@
     <div class="r"><b>Suite copilots commoditise drafting</b><span class="mit">Let them — drafting was never the product. Copilots make our COGS cheaper and still leave the customer as their own reviewer at 11 p.m. Accountability plus service doesn't ship in a suite SKU.</span></div>
     <div class="r"><b>Services margin drags the multiple</b><span class="mit">Mix-shift is the plan: supervised seats are the wedge and the data engine; platform seats at ~85% margin become the majority of ARR by Year 3 (illustrative).</span></div>
     <div class="r"><b>No security certification yet</b><span class="mit">Stated plainly on the public site today; DPA before anything connects; scoped access, revoked on disconnect; SOC 2 Type I funded in the ask. Honesty now beats implication later.</span></div>
-    <div class="r"><b>Model/vendor dependency</b><span class="mit">Multi-model failover is in the architecture (Claude primary, second provider or local fallback), and the review layer means a weaker model degrades speed, not safety.</span></div>
+    <div class="r"><b>Model/vendor dependency</b><span class="mit">In-family failover is already live (Opus &rarr; Sonnet on transient failure); a second provider is in the funded plan — and the review layer means a weaker model degrades speed, not safety.</span></div>
     <div class="r"><b>A public error under a customer's name</b><span class="mit">Tier design makes it structurally hard (Tier 3 waits indefinitely); full action logs make any incident auditable; the correction becomes a product change the same week — and we say so.</span></div>
     <div class="r"><b>Beta converts poorly at $1,500</b><span class="mit">Price is published now specifically so conversion is tested honestly from day one; fallback positioning against $1,600–3,400 part-time VAs holds margin; platform seat catches the price-sensitive.</span></div>
   </div>
@@ -400,7 +400,7 @@
   <h2>Green-light the venture. The beta is already live.</h2>
   <ul>
     <li><b>Decision:</b> constitute Donna as a NEWT product line with <b>$1.5M CAD / 18-month</b> allocation and Dawn formally allocated to it.</li>
-    <li><b>Next 30 days:</b> fill the 5 beta seats from live intake (first target: the warm Delta Air lead); start the Desk dashboard build; begin reviewer #2 recruitment from the EA community funnel; kick off SOC 2 scoping.</li>
+    <li><b>Next 30 days:</b> fill the 5 beta seats from live intake (first target: the warm Delta Air lead); put the shipped Desk dashboard in Dawn's daily hands; begin reviewer #2 recruitment from the EA community funnel; kick off SOC 2 scoping.</li>
     <li><b>Success gate for the next tranche:</b> 5 seats live with retention through day 30, review SLA held, and a signed VAR pilot — evidence, not projections.</li>
   </ul>
   <div class="callout"><div class="lab">The one-sentence case</div>

donna/summary.html

@@ -188,7 +188,7 @@
 
   <div class="ask">
     <div class="lab">The ask</div>
-    <p><b>Constitute Donna as a NEWT product line: $1.5M CAD over 18 months, Dawn formally allocated.</b> Next 30 days: fill the five beta seats from live intake, start the Desk dashboard, recruit reviewer #2. Gate for the next tranche: five seats retained through day 30 with the review SLA held &mdash; evidence, not projections.</p>
+    <p><b>Constitute Donna as a NEWT product line: $1.5M CAD over 18 months, Dawn formally allocated.</b> Next 30 days: fill the five beta seats from live intake, put the shipped Desk dashboard in Dawn's daily hands, recruit reviewer #2. Gate for the next tranche: five seats retained through day 30 with the review SLA held &mdash; evidence, not projections.</p>
   </div>
 
   <div class="foot">