📦 Products Dashboard

🔓 The Unlock: Charlie's insight — "Linear tickets are like a title and one sentence. That's not enough. You need to inflate them into actual work." Akira's job is to take sparse input and produce fully-specified, implementation-ready tickets for Jarvis.

Dedicated Machines Required

Need separate infrastructure for clean IP separation from Kindo. Cannot use Kindo resources for Akira/Jarvis work.

Status: Must resolve before Basis alpha engagement.
Options: Personal machines, cloud instances, or Tony's existing hardware.

Files Downloaded & Analyzed:
• portfolio-manager-with-context.zip ✅
• portfolio-manager-standalone.zip ✅
• reklaim-estimator-vercel.zip ✅
• reklaim-estimator-lovable.zip ✅

What Greg Built from Joana's Source:
• 3 OpenClaw agents (Coordinator, Estimator, Executor)
• Ownership/empowerment model integrated
• Multi-agent handoff protocol

Key Insight Preserved: Story Points × 2 = raw hours → × 1.7 drag factor. 33/33/34 Promise/Stretch/Buffer model.

Executive Prioritizer works off the same engine as current estimator. Estimates level of effort vs value, then prioritizes based on that.

Leverages data from Greg's tracking of how long it takes user to develop idea, deploy it, and begin earning revenue.

Feature: Visualize user's AI chats into visual dashboards. Dashboard tracks AI chats and progress of discussions on a timeline.

Today's session WAS the Exec.AI workflow in action. Here's what we did:

The High Performer Hypnagogic Workflow:

Session Results:
• 41 items captured → 7 remaining in Stream (17 routed, 17 done/purged)
• 11 items in To-Do list
• 3 emails sent (Joana, Victor, Alex)
• 4 new dashboards built (Stream, Standup, Products, To-Do refresh)
• Akira PMO fully analyzed (4 zip files downloaded, architecture documented)
• Products dashboard created with Exec.AI + Akira sections

Key Insight: The "AI sells fastest when it doesn't require humans to change" principle in action — Tony's existing workflow (handwritten cards, WhatsApp photos) stayed the same. Greg adapts to capture and process.

No actual RAG/embeddings needed. The "RAG" is hardcoded context strings (KINDO_SECTIONS_SUMMARY, CYCLE_PLANNING_RULES) injected into prompts. OpenClaw can do this natively with skills and workspace files.

Estimator core logic: Story Points × 2 = raw hours → × 1.7 drag factor = drag hours. 33/33/34 rule: Promise (first 33%), Stretch (next 33%), Buffer (34%).

Portfolio Manager: Pure calculation engine for capacity (engineers × 250 days × 6 hrs ÷ drag), payroll coverage, and milestone-based cash flow.

🔓 The Unlock: PROFILER → PRIORITIZER → EXECUTOR → VERIFIER

• Jarvis (Charlie) executes perfectly but doesn't decide what matters
• Akira prioritizes but needs human-supplied priorities
• The Gap: Automating the Profiler function (turning customer calls → structured priorities)
• Charlie currently IS the profiler — scales linearly with calls he can attend

Akira = Profiler + Prioritizer. Jarvis = Executor. Together = The Unlock.

Location: /Users/greg/.openclaw/workspace/

• SOUL.md — Who Greg is (personality, style, boundaries)
• USER.md — Who you are (name, preferences, context)
• AGENTS.md — How Greg behaves (memory protocol, heartbeats, safety)
• TOOLS.md — Local notes (contacts, device names)

How to modify: Direct edit in any text editor — changes take effect next turn. Or ask Greg: "Update SOUL.md to be more concise"

Location: /Users/greg/.openclaw/workspace/memory/

• YYYY-MM-DD.md — Daily logs (what happened each day)
• MEMORY.md — Curated long-term insights

How Greg uses them: Every session loads today + yesterday's logs. Main sessions also load MEMORY.md. memory_search does semantic search across all memory files.

You can: Read/edit directly, ask Greg to update — they persist across sessions/restarts.

Best uses for your workflow:
• Akira documentation — Estimator/Portfolio Manager specs
• Chat exports — ChatGPT/Claude as text files
• Research papers — Gartner PDFs for summarization
• Reference docs — Anything for Greg's context

How: Use gog skill to search/read Drive files, or tell Greg "read the doc at [link]"

Recommendation: Create a Greg-Context folder in Drive for docs you want Greg to reference.

Current: Joana shared folder at Akira PMO: Rebuilding Estimator & Portfolio Manager

Recommendation for Akira:
• Short-term: Use sub-agents within Greg for Estimator/Portfolio Manager tasks
• Long-term: If Akira becomes a product for others, spin out to its own instance

📅 Reminder set: Revisit sub-agents decision in 24 hours (Feb 7, 2026 ~9:22 AM)

🎯 Core Principle: The 80/20 rule — 80% of human-perceived value comes from 20% of the features. This depends on judgment: skillfully assessing the situation and human preference.

💡 The Explicit Value Algorithm:

1. Human identifies HIGH PRIORITY features
2. Estimator calculates LEVEL OF EFFORT
3. Cross-reference: High Priority ∩ Low Effort = HIGHEST VALUE
4. Result: Deliver more perceived value in less time
This is also the core engine for Exec.AI Executive Prioritizer.

🔮 THE MISSING PIECE: To-Do Item #10 — "Explain Profiler & Multi-Agent Distribution Product Customizer"

This is where Tony will explain the Human Profiler Engine — the mechanism for assessing perceived value based on understanding the human. This completes the Pareto judgment loop.

Status: Waiting for Tony's explanation session

📝 Summary: The Estimator is really an effort estimator, not a value estimator. It assumes priority input from humans. The Profiler fills the gap by modeling human preference — turning "what does this person actually care about?" into actionable priority weights.

💡 Bottom Line: APP is built on verification, not trust. That's why it works better for AI agents than humans. It's one of the few human methodologies that translates because it doesn't rely on ego, peer pressure, or human judgment.

Contract System: Clear acceptance criteria with Promise/Stretch separation. Explicit "ACCEPT/MODIFY/DECLINE" gate before work starts. Defined success metrics upfront (41+ images = 50% coverage).

Two-Phase Protocol: Phase 0 (acceptance) prevented rushing into work. Gave opportunity to catch scope issues before burning cycles. Coordinator approval ("BEGIN WORK") created accountability checkpoint.

Technical Context: Contract included specific file paths, API details, naming conventions. Wger API integration delivered 24 valid images. The Promise deliverable (update exercises.json, update UI) was completed.

Frame Extraction Approach: Card photo cropping was fundamentally broken. ffmpeg grabbed random frames (carpet, wrong exercises). Only 1 of 14 extractions was correct (7% accuracy). Root cause: No validation step before marking as "done".

No Intermediate Checkpoints: Lost visibility during long-running work. Couldn't catch the frame extraction issue until final QA. No "show me 3 sample extractions before proceeding".

Session Label Confusion: Used both 10x-images-001 and 10x-images-002 without clear handoff. Hard to trace what happened in which session post-facto.

Stretch Became Mixed with Promise: Card cropping was listed as "Stretch" but became attempted as Promise. Should have stopped after Wger images + declared Stretch incomplete.

💡 Verdict: The contract system works — it forced clarity upfront and created accountability. The failure was in execution monitoring and validation gates. The Stretch item (card cropping) should never have been attempted without a proven approach.

Recommendation: For the next cycle, add explicit QA checkpoints at 25%, 50%, 75% completion. Stretch items require separate "STRETCH APPROVED" confirmation.