THARP Architecture Dossier (My Current Build)

Generated: 2026-03-04 (America/Denver) Owner: Patrick Smith Intent: Document my full architecture for THARP intelligence, end to end, as built today.

1) Architecture Scope

This dossier describes the working architecture I built to run THARP discovery intelligence into tomorrow-call execution artifacts.

Included architecture surfaces:

• Remote acquisition from Mac Studio over Tailscale
• Transcript and meeting-intel ingestion
• Operational log ingestion and redacted health extraction
• Normalized internal corpus in Wake
• Cross-session synthesis pipeline
• Decision artifact generation pipeline
• Tomorrow-call execution package

2) Topology (Systems + Locations)

2.1 Source Host

• macstudio (remote)
• Access path: Tailscale network
• Role: upstream source for call artifacts, prep packets, and gateway/runtime logs

2.2 Integration Host

• Local workspace at /Users/patricksmith/Work/wake
• Role: canonical integration and synthesis environment

2.3 Output/Distribution Layer

• Internal markdown package under docs/plans/tharp/internal-share/2026-03-04/
• HTML publish target on highline.work (private-share intent)

3) High-Level Architecture Graph

[macstudio over Tailscale]
  |- meeting artifacts (Whisper/OpenClaw/Read.ai)
  |- prep packets
  |- gateway + watchdog logs
  v
[Wake Ingestion Boundary]
  |- bounded 72h pull
  |- path normalization
  |- redaction boundary for ops logs
  v
[Canonical Corpus in wake/docs/plans/tharp]
  |- core/
  |- meeting-intel/
  |- ops/
  v
[Synthesis Engine]
  |- cross-call pattern extraction
  |- failure-mode mapping
  |- quantified burden capture
  |- build-wedge ranking
  v
[Decision Artifacts]
  |- PM live call sheet
  |- exec decision memo
  |- board brief + slides
  |- compiled master brief
  v
[Tomorrow Call Execution]

4) Ingestion Architecture (Remote -> Local)

4.1 Transport Layer

• Primary transport: Tailscale remote access to macstudio
• Pull strategy: time-bounded (72h window)
• Reliability fallback: find/grep fallback when remote rg unavailable

4.2 Acquisition Targets

• THARP planning docs and decision memos
• THARP prompts and prep artifacts
• Meeting intelligence bundles:
  • summaries
  • transcripts
  • one-pagers
  • coaching notes
• Ops logs:
  • OpenClaw gateway logs
  • Read.ai refresh/auth behavior
  • watchdog event tails

4.3 Bounded Capture Pattern

• Continuously written files (watchdog/runtime) captured via bounded tail snapshots
• Prevents long-running sync blockage from actively appended logs

5) Canonical Data Model In Wake

5.1 Root Package

/Users/patricksmith/Work/wake/docs/plans/tharp/internal-share/2026-03-04/

5.2 Package Domains

• core/
• meeting-intel/
• ops/

5.3 Domain Responsibilities

core/

Executable decision-support docs:

• 72h integration record
• PM live call sheet
• board brief
• exec slide draft
• deep-dive/tomorrow plan
• exec decision memo

meeting-intel/

Conversation intelligence layer:

• call-level transcripts/summaries
• one-pagers and prep docs
• consolidated cross-session synthesis
• coaching + raw rollups

ops/

Operational reliability evidence:

• redacted event logs
• generated ops status summary
• connector health signals

6) Processing Architecture

6.1 Stage A: Structural Normalization

• Normalize filenames and section boundaries
• Align artifacts into deterministic package layout
• Separate source classes (core vs meeting intel vs ops)

6.2 Stage B: Evidence Extraction

• Pull explicit facts from transcripts/debriefs
• Separate direct statements from inferred claims
• Preserve directional metrics with source context

6.3 Stage C: Pattern/Failure Mapping

• Build handoff map across design, ERP, scheduling, PM
• Identify repeated manual bridges and person-dependency points
• Record operational failure modes and detection points

6.4 Stage D: Build Sequencing Model

• Candidate wedge scoring dimensions:
  • pain severity
  • universality signal
  • 30-day buildability
  • measurement feasibility

6.5 Stage E: Artifact Compilation

• Convert synthesis to audience-specific outputs:
  • operator-level call execution sheets
  • leadership decision memo
  • board/exec framing docs
  • full compiled brief

7) Intelligence Architecture (What “Whisper + Tharp Calls” Becomes)

7.1 Capture Inputs

• Whisper transcripts
• Read.ai artifacts
• internal call debrief transcripts
• meeting prep artifacts

7.2 Intermediate Intelligence Objects

• call summaries (__summary.md)
• full transcripts (__transcript.md)
• prep objects (__prep.md)
• coaching objects (__coaching.md)
• consolidated synthesis objects (tharp-cross-synthesis.md, tharp-consolidated-internal.md)

7.3 Derived Decision Objects

• call runbook (live sheet)
• decision memo
• board brief/slides
• compiled full-architecture read

8) Tomorrow-Call Execution Architecture

8.1 Control Objective

Turn synthesis state into one explicit Build-1 decision gate.

8.2 Run-Time Call Stack

1. Baseline workflow capture
2. Failure/exception mapping
3. Data-trust boundary mapping
4. Change-order loop mapping
5. Pilot acceptance criteria capture
6. Owner + due-date locking

8.3 Exit Artifacts (Same-Day)

• build-1 recommendation revision
• evidence table update (facts/assumptions/gaps)
• data request tracker with owners
• updated pilot metric frame

9) Reliability Architecture

9.1 Observability Inputs

• readai-mcp.events.redacted.log
• wake-mcp.events.redacted.log
• apple-notes-mcp.events.redacted.log
• gws-mcp.events.redacted.log

9.2 Health Pattern Used

• event count
• warning/error count
• token refresh cadence validation

9.3 Runtime Risk Controls

• bounded log capture strategy
• redaction at package boundary
• source inventory persistence for replayability

10) Security / Share Architecture

10.1 Separation Boundary

• Internal working corpus stays in Wake workspace
• Share layer is rendered output only (no raw runtime directories)

10.2 Exclusions By Design

• direct secret/token material
• unrelated repositories/workspaces
• non-THARP raw operational archives

10.3 Publish Objective

• single-share page
• private/unlisted access path
• no broad directory exposure

11) Artifact Graph (Current)

Raw Calls + Logs
  -> meeting-intel/* (normalized)
  -> ops/* (redacted + summarized)
  -> core/* (decision docs)
  -> THARP_COMPILED_MASTER_BRIEF_2026-03-04_observations.md (this dossier)

Key architecture-linked files in this package:

• core/THARP_72H_MACSTUDIO_GATEWAY_INTEGRATION_2026-03-03.md
• core/THARP_CHRIS_PM_LIVE_CALL_SHEET_2026-03-04.md
• core/THARP_DEEP_DIVE_AND_TOMORROW_PLAN_2026-03-03.md
• core/THARP_EXEC_DECISION_MEMO_FOR_SANTIAGO_GARTH_2026-03-04.md
• meeting-intel/tharp-cross-synthesis.md
• meeting-intel/tharp-consolidated-internal.md
• meeting-intel/tharp-coaching-full.md
• meeting-intel/tharp-raw-full.md
• ops/OPS_STATUS_SUMMARY_2026-03-04.md

12) Current Architecture State (Today)

Built and Active

• Remote ingestion over Tailscale from Mac Studio
• 72h bounded integration pipeline
• canonical corpus structure (core/meeting-intel/ops)
• cross-session synthesis pipeline
• decision-asset generation pipeline
• ops health extraction and redacted summary layer

Built but Iterating

• evidence confidence labeling granularity
• repeatable wedge-scoring automation
• tighter PM-specific metric normalization

Pending in Immediate Next Cycle

• post-tomorrow-call corpus refresh
• build-1 gate lock with final acceptance criteria
• refreshed share artifact based on tomorrow evidence

13) Architecture-Only Bottom Line

This is the full architecture I have running today for THARP.

• It is source-integrated (macstudio -> wake)
• It is evidence-structured (core/meeting-intel/ops)
• It is decision-producing (call sheet + memo + board artifacts)
• It is built for tomorrow-call execution, not static reporting