docs: "what next" strategy deep dives + synthesis

[r3dbars]

Why

A focused "what should we do next" session. We looked at the app, then ran three grounded background deep dives — product moat, market position, and codebase health — and wrote them up alongside a capstone synthesis so the direction is durable and reviewable, not just chat scrollback.

Product Impact

• Affects: docs only
• Lane: agent workflow
• Why this matters: the three lenses converge on one answer (lean into the agent-memory loop), and capturing that with its evidence and sequencing makes the next planning step concrete instead of vibes.

What changed

• docs/strategy/agent-loop-activation-2026-06-14.md — product/moat deep dive. The captures→agent→answer loop is the moat but is shallow today: lexical FTS5 only (no semantic), recap returns raw transcript lines, and the summarizer's structured Decisions/Action-Items/Open-Questions are extracted but never indexed — so the moat inverts as the library grows.
• docs/strategy/growth-positioning-2026-06-14.md — market deep dive. The defensible position is "memory for your AI agent"; make the MCP/agent connection the hero feature; don't fight Granola on AI-notes; free + local + MIT forever, with a Team-tier (E2EE sync) as the only trust-safe monetization if ever needed.
• docs/strategy/codebase-architecture-health-2026-06-14.md — engineering deep dive. Overall health B; clean bones but four god objects are the velocity tax (worst: TranscriptedSettingsView.swift, 4503 LOC / 187 commits). Raw-swiftc build is worth keeping; Swift 6 strict-concurrency is deferred debt.
• docs/strategy/SYNTHESIS-2026-06-14.md — capstone. All three converge: the moat, the pitch, and the one big refactor are the same project. Includes the two key collisions (the summary paradox; the hero move trapped in the worst god object), a sequenced plan (Phase 0 quick wins → index the summaries + cross-meeting MCP tools → semantic search + Settings decomposition → launch), and explicit don'ts.

These are analysis artifacts under a new docs/strategy/ subtree. No product code, no contract/test changes, no runtime behavior change.

How I checked it

• scripts/dev/agent-preflight.sh — docs-only change; only suggested check is preflight, which is clean
• Selected checks from .agents/test-matrix.yml — n/a (no code, no agent-contract files touched)
• bash build.sh --no-open — n/a (docs only)
• bash run-tests.sh — n/a (docs only)
• Manual check: rendered Markdown + internal links between the four docs

Risk Review

• Privacy / local-first behavior reviewed — analysis only; no telemetry/payload changes
• Storage path or migration impact reviewed — none
• Public-facing copy stays concrete and matches current product scope — internal strategy docs, not user-facing copy
• Release/update impact reviewed — none
• No private transcripts, audio, tokens, personal paths, or customer data are included — synthesized analysis only; competitor figures are public

Notes

Recommendations only — no direction is locked by this PR. Open flags called out in the synthesis: live GitHub star/download metrics were not pulled (keep live-release truth separate before any launch-timing call), the agent-memory window is contested (Granola + Plaud shipped MCP servers in 2026), and model strategy (extraction model + bundled embedding model) is now on the critical path and worth its own follow-up deep dive.

Agent handoff

COORD_DONE: GREEN | https://github.com/r3dbars/transcripted/pull/<this> | added docs/strategy/ (3 deep dives + synthesis) | none | pick next step: model-strategy deep dive, spec the agent-memory build, or start Phase 0 quick wins | agent-preflight (clean), manual link/render check | start recap-summary-aware + formatter↔parser agreement test

🤖 Generated with Claude Code