Skip to content

SESSION-SUMMARY-2026-07-17.md

Latest commit

 

History

History
82 lines (72 loc) · 3.91 KB

SESSION-SUMMARY-2026-07-17.md

File metadata and controls

82 lines (72 loc) · 3.91 KB
Raw

Working Session Summary — July 16-17, 2026

Shipped v2.2.0: market-intelligence grew from a directory of investigation prompts into a complete intelligence system. This note is the handoff — what changed, the architecture that now exists, and what's open.

What shipped (see CHANGELOG.md for full detail)

v2.2.0 — The intelligence release:

  • Tradecraft shelf (market-intelligence/reference/): Dean's Competitive Research on Steroids compendium — eight intelligence-community collection disciplines mapped to PM artifacts — plus the EU/MENA regional source overlays. Doctrine, not prompts: no comment block, outside the validator and catalog (non-recursive glob skips subdirectories; future reference docs go here, not in SKIP_NAMES).
  • The collection floor: seven discipline sweeps (osint-, finint-, geoint-demoint-, techint-, humint-, sigint-, masint-collection-prompt.md). All seven emit the same fusion-ready Signal Inventory schema — that shared schema is the system's spine. Existing prompts (VoC miner, earnings refresh, pricing tracker, TAM/SAM/SOM...) are now the deep-dive layer each sweep hands off to.
  • The situation room: all-source-fusion-prompt.md — confidence stacking (1 discipline = watch, 2 = hypothesis, 3+ = actionable, conflict = dig), independence test (same-source signals collapse), ambition-vs-commitment corollary (announcements are intent until money/hiring/permits corroborate).
  • Ground truth: prompts/win-loss-analysis-prompt.md — debriefs -> per-deal drivers, evidence-ranked patterns, and the confirm/refute ledger that closes the loop on public-signal inference. Guide mode when no interviews exist yet.
  • The rhythm: loops/fusion-cadence-routine.md — weekly SIGINT, monthly OSINT+HUMINT, quarterly FININT+TECHINT+fusion, annual GEOINT/DEMOINT, event-driven MASINT. Tiered by source velocity: poll at the speed the source changes.
  • Also: workshops/product-sunset-workshop.md (checkpointed EOL plan) and prompts/incoming-request-breakdown.md (inbound triage), both landed earlier in the cycle.
  • Repo hygiene: main branch protection enabled (no force-push, no deletion; direct commits still allowed — no PR/status-check gates).

The architecture in one breath

Instantiate the engagement (six compendium variables) -> run the collection sweeps that matter -> fuse when 2+ disciplines hold signals -> actionable stories update artifacts (battle cards, roadmap bets, positioning, TAM) -> win/loss interviews confirm or refute -> the cadence routine keeps it all turning. The worked example in market-intelligence/README.md walks a launch caught before its press release.

How to work in this repo now

Unchanged from v2.0/v2.1 (see SESSION-SUMMARY-2026-07-03.md): AGENTS.md first, validator + catalog after every edit, new prompts declare an interaction mode. One addition: collection-floor prompts must keep the Signal Inventory schema identical across all seven — it is a contract, not a convention.

Open items (in priority order)

  1. Licensing — DONE same day as v2.3.0: MIT -> CC BY-NC-SA 4.0, LICENSING.md plain-English guide, commercial use requires Dean's written permission, MITRE/dataset provenance corrected, validator now blocks MIT regressions.
  2. Article URL: when "Prompts Aren't Dead" publishes, link it in loops/README.md and the root README.
  3. Agent Strategy Canvas Prompt 1 of 2 — Dean to provide (Prompt 2 is in prompts/).
  4. Engagement-brief prompt (candidate): fill the compendium's six instantiation variables once, carry them across sweeps and into the routine identity block.
  5. Regional overlay build-outs (on demand): APAC, LATAM, Sub-Saharan Africa, following the EU/MENA pattern.
  6. 79 grandfathered validator warnings — migrate-on-touch worklist, no mass rewrites.
  7. Literal Jinja2 rendering pipeline — future candidate for agent builders.