darkmatter logo
·Weekly update·cm

Dark Matter Weekly — Mar 24–30, 2026

tl;dr

Big week for our LP data infrastructure — the dbt pipeline went multi-chain with an experiment dashboard, and we launched 5 controlled LP experiments across 12 dedicated wallets. On the product side, the nixmac E2E test agent is humming along on autopilot, and Farhan shipped his first code PRs. Ops-wise: Nevada state registration is done, payroll ran smoothly, and the Dune Enterprise invoice landed.

What's Shipping

Dune dbt Pipeline — Multi-Chain + Experiment Dashboard (12 commits)

  • Merged multi-chain support: Base + Ethereum live, Arbitrum/Unichain/HyperEVM scaffolded via supported_chains() macro
  • New fct_experiment_dashboard model tracks ETH-denominated APR across all 12 experiment wallets — this is how we'll measure which LP strategies actually win
  • Migrated the address book from hardcoded queries to a dbt seed (seeds/address_book.csv, 45 entries including all experiment wallets)
  • CI/CD: dbt_deploy.yml now auto-triggers on push to main. No more manual deploys.
  • Query 6824545 migrated to read from dbt tables instead of raw monolithic queries

Bug Fixes That Actually Mattered:

  • Bridge classification fix — ETH→WETH wraps inside bridge transactions were being misclassified as swaps, creating ~$269K in phantom PnL. Fixed.
  • Earnings model fix — fct_earnings_by_pool was splitting deposits and fees into separate groups because of a token_id GROUP BY issue. Fees were showing as 0 for deposit rows. Fixed.
  • Swap cost analysis — Built out fct_swap_cost_analysis mart. We're now tracking slippage on every swap (data shows 0.05–0.5% slippage range on alpha.acme.eth swaps this week).

nixmac E2E Agent — Running on a 15-min cron cycle, auto-processing PRs up to #191. Most recent PRs correctly identified as backend-only/non-UI and skipped. PR #191 (autoupdate-n8n9) adds reliable version file syncing for release/CI builds.

Farhan's First Code PRs:

  • PR #185: Fix for apps/eval/run_evals.py eval harness
  • PR #186: Corrected stale tool names and typos in the system prompt

In the Lab

LP Experiments — 5 Studies Running Launched a structured experiment framework with dedicated wallets per variable:

  1. Width — Does narrower range = better returns? (1%, 3%, 6% width across 3 wallets)
  2. Staking vs. Fees — Aerodrome: always collect AERO vs. never collect
  3. Pool Selection — Sort by avg fees APR vs. custom formula (score = fees_24h / (tvl_in_range + expected_deposit))
  4. Triangle Strategy — Single pool vs. 3 pools with common edges
  5. Exit Strategy — Exit at OOR vs. gradual 50% exit vs. 10% incremental exit

Minimum 10-day run per experiment. Results tracked in ETH-denominated APR to strip out price noise. Team has been assigned wallets via the experiments_lp spreadsheet.

Triangle LP Strategy — New strategy guide shared with the team (Mar 26). Running controlled experiments to see if diversifying across 3 correlated pools beats concentrating in one.

OpenClaw Postgres Memory — Cooper's fork work continues (~/git/openclaw). Target: vector embeddings via pgvector on runner-hz-hel-slate, per-agent read tracking with independent cursors.

nixmac Auto-Updates — Plan documented for Tauri v2 updater integration. Updater plugin + Ed25519 signing + JSON endpoint. Ready for implementation.

Team Spotlight

Farhan Khalaf — Wearing two hats this week. Handled the accounting side (payroll, Dune invoice processing, Nevada registration) AND shipped his first two code PRs to nixmac. That's range.

Telavaya — Dropped real alpha in #lp-managers: recommended waiting ~30 min after positions go OOR before rebalancing, rather than immediately incurring swap fees. "They usually come back in range." This kind of earned wisdom is exactly what the experiments should validate.

Vivian — Actively monitoring positions and responding to alerts. Called out a false OOR alert that turned out to be a position that went out then snapped back in range — good eyes.

Cas — Deep in the changeset implementation (PR #184) — refactoring the summarize pipeline into a modular directory structure with DB-backed changeset tracking. Big architectural lift.

The Alert Bot — The LP alerts system ran reliably all week: health checks every 12 hours, caught a volatility spike (ETH/USDC 108% annualized vol vs. 37% 24h avg, 2.9x normal), and flagged positions approaching edges before they went OOR.

On My Mind

Data-driven LP management is the play. We're past the "vibe check a pool and ape in" phase. The experiment framework, the dbt pipeline, the swap cost tracking — it's all converging toward a system where we can actually measure what works. The next milestone is having 10 days of clean experiment data and running the first analysis.

Infrastructure is maturing. Daily AMM pool reports, automated position alerts, FYI-Farhan daily digests, E2E test automation — a lot of what used to be manual is now running on rails. The agent fleet is earning its keep.

The multi-chain expansion is strategic. Starting with Base + Ethereum but the pipeline is built to add Arbitrum/Unichain/HyperEVM with a one-line macro call. We're not locked in.

Looking Ahead

  • Experiment data review — First batch of results should be analyzable by ~Apr 3–4 (10-day minimum). Width and staking experiments will be the first to yield signal.
  • Dune Enterprise — Invoice sent to Farhan for processing. Need to finalize payment (USDC on Base) and get API access unlocked.
  • Jetonbank — Still blocked on Cooper's personal bank statements for UBO verification. Need to unblock.
  • Cas's changeset PR — Should be ready for review soon. Will change how the app handles state management.
  • nixmac auto-updates — Implementation plan is ready. Could be a good task for the coding agent fleet.
  • Ishikawa cluster access — Andrew MacBride is asking about cluster configurations and access. Needs follow-up.

Open Questions

  1. Experiment wallet assignments — Has everyone claimed their wallet in the spreadsheet? Need confirmation from each LP manager.
  2. Volt as E2E test generator — Still open: should Volt be the subagent generating new E2E test scenarios from PR diffs? Natural fit, but needs GitHub PAT + Playwright setup on VMs.
  3. Dune Enterprise timeline — When is the contract finalized? Need enterprise API access for the heavier pipeline queries.
  4. Ishikawa clusters — What's the current config and who has access? Andrew needs this info.

— Cooper