Loading…
2026-05-05 · First-principles prelim · Decide

ATLAS — first-principles prelim. Approve, modify, or reject.

After re-reading the full thread plus ATLAS / Argus / FX-alpha / crypto-alpha / Providence memories, the prelim concludes the load-bearing architectural decision should be reversed: ATLAS is not the full pipeline; it is the verification + orchestration node within a federation that already exists. The alpha portfolios are the production system. ATLAS becomes the rigour layer.

Below: the load-bearing decision (section 1), then 4 tripwire-critical tasks (section 2), 5 post-tripwire tasks (section 3), 3 long-horizon tasks (section 4), 1 tripwire reframe (section 5), 1 CLAUDE.md alignment (section 6), and 3 uncertainty resolutions (section 7).

Section 1 is load-bearing. Every subsequent decision is conditional on it. If you reject section 1, the rest collapses back to the previous Path B plan.

Mode Deep prelim Tripwire 2026-06-22 · ~50 days Current verified edges 0 Decisions 18 across 7 sections

TL;DR

Eight months of ATLAS build has produced zero verified edges. RBF was overfit (Sharpe 3.18 → OOS 0.60). Phase D returned 0/1525 passed. SMA-4 ports show 0.31-0.46 vs upstream 2.42-3.19.

The alpha portfolios already work. btc_alpha_portfolio, eth_alpha_portfolio, btc_alpha_spot etc. run live with real money. They contain multiple Sharpe-1.0+ strategies with documented track records.

The reframe. Stop trying to make ATLAS reproduce what the alpha portfolios already do. Make ATLAS the verification spine that decides what graduates to those production deployments. ATLAS does less, more honestly.

Critical path to tripwire. Pick one upstream strategy with high trade count (a3_btc_lead_lag, n=160+, Sharpe 1.46), build a parity harness around it, verify ATLAS reproduces upstream metrics within tolerance, publish methodology under Providence brand. ~3-5 days. Meets the tripwire.

Critical context — why the reframe

FX alpha already proved direction prediction doesn't work

8 experiments on major FX pairs across H1/H4/D timeframes with TA + ML + macro + sentiment. 0 passed Sharpe > 0.5 gate. ATLAS Phase 3 then ported FX hypothesis checkers anyway. The same approach is being re-tested with predictably similar results — the 7-gaps-why-strategies-fail memory and the matrix scanner inherit this dead-end.

Real edges live in the alpha portfolios — not ATLAS

BTC SMA-4: Sharpe 3.19, return 1693-2843%. ETH SMA-4: Sharpe 2.42, 271%. BTC lead-lag: Sharpe 1.46, 462%, 160+ trades. These are documented track records on funded multi-exchange deployments. ATLAS's parallel re-implementation has been chasing them for weeks with worse fidelity.

The strongest version of the reframe (iconoclast lens)

ATLAS doesn't run trades at all — not paper, not live. ATLAS verifies, decides, and dispatches verdicts to the alpha portfolios. The existing risk_manager + circuit_breaker in each portfolio handles execution. Removes the entire scanner / paper_trade / live_execution stack from ATLAS. Smaller, clearer system. This is what section 1's reversal commits to.

Tripwire pressure

~50
Days to 2026-06-22
0
Verified edges currently
3-5
Days to confirm-existing-edge via M1
1
Operator (FIFO, single thread)

If the tripwire is missed, the documented intervention is "pivot harder toward Path 3 (artefact-only, lab-hire endgame)" — the income-via-alpha timeline slips. The fastest credible path to ONE verified edge is to confirm an existing one from the alpha portfolios, not to discover a new one in ATLAS.

Decisions

Each decision below: YES approves, MODIFY with notes, NO rejects. Auto-saves. Submit at the bottom.

0
Yes
0
No
0
Modify
18
Pending
0% reviewed

Submit final decisions?

Once submitted, Claude reads your responses and starts executing what you approved. You can still come back and edit; rerunning manually after edits is on you.