Aegium
Our journey · recorded in Git

We didn't start
with the answer.

We started with a dashboard, kept the parts that survived the evidence, and rebuilt the rest. This is how Aegium became a market-neutral flow desk with its proof running live.

Follow the build ↓See the live proof
01
27 Feb
aegium-dashboard

the first recorded system

02
06 Mar
aegium-dashboard-clean

a clean production snapshot

03
15 Mar
dashboard-new

a modular rewrite explored

04
15 Mar → now
dashboard

the platform that earned its shape

4
repositories

one continuous product lineage

1,500+
recorded commits

as of July 2026

2
full rewrites

because the first shape was not sacred

1
public record

the result now has to prove itself live

The honest version

Not a victory lap.
A record of iteration.

Commit volume is not the same as progress. The proof is in the decisions: what we repaired, what we threw away, and what we refused to promote before it earned the right. These are the moments that changed the product.

27 February 2026
The first recorded snapshot

It began as a way to see the market clearly.

The earliest Git snapshot was already a working three-part system: a live data ingest, a FastAPI service, and a Next.js dashboard. It tracked price, flow, open interest, funding and market structure across perpetual futures.

The idea was simple: stop scanning disconnected charts and put the market's moving parts into one calm, usable view. The first commit captured 2,412 lines. It was useful — and nowhere near finished.

selected history
5e3560dsnapshot: server state
094a0ecrelax HTF blocks and cache best trades snapshot
40cd69eprevent infinite warmup on slow requests
Private repositories · excerpts shown for chronology
What changedThe first product was not a strategy. It was an instrument for asking better questions.
1–15 March 2026
Reliability before ambition

We rebuilt the foundations while the data kept moving.

Higher timeframes stalled. Cache hydration raced the interface. Charts could show stale windows. Instead of hiding those cracks, we worked through them: cache-first reads, shared Redis state, safer backfills, clearer freshness, tighter security and hundreds of small loading-path repairs.

Then we tried two cleaner architectures. One became a production snapshot; another became a modular FastAPI and HTMX rewrite. Neither was treated as destiny. The useful ideas survived and the product moved forward in the current repository.

selected history
f6a461bdashboard consistency refactor: phases 1 through 7
221218badd Redis shared SWR cache
52f2274initial commit: Aegium dashboard v2
Private repositories · excerpts shown for chronology
What changedA rewrite is not progress by itself. The version users can trust is the version that wins.
15 March–29 April 2026
From dashboard to platform

The screen became a research and execution system.

Research runs, outcome labels and multi-timeframe backtests arrived first. Then grid search gave way to Bayesian optimization, replay charts, train/test splits, realistic fees and stricter promotion gates. We kept connecting the thing being tested to the thing being shown live.

The platform expanded around it: authentication, plans, multi-user accounts, live executors, portfolio views, mobile layouts, notifications, public product pages and operational controls. Every new surface created another place where assumptions had to match reality.

selected history
b8505a5add MTF backtest engine
31f3629add Bayesian auto-optimizer
a631f1aadd live execution layer
ef9cc53add complete public-facing marketing layer
Private repositories · excerpts shown for chronology
What changedA backtest, a live engine and a dashboard are one product only when they tell the same truth.
May–14 June 2026
Reality over theory

We stopped asking what looked good and asked what survived.

Live replay gates, cost audits, shadow observers and human-review dossiers turned research into a process rather than a promising chart. A dedicated Research Lab tested broader families, expanded the dataset, rejected weak ideas and recorded why they failed.

On 12 June, a subtle timestamp-keying flaw was found: the backtester could see one bar ahead. We fixed it, invalidated the affected evidence and re-ran the work on leak-free data. That correction mattered more than preserving any attractive number.

What survived was not another directional signal. It was a diversified cross-sectional approach: rank the market, go long the strong, short the weak, and keep the book neutral to the broad move.

selected history
0703f65gate optimizer promotion on live replay edge
10cd86badd no-lookahead leakage audit
1fc27a6eliminate lookahead via end-keyed timestamps
c2df978multi-book paper trading and universe SSOT
Private repositories · excerpts shown for chronology
What changedFinding a flaw is not a failure of the process. Refusing to correct it would be.
15 June–July 2026
The product earns its shape
still building

We removed the noise and put the evidence in public.

The dashboard was realigned around the validated stack: Flow Board, Market State, Rotation, the market-neutral book and a Track Record that separates a blue backtest claim from green live proof. Pricing and onboarding followed the product that actually existed — not the one we had imagined months earlier.

Aegium went public on 19 June. Since then, the work has continued in the open: shadow engines compete on common windows, the liquid universe maintains itself, rejected research stays rejected, and execution fidelity is measured across venues instead of assumed.

The story is no longer about how much was built. It is about how much survived contact with data, markets and time.

selected history
8603b2crealign dashboard to the validated stack
cfbbc3cgo live — public launch
cdbb46fstaggered shadow research engine
24f073aself-maintaining liquidity-driven universe
Private repositories · excerpts shown for chronology
What changedThe edge is never finished. The bar for changing it only gets higher.
The invisible work

What didn't ship
shaped what did.

A serious research product is partly defined by what it says no to. The graveyard is not hidden; it is part of the method.

Rejected

Ideas that only worked in hindsight

Leakage audits and holdouts turn attractive curves into evidence—or into lessons.

Removed

Systems the evidence outgrew

Legacy signal, weights and paper-engine paths were stripped when the validated stack replaced them.

Rebuilt

Interfaces that hid uncertainty

Freshness, coverage, drawdowns and live-vs-backtest gaps became visible instead of being smoothed away.

The next commit

The story is still running.

The repositories are private. The outcome is not. The live paper book, its positions, drawdowns and competing shadow engines are where the work has to keep proving itself.

View the Track Record ↗Explore Aegium
History reviewed across all reachable branches · snapshot 12 July 2026