🚀 NVIDIA Inception Program Member — ReDevOps is a member of the NVIDIA Inception Program, supporting startups advancing AI and accelerated computing. Membership provides access to NVIDIA technology, technical resources, and the startup ecosystem. It does not imply product endorsement by NVIDIA.
Quantify compiles financial scenarios into transparent, versioned simulations. It records the methodology, evaluation protocol, calendar, tax and account treatment, cash flows, market-data policy, realized data, statistical assessment, and modelling limitations behind every result.
Not an AI portfolio optimizer, and not a backtesting dashboard. An event-driven financial simulation and research runtime built on versioned knowledge and execution artifacts.
857 tests passing · AGPL-3.0
A version-controlled knowledge system for investment research: what is claimed, what supports it, what assumptions it depends on, how it was tested, what changed, and whether the result may responsibly be published.
Impersonal by construction. No holdings, income, taxes or objectives; no personalized ranking.
Describe a financial workflow, confirm exactly what will be simulated, compare symmetric historical outcomes, save the plan privately, and track what happens forward. Quantify does not place orders or choose a course of action for you.
The boundary runs one way — a private plan may cite public research; no public artifact may ever cite a private one.
Every meaningful behaviour is declared and realized; every declaration names the mechanism that enforces or checks it.
Two failure modes are closed by construction: behaviour without declaration, and declaration without behaviour. The second is the harder one — a methodology that declares a rule its executor ignores looks exactly like one that enforces it.
This is not theoretical. Every release that made a hidden choice explicit immediately exposed a real defect:
| Made explicit | Found |
|---|---|
| Execution lag and costs | A reported 13.00% return was actually −2.83% |
| Trading calendar | 31.1% weekend padding inflating annualized figures |
| Rules naming their realization | Declared rules were inert — never executed |
| Tax treatment as a runtime | A Roth and a taxable account compared as identical |
pip install -r requirements-core.txt # or requirements-core.lock, pinned
python3 -m pytest tests/ -q
uvicorn src.api:app --reload # /ui library · /workspace private scenariosThe suite runs entirely on a committed synthetic price fixture — invented, deterministic, no credentials and no network. Nothing measured on it is a claim about any real security.
Licensed market data lives in a private, versioned S3 bucket, pinned by snapshot
id, object version and hash in data/manifests/. The bucket name is supplied by
.env.market-data (gitignored) so it stays out of a published repository; the
manifest carries everything needed to review what a result was computed against.
It is used by the opt-in tier:
source .env.market-data
pytest -m market_data_integration # requires credentials; fails, never skips
python3 scripts/provision_market_data.py --dry-run # plan a new snapshot uploadDescribing a scenario in prose uses a language model for stage 1 of the
compiler only — recognising phrases, never deciding anything. Set
ANTHROPIC_API_KEY to enable it, and QUANTIFY_PARSER_MODEL to choose a model.
Without a key the compiler uses its deterministic phrase rules, recognises less,
and asks more questions; it never guesses to fill the gap.
python3 scripts/run_methodology.py # execute a methodology under a protocol
python3 scripts/evaluate.py # assessment → policy → publication
python3 scripts/publish_run.py # record a run in the ledger
python3 scripts/assess.py # statistical assessment only| Architecture.md | Artifact model, runtime lifecycle, boundaries, comparability, regulatory posture |
| Features.md | What the system does, by surface |
| Implementation.md | Status, defect history, acceptance criteria, the architecture freeze and Closed Pilot v1 |
| Performance.md | Measured latency, HarnessBench, the Polars crossover |
| docs/errata/ | Published corrections |
The one place status is easy to misread: Investigation is implemented as a
knowledge artifact — persisted, queryable, rendered — and not yet implemented
as a durable unit of work. Lifecycle transitions, Discovery-driven creation and
conclusion-to-Finding routing are the remaining work. See
Implementation.md §8.
Source papers are artifacts under evidence/ rather than a bibliography, so a
claim can be supported, qualified or contradicted by them and the relationship is
queryable:
- López de Prado, M. (2016). "Building Diversified Portfolios That Outperform Out of Sample." Journal of Portfolio Management 42(4), 59–69. doi — hierarchical risk parity
- Jegadeesh, N. & Titman, S. (1993). Returns to buying winners and selling losers — cross-sectional momentum
- Bailey, D. & López de Prado, M. Deflated Sharpe ratio, probability of backtest overfitting, minimum track record length
- Vuletic, M. (2025). Multi-asset financial markets: mathematical modelling and data-driven approaches (Oxford DPhil thesis) — regime detection
- CFA Institute Research Foundation (2025). AI in Asset Management — monograph
- Guo, J. & Li, Y. (2026). Salience Theory and Risk Anomalies — SSRN
The cash-proxy finding shows how these are used.
finding/hrp-degenerates-to-cash-proxy@1 refutes one claim, qualifies
another — López de Prado's result holds for comparable-risk universes, not any
universe — invalidates the results of two methodology versions, motivates
a third, and introduces a constraint-precedence assumption. One conclusion,
five typed impacts.
The static dashboard builds to reports/ and deploys to Cloudflare Pages:
python3 -m src.history --start 2015-01-01 --end $(date +%Y-%m-%d) --step 5
python3 -m src.visualization.bokeh_app --output reports/regime_dashboard.html
export DOMAIN="quantify.club"
./deploy_cloudflare.shDNS is a CNAME from @ (or www) to the Pages project; Cloudflare provisions
TLS automatically. Daily rebuilds run from
.github/workflows/daily-deploy.yml and need
CLOUDFLARE_API_TOKEN, CLOUDFLARE_ACCOUNT_ID and CLOUDFLARE_EMAIL as
repository secrets.
Source-available, not open source. AGPL-3.0-or-later plus the Commons Clause condition.
AGPL §13 extends copyleft to network use: serving users over a network with this code obliges offering them the corresponding source of the combined work. The Commons Clause additionally removes the right to sell the software, including hosting it for a fee — which AGPL on its own permits.
See LICENSE.md and LICENSE-COMMONS-CLAUSE.