01
What is the number?
The composite score — ranked cross-sectionally against the universe on the day. A rating means "top of this field today", not "cleared a hard-coded threshold that drifts with the market."
composite
+0.74
Catalyst-driven equity research
Ratings that showtheir working.
Sentinel rates any listed company on estimate revisions, earnings surprise, momentum and classified news catalysts — then tells you exactly how much real data went into each one. Every position it would open must clear five gates, and every refusal is written down.
Research and automation tooling. Not investment advice.
sentinel · ideas
| Symbol | Rating | Score | Confidence | Shari'ah |
|---|---|---|---|---|
LLY Eli Lilly | Strong Buy | +0.88 | high · 94% | Compliant |
APH Amphenol | Buy | +0.74 | high · 92% | Compliant |
BRK.B Berkshire Hathaway | Buy | +0.62 | high · 85% | Review |
NVDA NVIDIA | Hold | +0.51 | high · 96% | Review |
KO Coca-Cola | Hold | +0.18 | high · 91% | Compliant |
TSLA Tesla | Sell | -0.42 | medium · 78% | Not compliant |
33
catalyst categories
6
data providers
5
gates before any order
39/39
engine checks passing
§02 · Evidence
Four questions, answered with evidence.
Most screeners give you a number. Sentinel gives you the number, the factors underneath it, how much real data each factor actually had, and the headlines it read to get there. A score built on two factors is not presented as though it were built on five.
02
What factors sit underneath it?
Six factors, with their weights visible. Estimate revisions carry the most weight because a revision is new information about the future. The other factors restate the past.
weights (%)
32 / 20 / 18 / 15 / 8 / 7
03
How much real data did each factor actually have?
Coverage is the share of total factor weight that had real data behind it. A missing factor is never filled with zero — it is renormalised out of the weighting and subtracted from confidence.
coverage
92%
04
What headlines did it read to get there?
Classified into 33 typed catalyst categories, ranked by impact not recency. Each item shows source, catalyst type, and relevance. Headlines about a competitor are scored down and discarded.
lead catalyst
FDA approval · +0.18
01Ideas
The ranked list. Buy / Sell / Hold, confidence, coverage, Shari’ah, lead catalyst, price, 21-day drift.
02News
Classified by catalyst, ranked by impact not recency. Competitor headlines scored down and discarded.
03Movers
Biggest gainers and losers, each with the catalyst that explains it where one can be found.
04Calendar
Upcoming earnings dates — the events the model explicitly does not claim to predict.
05Portfolio
Connect Trading 212. Every holding re-rated on the same engine, including the Shari’ah screen.
06AI Advisor
A chat panel that answers questions about the ratings using the app’s own data, and states its confidence.
§03 · How it rates
The factor table, in the open.
Revisions carry the most weight because a revision is new information about the future, whereas most other factors restate the past. Scores are ranked cross-sectionally against the universe on the day — a rating means “top of this field today” rather than “cleared a hard-coded threshold that drifts with the market.”
Factor
Weight
Bar
Estimate revisions
32%
Which way analysts are moving their numbers. New information about the future.
Momentum
20%
12-month return excluding the last month, with a short-term reversal guard.
Earnings surprise
18%
History of beating or missing consensus.
Catalyst score
15%
Classified, relevance-weighted, recency-decayed news.
Quality
8%
Balance-sheet and profitability checks.
Sentiment
7%
Directional tone across coverage.
Total · 100%
From score to rank
1
Strong Buy
Top 5%
2
Buy
Next 15%
3
Hold
Middle 60%
4
Sell
Next 15%
5
Strong Sell
Bottom 5%
Levels
Every rating carries a reference price, a stop at 2× ATR and a target at 3× ATR. These are what the automation sizes positions against.
Coverage & confidence
Absent data reduces confidence. It is never filled in with zero.
When coverage is thin, the composite is shrunk toward zero in proportion to coverage, and factor agreement is damped until enough factors have actually spoken. The confidence label says so — on screen, every time.
Example coverage
0%
Rank
02
Buy · High confidence
§04 · Automation
Five gates and a paper account.
Off by default. Every candidate must clear all five gates — signal, eligibility, freshness, risk, account. Any refusal is written to the decision log with the gate that stopped it and the reason.
01
Signal
A rating must clear the signal gate before anything else. Rank, confidence, an absolute composite floor, a data-coverage floor, and — if you require one — a specific catalyst type.
- Rank threshold
- Confidence floor
- Absolute composite floor
- Data-coverage floor
- Optional required catalyst
02
Eligibility
The name itself must be tradeable. Shari’ah verdict, minimum price, minimum average volume, an earnings blackout, and not already held in the book.
- Shari’ah verdict
- Minimum price
- Minimum average volume
- Earnings blackout
- Not already held
03
Freshness
Refuses a rating whose price has drifted away from the live market. Sizing and stops come from the scan price, so a stale rating produces a wrongly sized position.
This is not hypothetical: a scan once captured Amphenol at an unadjusted split price of 160.08 while the market was at 82.14, and the resulting long carried a stop above its entry.
- Price-drift check
- Scan price vs live market
- Stale rating refusal
- Sizing comes from scan price
04
Risk
Position size is derived from your per-trade risk budget and the actual ATR stop distance — so a volatile name gets a smaller position for the same account exposure. Then capped by max position size, sector exposure, and cash buffer.
- Per-trade risk budget
- ATR stop distance
- Max position size
- Sector exposure cap
- Cash buffer
05
Account
The final gate. Master switch, kill switch, daily loss limit, market hours, and — for live — explicit arming that expires on its own.
- Master switch
- Kill switch
- Daily loss limit
- Market hours
- Explicit live arming
Presets
Three presets configure all twenty rules at once, followed by a plain-English summary of exactly what the configuration will do. The individual gates sit behind an “Advanced” disclosure — because a novice changing a number they do not understand is the single most dangerous thing that can happen here.
Preset 01
Conservative
BuysRank 1, high confidence, ≥80% coverage
Risk / trade0.5%
Max positions5
Preset 02
Balanced
BuysRank 1, high confidence, ≥70% coverage
Risk / trade1.0%
Max positions8
Preset 03
Aggressive
BuysRank 1–2, medium confidence, ≥60% coverage
Risk / trade1.5%
Max positions12
Three execution modes
Practice, demo, live — and live has to be armed on purpose. Practice fills model slippage and FX against you, so a paper track record is pessimistic rather than flattering.
Practice
None moneyLocal simulation only. No broker, no network.
8 bps slippage per side + 15 bps FX fee. Zero commission. Pessimistic by design.
Demo
None moneyTrading 212 practice account. Real API, same code path as live.
Live and demo share one code path; only the hostname differs.
Live
Real moneyReal orders in your real Trading 212 account.
Armed separately from being selected. Arming expires on its own.
Verification
The engine ships with an end-to-end wiring test that exercises every gate, the sizing maths, and a full paper round trip.
39 / 39checks passing
§06 · What it does not do
The most credible section on this page.
A research tool that only lists its strengths is selling something. These limits are stated in the app too — put on the page, not buried in a footer.
Validation · momentum factor
When a factor fails, the engine reports it.
The information coefficient is measured against a shuffled-noise control on every run. On recent data the momentum factor scored −0.1883 against a 0.1668 noise control — i.e. currently inverted. Reported in the app, not hidden. The weights have not been retuned to flatter one period.
Information coefficient
−0.25
0.00
+0.25
Momentum factor
−0.1883
Inverted
Noise control
0.1668
Shuffled baseline
01
It has not been shown to predict returns.
The engine measures its own information coefficient against a shuffled control on every run. On recent data the momentum factor scored −0.1883 against a 0.1668 noise control — i.e. currently inverted. That is reported in the app rather than hidden, and the weights have deliberately not been retuned to flatter one period.
02
Catalyst coverage depends on your keys.
On the free fallback source alone, news coverage is thin and many names show no catalyst at all. The coverage figure is always on screen so you can see which situation you are in.
03
The Shari’ah screen is a first pass.
It is not a fatwa and does not replace your own scholar. Where the answer is scholar-dependent the verdict says so.
04
Credentials are encrypted, not end-to-end.
The encryption key is held by the server, which is what allows keys to roam between devices. An operator with both database and key access could read them. This is stated in the app too.
05
No system is unhackable.
The security work is real and documented — authentication, encrypted credentials, row-level security, Content-Security-Policy with a per-request nonce, rate limiting, security headers, and an OWASP Top 10 review. But the claim that any system is unhackable is not one anyone can honestly make.
§07 · Begin
Start on the practice account.
Nothing reaches a broker until you connect one, and nothing reaches real money until you arm it deliberately. Run it for a month first and see whether you believe it.
Yahoo Finance works with no key at all, so the engine is useful before you configure anything.
- No key required
- Yahoo Finance
- Optional keys
- Finnhub · FMP · Polygon · Tiingo · Alpha Vantage
- Scan sizes
- Quick 40 · Standard 120 · Deep 250 · Full
- Broker
- Trading 212
PracticeSafe
Simulated on your machine. No broker, no network, no money. Fills assume 8 bps of slippage against you plus a 15 bps FX fee, so nothing here flatters the strategy.
DemoNo money
Your Trading 212 practice account. Real API, practice money — the same code path live uses, so it is the honest rehearsal.
LiveReal money
Real orders in your real account. Selecting it is not enough — it must also be armed, and the arming expires on its own.