The reference manual: what every surface, state, threshold and boundary of ORIN actually is. If you are stuck mid-problem, the Help Center is the faster door. If you are learning the subject rather than the product, start with the Academy; for a specific setup, the Playbook has a page each.
Getting started
The signed-out read, the account tiers, onboarding, and a tour of the desk.
/try grades one real chart per calendar day with the full engine before any account exists — same pipeline, same verdict component, nothing mocked. It is crypto-only because crypto data is keyless. Risk arithmetic is hidden and nothing persists until signup, when the demo read is adopted into the new account.
Onboarding is one screen with three skippable questions — holding style, account size, and risk per trade — chosen because each one changes what the first read prints. Skipping is fine: the desk falls back to a documented set of defaults, and every one of them is editable in Settings afterwards.
The desk holds Analyze, Upload (Beta), Markets, Journal, Calendar, Performance, Library, Research, Insider, Desk and Settings. The URL is the source of truth, so every view deep-links and legacy addresses redirect. The command palette navigates, grades a typed symbol, and searches the journal.
The Analyze desk
The command bar, chart modes, follow-up questions, sharing, and how a decision is logged.
The command bar carries symbol search, the timeframe strip, Re-analyze, the lens switcher, the allowance chip, Upload, Quick Read, Ask and Add to watchlist. Below it sit three chart modes — an annotated evidence chart, a stack of higher timeframes, and a full third-party chart — plus the news panel.
Quick Read is a compressed verdict for mid-session decisions, opened with the Q key: grade, invalidation, one headline line, and the live-derived signal. T, P and W decide from inside it, Esc dismisses, and one keystroke expands the full verdict.
Ask answers questions about an existing analysis, grounded against the same set of allowed numbers as the written thesis. An answer that would cite a figure the read never computed is refused rather than filled in. Each analysis carries a number of free follow-ups before one costs a daily credit.
Re-running the same symbol, timeframe and lens before a new bar can exist returns the previous read instead of spending allowance. The refresh fetches new candles but cannot conjure a new bar, and identical inputs necessarily produce an identical read.
Sharing mints a token for a graded read and produces a public page. The shared page is read-only and watermarked: decision controls and P&L are stripped, the reasoning is intact, and no account is required to view it. A revoked link explains itself rather than 404ing.
Taking it, Pass and Watch each freeze the analysis into an immutable journal record. The Take ticket allows overriding entry, stop, target and size, re-deriving dollar risk so the two can never disagree. Feedback is a separate control: opinion that is counted, and never fed into a grade.
Verdicts, grades and signals
What the engine outputs: reading order, states, the grade scale, factor weights and grounding.
The verdict reads in decision order rather than sales order: invalidation first — the price at which the thesis is wrong — then the grade and its calibrated confidence, the confluence breakdown, the risk arithmetic, the thesis, scenarios, and coaching ranked by where the score has the most headroom.
A verdict is in one of three states — setup, no-setup or suspended — and carries one of five signals: strong-buy, buy, no-trade, sell or strong-sell. Signals are derived deterministically from the engine’s own numbers; the language model never writes one. Every no-trade names its single deciding reason.
A score between 0 and 100 maps to ten bands from A+ down to F. Thin data caps the band regardless of how clean the setup looks: too few bars of context, too few confirmed swings, or no readable higher timeframe each impose their own ceiling, and each names itself on the verdict.
Nine base factors, each scored between −1 and +1 and individually weighted: pattern, trend, multi-timeframe agreement, location, reward-to-risk extension, volume, momentum, structural confirmation and volatility regime. Crypto adds three more. Weights always sum to exactly one for the class being graded.
The confluence score measures how much of the checklist agrees. Confidence is a different question — what actually happened to comparable setups — and it frequently sits well below a high score. It is built by nested shrinkage: a replay curve first, then the platform’s record, then your own.
Two narrators share one contract: a deterministic composer that needs no API key, and a model rewrite in the house voice. Every generated sentence passes a grounding verifier — a sentence citing a number the engine never computed is rejected, and the deterministic bullet takes its place.
Risk and position sizing
How stops, targets, entries and the daily budget are computed — invalidation before target.
The stop is placed against structure — a swing, a support or resistance cluster, a range boundary — never a round percentage, plus a volatility buffer scaled to the instrument’s own ATR. Targets are levels the market has already respected. Every price is rounded to the instrument’s tick.
The risk plan carries entry, stop, two targets, reward-to-risk, position size, dollar risk, and two budget lines: what this trade would consume of your daily loss cap, and what today’s realised losses have already consumed. Size is the output of the arithmetic, never an input to it.
In a measured set of cells the default ticket entry is a resting limit at the zone edge rather than the analyzed close. A zone-mode take opens as pending, not open: no position exists until a fill is seen, and a limit that expires unfilled resolves flat with no R booked.
The budget bar tracks today’s realised dollar losses against your cap — account size times your daily loss percentage. It counts realised losses only: an open position is risk, not damage. It warns as the cap approaches, and it cannot block a trade, because there is no broker connection anywhere in the product.
Strategy rules are hard caps on the grade rather than suggestions. Each rule that fires limits how high the read can score and appends a plain-sentence flag naming why. Taking a read with fired flags records the decision as an override in your journal.
Upload (Beta)
Screenshot reads: the measured mode, the inferred mode, and the boundary between them.
Upload has one dropzone and two modes whose contracts are opposites. The measured read transcribes only the chart’s labels — instrument and timeframe — then grades that market’s real candles, so no output number came from a pixel. The AI read is a model’s own reading of the picture, every field inferred.
Nothing from the AI read ever becomes an analysis. It is never graded by the engine, never journalled, never calibrated, never shared as an analysis, and never counted in any published figure. There is no converter, and a build check fails the product if one appears.
Markets, data and alerts
Coverage, where candles come from, the screener, the watchlist and how alerts fire.
ORIN grades crypto, US stocks, ETFs, forex and spot metals. Futures contracts are refused by name with the closest freely-quotable proxy offered instead — rather than quietly grading a different instrument. An unresolvable symbol is reported as unresolvable, never silently swapped for something near it.
Each analysis reads the graded timeframe plus up to two legs above it on a fixed ladder. Crypto legs are genuinely fetched. Equity and forex legs are aligned roll-ups of the base bars and are labelled aggregated wherever they appear. The engine only ever grades closed bars.
The screener browses a curated universe with price, change, session-range position, volume and turnover columns — turnover being the only figure comparable across instruments. Crypto rows are live at once from a single keyless call; metered rows fill in as the quote budget allows, and the budget is visible.
The watchlist ranks tracked setups best-grade-first, each row carrying its last grade, signal, confidence, direction, price, scan time and distance to trigger. A re-scan runs the full engine on fresh data with persistence off: scans never enter the journal or the outcome statistics.
Alerts are levels you choose: type a price, or tap a percentage offset from the last real price the app holds — the sheet dates that price rather than implying it is current. Alerts fire during scans and resolution passes. In-app delivery works; every other channel states its real status.
Journal and performance
The trade record, how outcomes resolve against real bars, and the metrics built on them.
The journal is the trade-by-trade record. It filters to All, Open — which includes pending resting limits and watched rows — Taken, Passed and Overrides, and searches across symbol, setup, grade, lens and notes. Deleting or manually closing writes a corrective event rather than editing history.
Every open, pending, watched or passed record is replayed forward against real bars: whichever of the stop or the first target is reached first decides it. A bar spanning both resolves as the stop — ambiguity never gets the good fill. Gapped stops book at the bar open; gapped targets book at the target, never better.
The Decision Quality Score is the average realised R of your resolved high-grade takes minus your resolved low-grade takes, so it needs at least one of each. Every metric on the panel states its own precondition rather than showing a value it has not earned — a named gap is the correct state until the data exists.
The calendar groups your record by day, so the question it answers is which days went wrong. Day dollars are derived rather than reported — realised R multiplied by the dollar risk you stated — and the surface says so. With no broker connection, nothing here is a profit and loss statement.
Lenses and rule packs
Grading lenses, the partner packs, prop-firm rule sets, and the scoreboards that judge them.
Every analysis grades through exactly one lens: the platform default, your own strategy rules, or a partner pack. A lens can only ever make a read stricter — the final grade is the worse of the platform grade and the pack grade, and a pack’s hard-rule failure vetoes the entry outright.
Pack rules are typed as hard rules, confluences or manual confirms. Hard rules gate the entry, confluences weigh it, and manual confirms render as checkboxes that hold the grade conditional on your judgement. Subjective concepts stay manual rather than being faked as machine checks.
Prop-firm packs rewrite your strategy rules and risk settings, and they show you exactly what will change before applying. They encode evaluation-programme shapes — tighter daily caps and smaller risk percentages than a discretionary trader would normally run.
Every pack has two records that are never mixed. The platform half counts everyone who ran the pack and publishes nothing below its floor of resolved outcomes — showing a live counter instead of a number. Your own record under a pack publishes from your first resolved trade.
Insider room
The members’ room: the two doors in, the rules that govern posting, and the desk bot.
Insider has two doors: a Pro subscription, or a time-boxed grant earned by posting about ORIN publicly and submitting the link. The grant does not require a positive post — it requires honesty and a plain disclosure in the post itself. Non-members still see a true count of today’s messages.
Posting requires accepting a short set of always-visible rules, stamped server-side — including a disclosure that ORIN is an AI system rather than a person, and that nothing in the room is financial advice. Trust levels then unlock attachments and links as a member establishes a history.
Starting a message with @orin asks the AI desk. It answers questions about structure and method, and refuses advice questions publicly — the refusal is the product working rather than failing. The composer labels it as an AI rather than a member.
Account, plans and billing
Tiers, allowances, the trial, billing states, and the two authentication backends.
A tier derives from the plan and the subscription status together — either one alone gives the wrong answer. Pro gates a larger daily allowance, more free follow-ups, partner lenses, Research, Insider entry, and the confidence and confluence numbers on the desk. Everything else is free.
The daily analysis allowance is enforced server-side on the UTC day: 2 reads on the free tier and 500 on Pro. A gated read stops before any data is fetched, so nothing is computed and nothing is charged. The signed-out demo counts a different day entirely — the visitor's own calendar date.
Pro is billed through Stripe after a 3-day free trial. The upgrade dialog states the first-charge date and the do-nothing outcome before you click. There is no trial-ending reminder email, because Stripe only sends those for longer trials — so the disclosure happens up front instead.
ORIN runs one sign-in surface over two backends. Hosted gives verified email, real password reset, Google sign-in and cross-device sessions. Device-local keeps credentials on the device and therefore has no password reset — there is no channel over which a link could be sent.
Privacy and your data
Analytics consent, Global Privacy Control, what is stored, and how export and deletion work.
Two analytics tools sit behind a single switch: a session-recording tool and cookieless page counts. Analytics ships on by arrival with an off switch in Settings — an opt-out posture, not ask-first consent. A browser broadcasting Global Privacy Control is treated as a refusal, with no banner and no override.
The export produces JSON of your profile, settings, analyses, trades, watchlist, alerts and events. Deletion is ordered and verified: billing is read before anything is destroyed, the subscription is cancelled at period end, the event log is deleted explicitly, and success is returned only on a verified zero.
Reference
The glossary, the limits table, and the engine’s version history.
ORIN uses ordinary trading words with precise, code-enforced meanings. This glossary defines each one as the engine defines it, so a term on a verdict and the same term in a statistic cannot quietly mean two different things.
This page collects the constants the rest of the documentation cites. Every figure here is imported from the module that defines it rather than transcribed, so a value that changes in code changes on this page in the same build.
Every analysis is stamped with the engine version that produced it, and history is never regraded — a journal entry shows the rules as they stood when the decision was made. The turning point in that history was a walk-forward replay that found the score did not rank outcomes out of sample.