ORIN
Sign in

Changelog

Everything that shipped, newest first. Changes to the scoring engine are marked Engine and kept separate from interface work — they change what a grade would have been, and that deserves its own label rather than a line in a list of improvements.

  1. NewOct 17, 2026

    Every read carries its plan again; Alpha 1.7, the latest model, and a desk that opens without an account

    Since 1.12.0 a chart with no tested setup got a chart read: no grade, no entry, stop or target. From today that gate is off. Every read the verdict and the higher-timeframe veto let through carries its grade and its plan, on every market — gold included. Nothing else about the engine changed: scores, grades and levels are computed exactly as before, and the evidence tier is still recorded on every read. The desk also names its models now. A free account grades on Alpha 1.7 — ORIN's default rulebook, with the read written by the built-in narrator. Pro grades on the latest model: the same engine, with the read written by the reasoning chain, the evidence tier shown beside the plan with its report card behind it, your own strategy rules and the prop-firm rule packs. The guru packs — six third-party methodologies offered as Pro lenses — are retired; reads graded through one keep the name they were written with, and their pages redirect. The desk itself opens without an account: a visitor grades 3 real charts a calendar day on Alpha 1.7, kept in their own browser until they sign up, when the reads move into the account. The Setups feed lists today's Grade A and B reads on default settings, with or without a tier. Reads saved as chart reads now show the plan they carried; nothing was regraded, and no rulebook was rewritten — rules an account set stay in Settings and apply on Pro.

    Verdict states and signals →
  2. EngineSep 14, 2026· v1.12.0

    Plans only for tested setups — every other chart gets a chart read

    Until now almost every chart ended in a grade and a trade plan. On the replayed record most of those plans had no skill: across the platform they played into their plans no more often than the same trade taken at a random moment, and the top grades were timed worse than the bottom ones. From this release ORIN gives an entry, stop and target only where a setup has beaten random entries in a registered test, and shows the evidence beside it as a tier — Beta, for a setup that passed on past data and is being tested again on new bars, or Proven, once it has passed that too. At launch everything is Beta: the failed breakout on crypto and forex 1H and 4H, and the top-ranked reads in crypto, forex and stocks under the weights fitted in 1.9.0. Every other chart — including every gold chart, where nothing has passed yet — gets a chart read: the trend, structure, location and stretch of the chart, with no plan, grade or confidence. Each tested setup has a report card that tracks how often it plays out on new bars against thirty random moments per result, fed by traders' reads and a daily read of a fixed list of markets; a card shows a rate only once it holds 100 results over 30 market days. The "trade with the higher-timeframe trend" rule is now off by default for new accounts and for the Setups feed, because in the markets graded by fitted weights it capped mostly the setups that played out best; if you have it on, it stays on, and a tested setup it capped offers a one-tap switch. The Setups feed now lists tested setups, whatever their grade from A+ to C.

    Effect on past gradesNothing has been regraded, and reads saved before this release keep the verdict they had. Scores and grades are computed exactly as before — replaying the full history produced identical grades, patterns, verdicts and plans, with the tier recorded beside them. On that replay about 29% of issued crypto setups, 28% of forex and 22% of stocks carry a tier; the rest become chart reads. Replay figures under the harness fill model, never expected returns.

    Report cards and tiers →
  3. NewSep 14, 2026

    Setups — today's Grade A and B reads, as the platform's own read of the chart

    A new desk page for subscribers, with the day's count in the rail for everyone. When a trader finishes an analysis, the server re-runs ORIN's standard read of that symbol and timeframe on default settings — no journal, rulebook, lens or account behind it — and lists the result if it grades A or B with a verdict of setup. The listing is the platform's read, not the trader's: it carries no user, account or analysis identifier and no request time, and the table that holds it has no column that could; a build gate runs the server engine on deliberately unusual personal settings and searches the projected row for every one of them. Ten people reading the same bar produce one listing. Each opens to the full read-only report — chart, thesis and its evidence, entry, invalidation, targets, confidence and confluence — worded as ORIN's, with a button to run the same chart under your own rules. A listing is live for five bars of its timeframe and then expires; listings are kept a week. A grade describes the chart, not the odds, and the page says so. Your activity prompting a re-read can be switched off in Settings → Data & privacy.

    How the feed works →
  4. EngineSep 14, 2026· v1.11.0

    A first beta setup: the failed breakout, on crypto and forex 1H and 4H

    Five setup definitions — a failed breakout, a sweep that reclaims, a rejection at a range edge, a second entry with the committed side, and a breakout the next bar confirms — were written down with the test they had to pass before any result was read. On the replayed record one passed: in crypto and forex, a failed breakout played into its plan more often than the same trade taken at a random moment of the same month (by 4.0 and 3.3 percentage points, judged against thirty random moments per setup). The with-trend definitions did not: second entries and confirmed breakouts played out no more often than chance, and in stocks the second entry did worse. The failed breakout now appears as a labelled beta — "Failed breakout (beta)" — on crypto and forex 1H and 4H charts, where it sets the read's direction and plan. The grade, your rules, the refusals and the higher-timeframe check apply to it as to any other read. It is not confirmed: the same test runs again on bars from after this release, no earlier than 7 December, and the beta is withdrawn if it fails there.

    Effect on past gradesNothing has been regraded. On the replayed record the beta changes the setup named on about 2 reads in 100 across the platform; about four in ten of the beta setups it issues are on charts that previously refused, and together they played into their plans 2.9 points more often than random moments. On charts where the engine already issued a setup at that bar, the beta read did no better than the read it replaced. Replay figures under the harness fill model, never expected returns.

    How we grade →
  5. EngineSep 14, 2026· v1.10.0

    Four detector faults fixed — and the breakout they were hiding points the wrong way

    A correctness release: no weight, threshold or ladder was refitted. Structure breaks are now ordered by when the close went through, not by when the pivot printed — the old order could name an older break as the latest one, and on a V-reversal call a break down while price had just closed back above structure. Order blocks and fair value gaps now share one rule: a zone is spent once an earlier bar returns to it, the bar being read is never the one that spends it, and a reaction names the zone price is closest to. Before, the bar trading into an order block marked it used, so order-block reactions almost never appeared, and gaps won every tie. The breakout check measured the range including the breakout bar itself, which no close can clear, so breakouts never fired. And a chart whose only pattern was an extended chase, or no pattern at all, is refused as no setup instead of being issued with a plan. Measured by replaying the engine against the same trade taken at a random moment of the same month: pooled accuracy is unchanged (−0.2 against −0.1 percentage points across about 146,000 setups; within noise on the five weeks after the 1.9.0 freeze), order-block and gap reactions now play out more often than the same trade taken at random moments (+2.5 [+0.4, +4.7], each setup judged against thirty of them; 0.0 before), and the refusal removes a contradiction rather than adding measured accuracy. The breakout detector, once it worked, measured worse than random — on the five weeks after the freeze the market reversed those breakouts far more often than it followed them — so it is recorded but never issued as a setup until a market clears it on its own record. Every read now also records a second, structural read of the chart — which side the market is committed to, how deep the pullback is and which attempt at resuming this is, whether it is breaking out, channelling or ranging, where price sits against the prior day, the session mean and round numbers, whether volume is joining the move, and breakouts that failed. Nothing scores it yet: it is recorded so the next setup definitions can be built and tested on what live reads actually saw, and replaying the full history with it produced byte-identical grades, patterns, verdicts and plans.

    Effect on past gradesNothing has been regraded and no stored read changes its verdict. New reads name a different primary pattern about three times in a hundred and a different direction about once in a hundred; grades move on about four in a hundred. Rule packs that check for a recent break of structure or change of character, or for an order block or gap at entry, now read the corrected detectors. Replay figures under the harness fill model, never expected returns.

    How we grade →
  6. EngineSep 13, 2026· v1.9.0

    Crypto, forex and stock grades are re-weighted from evidence — and the evidence says confirmation is late

    The replayed record was asked a stricter question than profit: whether the market played into a setup — reached the first target before the stop — more often than the same trade taken at a random moment of the same month, so that the distance to the target and the market’s drift cancel and only the timing the analysis added remains. Under that test the shipped grading was slightly worse than random across 184,852 replayed setups, and its A grades were its worst-timed ones: by the time trend, momentum, the higher timeframes and a fresh structural break all agreed, the move was mature. The engine’s own nine factors were re-weighted with the signs the evidence asked for. Trend alignment, multi-timeframe agreement, location and structural confirmation now count against a setup in crypto, forex and stocks, momentum and volume in some of them; entry location relative to the mean and the volatility regime count for it. Refitted quarter by quarter on data it had never seen, the top fifth of setups under the new weighting beat random on every unseen quarter in crypto (+2.4 points, 7 of 7) and forex (+3.6, 10 of 10) and on 7 of 9 in stocks (+2.5), and sat 7 to 10 points ahead of the old A family. Letter grades in those three markets are now set per market and timeframe: the A family is the top 15% of the setups the engine issues there, and the bottom 20% — the setups the record says play out less often than random — grade D or F and are refused with the reason shown. Stops and targets were tested across a grid of distances and left where they were. Gold and indices are unchanged. Disclosed in full: the frozen weights were checked on five weeks of bars fetched after the freeze, and the registered test on the 1H and 4H timeframes did not pass — every estimate pointed the walk-forward’s way (+4.1 points top fifth over bottom, +3.5 over random), but 35 trading days cannot close the interval that test asked for, and the 5m and 15m timeframes cleared the same bar on their own. The rule as registered said nothing ships; this version ships on the owner’s decision, the same check is re-run at twelve weeks, and any market that fails it then reverts to its declared weights.

    Effect on past gradesNothing has been regraded and no open journal row changes. New crypto, forex and stock reads grade under the fitted tables and their own letter bands, so most will score lower on the 0–100 scale than before while the letter is calibrated to the market; roughly one read in five that used to be a setup now says no-trade. One thing to know about your rulebook: the default “Trade with the higher-timeframe trend” rule caps a read at B− when the charts above lean the other way, and in these three markets that is exactly where the evidence now puts the best setups — on the replayed record the rule caps about three-quarters of the reads whose score reaches the A band. Every capped read says so in the cap’s own sentence. The rule is yours: keep it and the A band stays small, or switch it off in Settings and the evidence-based grade shows uncapped. Confidence curves are refit on the new score. Replay figures under the harness fill model, never expected returns; no spread or cost figure appears in this version’s evidence, because spreads differ per trader and are not a property of the analysis.

    How we grade →
  7. EngineSep 9, 2026· v1.8.0

    Gold is graded against its own market, the daily read becomes a measured gate, and a bar-by-bar layer ships measured but unscored

    Three things for gold, each with its evidence. First, spot gold was being graded against equity volatility thresholds it sits at less than half of, so the volatility-regime factor read almost half of all gold setups as abnormally quiet; it now has thresholds fitted from its own candles, and that share fell from 49% to 13%. Second, the walk-forward record was asked, under criteria written before the corpus was regenerated, whether the daily aggregate predicts a 4H gold read — and it does, in both directions and in every year measured: reads with the daily earned +0.19R on held-out data, reads against it −0.26R. That rule lives in a new lens, “Price Action — Spot Gold 4H”, a staff adaptation of Al Brooks’s continuation method at reviewed tier; replayed with the lens on, its 4H passes earned +0.33R at a 50% hit rate against −0.03R for what it refused, and below 4H it helps nothing, so it is built for 4H alone. Third, the engine now reads every bar as a bar — signal-bar close, follow-through, always-in, second entries — and measures all of it on every read without scoring any of it; on gold none of it cleared a floor (second entries were worse than first ones on the 1H), so those two Brooks concepts stay manual confirmations in the lens. Elsewhere, the same measurement promoted a “second entry” pattern in two non-gold cells (forex 5m, stock 1H) that cleared the registered floors. Weights were refused for every class again. One gold resting-limit cell (5m trend pullbacks) fell seven fills short of its floor on the regenerated record and reverts to close entry. Every confidence curve is refit on the current engine, and gold gets one of its own.

    Effect on past gradesNothing has been regraded and no open journal row changes. New gold reads will grade differently from today — the regime penalty is gone, so more of them land in the B and A bands — and their confidence figure comes from gold’s own curve. If you apply the gold lens on 4H, reads against the daily aggregate are refused with the reason shown; applying it on any other timeframe warns first, because the record says it would filter in the wrong direction there. Replay figures under the harness fill model, never expected returns; the live record holds 150 resolved gold reads, far too few to confirm any of this yet.

    How we grade →
  8. EngineSep 1, 2026· v1.7.0

    Gold joins the resting-limit cells; two new placement questions get measured and answered “no”

    Two things happened, and they point in different directions. First, the walk-forward corpus was rebuilt and grew to 268,637 replayed setups, with commodities now carried in the main record rather than a side file. Re-running the unchanged entry criteria on it re-cleared all seven existing cells and cleared two more — liquidity sweeps on commodity 1H and trend pullbacks on commodity 5m — so gold reads in those two cells now enter at a resting limit at the zone’s edge instead of at the close. Nine cells in total. Both new cells are the smallest in the set, at 365 and 304 simulated fills against a floor of 300, and they cleared the same three thresholds every other cell did. Second, the engine now measures two questions it could not answer before: where inside the zone a limit should rest (the midpoint and the far edge, against the near edge it currently uses), and whether the target ladder should be walked from the limit rather than from the close. Both were registered with their thresholds written down before the data existed. Neither cleared. Deeper limits earn more per fill and fill far less often — 44–54% at the midpoint and 31–41% at the far edge, against a 60% floor — and in two cells the extra return per fill would have passed on its own. The floor answered them, and the floor was not moved. Targets from the limit beat targets from the close in every cell measured, but never by the required margin on enough pairs.

    Effect on past gradesNothing has been regraded and no open journal row changes. If you trade gold, new reads on 1H liquidity sweeps and 5m trend pullbacks will now show “Entry · resting limit” with the band drawn on the chart, and can expire unfilled — that is the one behaviour change here. Crypto, forex, stock and index reads are unaffected. The two placement experiments changed nothing at all: every plan still rests its limit at the near edge and still takes its targets from the close.

    How we grade →
  9. EngineAug 25, 2026· v1.6.0

    Entries move to where the pattern said they belong — where the evidence cleared it

    Every plan used to enter at the price the chart was analyzed at, whatever the setup described. The pattern engine has always computed the band it actually meant — the swept liquidity pocket, the order block, the level a breakout retests — and since 1.4.0 that band has been simulated in the walk-forward record as a resting limit, paired against the close entry on the same rows. Four liquidity-sweep cells cleared the pre-registered criteria at release (at least 300 simulated fills, a fill rate of 60% or better, and at least +0.10R over the close entry on held-out data) — crypto 15m and 1H, forex 5m and 15m — and the fit on this version’s own corpus adopted three trend-pullback cells alongside them (crypto 5m and 15m, forex 15m): seven fitted cells in all. In those cells the plan’s entry is now the limit at the zone’s near edge, the reward:risk and position size are computed from that limit, and the plan states its terms — the limit rests for eight bars, and if price never comes back, the plan expires unfilled. No chase. A limit that never fills is recorded as exactly that: not a win, not a loss, excluded from every statistic, with a notification when a trade you took expires flat. Everywhere the evidence did not clear — including stocks, where the same comparison showed zone entries losing money — the entry stays at the close and the zone is shown for what it is: where the pattern would rather enter, unproven. CHoCH reversals derive their bands too, measured from this version onward so a future fit can judge them on the same criteria — that comparison has so far adopted nothing for them.

    Effect on past gradesNothing has been regraded and no open journal row changes. New reads in the seven flipped cells will show “Entry · resting limit” with the band drawn on the chart; taking one logs a pending limit that either fills and resolves like any trade, or expires “never filled” with no R booked. Reads everywhere else look exactly as they did last week.

    How we grade →
  10. EngineAug 5, 2026· v1.2.0

    Higher timeframes now carry the weight the label always implied

    Two problems, one cause. The multi-timeframe factor was weighted 0.08 on equities and forex — sixth of nine, tied with momentum — so a 4H flatly opposing a 15m long moved the read by about three points, while the engine described that factor as the heaviest after setup quality. It only ever was on crypto. And the default strategy rule labelled “Trade with the higher-timeframe trend” read the structure of the chart being graded, not the ones above it, so a trader who left the defaults alone was protected against fighting the 15m and told nothing about the 4H. The factor is now 0.14 on both tables, taken a hundredth at a time from the six factors that describe the base chart alone; the default rule walks each higher leg and names the one that objected; and a read with no higher timeframe at all is capped at B with the reason stated rather than scoring the absent context as a silent neutral. On an identical setup with only the higher timeframes flipped, the score now swings 12.3 points instead of 7.

    Effect on past gradesNothing has been regraded. Every analysis already in your journal keeps the grade it was given, stamped with the engine version that produced it. Two things will look different from now on. Setups that fight their higher timeframes will grade lower than the same setup did last week — a counter-trend read that came back A- may now come back B- with a flag naming the 4H or the daily. And if you never edited your strategy rules, the “higher-timeframe trend” rule starts doing what its name says the first time your settings load; your cap and your on/off switch are carried across untouched.

    How we grade →
  11. EngineAug 3, 2026· v1.1.0

    The engine now knows which market it is looking at

    Every scoring threshold used to be one number applied to everything: a volatility regime centred on 0.8% ATR, overextension at RSI 78/22, a chase cap at 1.5 ATR. Fitting those to real candles showed 0.8% is the median for hourly equities and almost nothing else — forex runs at 0.02–0.79% and crypto at 0.12–4.93%. The volatility factor was therefore penalising every forex read, and crypto at both ends of the timeframe range, at maximum, permanently, for behaving normally. Thresholds are now measured per asset class and timeframe from that market’s own history, with the sample size and fitting date recorded. Crypto additionally scores perpetual funding, open-interest change and crowd positioning, and reads genuinely fetched 4H and daily bars rather than a roll-up of the timeframe being traded. Equities are capped when a setup sits inside an earnings window.

    Effect on past gradesNothing has been regraded. Every analysis in your journal keeps the grade it was given, stamped with the engine version that produced it — a record of a decision has to show the rules as they stood when you made it. Expect new reads on forex and on 5m or daily crypto to score higher than comparable ones from last week; that gap is the correction, not an improvement in the setups.

    How we grade →
  12. FixAug 3, 2026

    Crypto prices no longer start with a request that cannot succeed

    Binance was the primary crypto source and answers 451 from the United States, so every crypto read paid a doomed round trip before falling through to Coinbase. The fallback worked, which is why nobody noticed. Coinbase now leads, OKX is second, and a source that refuses the region is dropped for the session instead of being retried on every analysis.

  13. NewJul 25, 2026

    Live market scanner, public and keyless

    The whole USD crypto market ranked by an attention score computed from 24-hour movement, range position and real traded turnover. No key, no account. Opening a symbol runs the actual grading engine on real candles.

    Open the scanner →
  14. ContentJul 25, 2026

    ORIN Academy

    Seven tracks, sixty-two lessons in the syllabus, each ending in a checkpoint you have to do rather than a quiz you can guess — mark the invalidation, grade the setup against the engine, size the position, spot the trap, order the sequence. Unwritten lessons are listed rather than hidden.

    Start the Academy →
  15. ContentJul 25, 2026

    Help Center

    Every article opens with the answer in one or two sentences that survive being extracted alone, because increasingly the destination for this content is an answer engine rather than a click.

    Open help →
  16. ContentJul 22, 2026

    Trade Playbook

    A page per setup family, each rendering a generated instance from the same shape data its description is written against — so the annotations point at the bars they claim to. Every entry leads with invalidation before target.

    Open the Playbook →
  17. FixJul 18, 2026

    Build gates for published content

    Three checks now block the build: demo data across 1,350 generated fixtures, five uniqueness rules per solutions page, and the content doctrines. Three real defects were caught on first run, including a worked example structurally incapable of the risk-reward it claimed.

    Editorial standards →
  18. NewJul 15, 2026

    Solutions and Platform clusters

    Sixteen public pages, each running a working tool above the fold rather than describing one. Nothing behind a signup wall.

    Browse solutions →
  19. EngineJul 8, 2026· v1.0.0

    Grading engine 1.0.0

    The scoring pipeline as it stands: features computed deterministically from candles, risk plan derived from structure, grade from nine weighted factors read across up to three timeframes, then calibrated against your own resolved history.

    Effect on past gradesNothing in your journal was regraded, and nothing ever will be. Each analysis snapshots the engine version and the lens it was graded through, because a record that silently rewrites itself is not a record.

    How grading works →

ORIN is analysis software, not investment advice. Markets carry risk of loss. Read the risk disclosure.