Glossary

The newer segment program that behaves differently after a stop.

mamba s01's segment engine (since 27–29 Oct): two S0 from candle bodies after STOP and TARGET, technical and real trades, P&L in % and R. Default on cryptotrader /s0/ since 31 Oct.

We check whether we just got lucky, like with a coin.

Comparing selected trades with random sets of the same size. “Rand.” is the share of random sets worse than the rule (about 0.5 — no better than chance); p in Profiling is the share not worse.

⚖ The scales are opposite: a high “Rand.” and a low p mean the same. In cryptotrader /dynstep “Rand.” uses 5000 random sets.

A system stitched from pieces: market quiet, past floors and a second try.

A planned combo: a DH signal + Segment context (−1/−2/−3) + two entry attempts (dh_signal.py, signal_context.py, two_attempts.py, analyze_signal.py).

⚖ Simple/complex signal, entry timing, two attempts and the meaning of “context −2” are open; not a finished implementation.

The floor “switched on”, but that doesn't yet mean we bought.

The geometry turns into a directed segment on a close at or beyond KEY. Not a trade entry: trade admission (technical/real) and execution are separate states.

We compare the rule with entering “at random”.

In the Fable study the baseline is entry on a random candle along the large chain with the same exit; hit — which comes first: +1 or −1 STEP M within 8 hours.

⚖ The s01 baseline is all trades without a filter; hit is not the win rate of a given exit.

The “quiet zone”: the market has squeezed.

The DH line within [B0, B1 + R/16] — the lowest 3/16 of the profile, a squeeze. s_abs sits there 42–53 % of the time.

Pick the stair using weekdays only, when the market is livelier.

A STEP calculation mode: profile, window start S0 and candidate runs use Mon–Fri UTC candles only; the chain itself and the phase use all candles. Not for 1W. STEP comes out 2–24 % larger.

A thread tying the fast and the slow “quiet” together.

A line from an M min to the nearest S min within the M window: shows where the S→M sequence began, gives no trade direction.

Trades at a “good moment” count for more.

In stepfit a trade weight = 1 + bonuses for zone, signal and Charge. A trade counts as successful with PARTIAL/TARGET or a profitable opposite KEY; success weights are tunable.

⚖ Structural “success” is not the same as PnL > 0 in the core's win rate.

Another “quiet” inside a quiet that is already going on.

A repeated M min while a min segment is already active (an in-between M max may have been dropped by min_bars). Drawn grey.

Ideas about what the market might pay us for, and how to test them.

“Own system” candidates with a pre-written rejection criterion: H1 — overheated crowd (funding, OI, long share), H2 — pullback after a liquidation cascade, H3 — direction-free funding harvest, H4 — trendline breakout. H1 rejected, H2 passed 2021–2024.

The deposit the broker holds for each contract.

Finam's collateral per contract, direction and account. Trade size is the smaller of the risk allowance and the available margin.

⚖ Account categories KNUR/KSUR/KPUR are named in SPEC without explanation.

Price “jumped” overnight without passing the middle.

A jump between the previous close and the new open. The engine first treats the open as a virtual close of the skipped move, then processes the real close.

A red candle that first went up counts as a green one plus a red one.

A counter-trend candle that set a new trend extreme is split in two: open → high (with the trend; its open becomes the new level 1) and high → close. Mirrored for downtrends.

Two staircases on one chart: with small and with big stairs.

A mamba lab tab: a small chain (STEP S) and a large one (STEP M) from one S0 on one chart, plus golden trend M5/H1 and DH.

Change the stair height when the market changes character.

STEP is recalculated by the calibrator as soon as the Volatility phase returns rescan_step; the new STEP applies from the next segment. BTC 30M 2026: −202.5 R vs −75.5 R for a fixed STEP (with fees).

⚖ The author's protocol intends a fixed STEP; both dynamic runs (23 Oct and 2 Nov) were worse or no better than fixed.

How long a trade lasts and how many happen in a year.

avg_bars — the average number of candles from entry to exit inclusive; per_year — completed trades per history length in years (denominator at least 0.01 year).

The width creeps one way while we wait for it to turn.

The search for the current DH width extreme until a k·STEP move back. A run has a length run_bars, a depth run_depth and an unfinished state.

We take the meter reading from the previous candle so as not to peek into the future.

A cryptotrader owner decision: the DH point of candle i equals the calculation on closed candle i − 1. The current OHLC does not move bounds, width or the squeeze flag; a candle's own result appears only at i + 1.

⚖ The author's DH (mamba, S0) computes on the closed candle itself and signals on the confirmation candle — it differs from frozenDH.

The longer the market “stored up quiet”, the bigger the charge.

How long the width last stayed in the minimum zone without a later visit to the maximum. Gives a normalised bonus to the DH weight in stepfit.

The moment both the fast and the slow DH lines say “quiet”.

A min of line M after a confirmed min of line S; a max of M resets the series. The M5 chain guesses the side after a star 49–57 % of the time — a coin flip.

A trend that lasts until price breaks both safety lines.

A run from a reference candle until a close beyond both lines of “Andrey's trend” (level 1 gold, level 2 blue). M5 and H1.

⚖ cryptotrader only has “Andrey's trend” without golden mode. Run start: the code uses the reference candle, Andrey said any crossing of both lines.

Read more →

Small width is “quiet”, big width is “noisy”.

DH width areas in STEP set by a given task; in the Fable study — below 3·STEP and above 7·STEP.

⚖ Not the same as the statistical B0…B8 barriers.

Shelves on the quietness ruler: which one the width landed on.

Classes of where the width sits on the B0…B8 profile with an R/16 buffer: below B0, B0 to B1 − buffer, B1 ± buffer (barrier), then AT_B1_BUF, AT_B2, ABOVE_B2; mirrored at the top — IN_BARRIER_TOP, AT_B8, ABOVE_B8. Right bounds excluded.

Marks of the quietest and noisiest places on the 4-hour chart.

A separate 4H layer: S6 width (open and body midpoint of 6 candles / close) and its max and min in every window of 25 values. An event is known at knownAt = window end + 4 h.

RTS Index data sits in the database apart from the trading instruments and serves only as background reference.

RTS Index candles (FINAM, RTSI@RTSX), M5 and H1, stored in market_data_hub.candles on S0 with asset_class='index'. Table readers select only 'futures' and 'equities', so the index stays out of chains, monitoring and dashboards. Appended every 5 minutes by the s0-rtsi-refresh.timer.

A program that tries hundreds of stair heights and picks the best on past data.

The author's protocol: candle ranges → two-Gaussian mix (μA, μB) → small and large ranges → run candidates through the chain → max score in the first non-empty (small) range.

⚖ STEP is fitted on the Core v2 model but traded by the chart engine — “optimal for the calibrator's model”.

A colour picture: which stair height made money later and which lost.

A cryptotrader page (v0.2–v0.7, 22 Oct): the author's calibrator from a start date, then every candidate of both ranges is run forward through the core; colour — R after costs or points over time.

When many get forced out, price overshoots and then comes back a bit.

At a 5m close, open interest fell below its 1st percentile over the hour while price moved beyond its 99th (coin thresholds over 90 days). Trade against the move, exit after 4 h: +0.53 % net, t = 3.04, 9/10 coins, 4/4 years (2021–2024).

⚖ The 2025–2026 holdout run was not done — the track is paused.

Like a bus fare: the more rides, the more you pay.

The exchange fee per trade plus slippage. Crypto models use 0.055–0.06 % fee and 0.02 % slippage per side (0.11 % round trip); on small STEPs this eats the whole edge (up to −0.58 R on 5m). “Net” figures are after costs.

A trade that didn't finish in time counts as neither a win nor a loss.

A position still open at the end of data is valued at the last close and flagged separately; a forced valuation, not TP/SL. Main tables count closed trades only.

What happened on the last three floors.

Type, direction and PnL of the three previous segments: −1 is the nearest one, then −2 and −3.

One “cell” of the system: which market, which candle and its settings.

In cryptotrader /s0/ a contour is a Bybit instrument + TF with its own parameters (s0/contours/<symbol>-<tf>.json) and one current chain. The word also names the trade accounting lines (real and technical) and the agents' work folders lab/claude and lab/gpt.

⚖ In mamba a “contour” is only the technical vs real trade within one geometric chain (Technical vs real trade); in cryptotrader the word mostly means instrument + TF.

Different ways to pick the stair — and none came out clearly best.

The bench.py rig (22 Oct): A — the author's rule; B — max score over both ranges; M_k — max score among STEP ≥ k·median; F_k — STEP = k·median; N — max R after costs. On 5m–1h all lose, none is reliably better.

A big order sitting in the queue — sometimes filled, sometimes pulled.

An order-book level at or above the 99th size percentile of its side (≈10 largest of 1000) that lived ≥ 10 s; stops being large below half the threshold. The collector logs its life and best bid/ask.

⚖ Bybit has no order-book history — only our own recording since 8 Oct 2026.

Where ideas are tested on history before trading.

The research side of s01 (mamba/lab): tabs “Golden trend”, “Two chains”, “STEP fit”, “DH plan”, per-task reports.

A — take half and keep waiting; B — take everything and just finish drawing the floor.

The author's modes: A closes 50 % at PARTIAL; B closes 100 % and then carries the segment as a “phantom” — the structure continues with no trading result.

⚖ The DH filter and the GMM components also have A/B — different things.

Two boxes of stair options: small ones and big ones.

Calibrator candidate ranges: small μA/4…1.5·μA, large 1.5·μA…2·μB. Grid step 10^⌊log10(width/50)⌋, 50–500 candidates. Logic A takes the first non-empty range — in practice always the small one.

⚖ In review #151 over 27 contours the small range always won, though in 19 the max score was in the large one. mamba uses the best large candidate as STEP M.

We entered and were already past the stop.

A STOP where the entry price is already beyond the buffered working stop. Profiling flags such cases for diagnostics.

How far price usually “jumps” in one candle.

Median high − low of window candles. The author's STEP ≈ 1 median (1.01–1.11 on 5m…1d); “STEP fit” searches 1…10 medians.

A list of markets worth watching right now.

An instrument's initial state in the DH 1H scanner before any label. READY moves it to the watch list, RIPENING keeps it there with a note.

⚖ Not the same as the s01 multi-chart “Overview” tab.

A big weekly staircase we look at for a hint.

A WeeklyEngine level grid from max close S0 with STEP_1W steps and half-steps, updated weekly; the filter measures distance to the nearest levels.

Three “lenses”: look at a short, medium and long stretch of history.

Short, medium and long windows (in candles) for DH width. mamba: plain filter 45/70/120, DH S/M 25/120, H1 scanner 50/250. cryptotrader DH-Levels: 5m 20/60/120, 1h 10/35/70, 4h 10/25/60, 1d 5/10/15; ratio = S width / L width.

⚖ There are no common windows: each module has its own defaults.

The time while this contract is the main one on the exchange.

The period a futures contract is the active one. Cycles and trends are counted per contract, warm-up takes history before activation (DH S/M — 30 days, forward — 14).

The point we wait from may “drift” a little — but by no more than two stairs.

The old Aiming scheme: from anchor A (e.g. the finished TARGET) whole-STEP price deviations shift the keys while U + D ≤ 2; the fraction is dropped.

⚖ Since 29 Oct the main s01 engine (aimcore) keeps two S0 from candle bodies after STOP/TARGET, and the drawn A is only the S0 of the closed side; cryptotrader Core v2 keeps one anchor and the U + D ≤ 2 budget.

A confident candle with a big body and almost no wicks.

A candle with body ≥ 70 % of its range (was 75 %). Opens a golden trend in its direction and sets level 2.

How many bets are open on the market right now.

The number of open futures contracts. Bybit gives it at 5-minute steps and up (since 2021), the long/short account ratio since 2021; cryptotrader stores it in db/derivs.sqlite every 5 minutes.

A piece of history hidden until the very end, like an exam ticket.

Rules are tuned on 2021–2024, and 2025–2026 is run once after the rules are frozen, without tuning. Shared by the Claude and Codex contours and step_prot.

The first try: trade the exit from a squeeze on 5-minute candles.

cryptotrader's first backtest (27 Sep): 1h/4h squeeze (OR or AND) + a 5m close breaking the S zone; SL A — the opposite S bound, SL B — the signal candle extreme; targets 3R/4R. 8 variants on 13 coins over a year.

How many failed floors in a row came before this one.

The code counts STOP, STOP_PARTIAL and REV outcomes as bad; bad_streak_before is the unbroken run of them before the current segment.

⚖ The profiling.py comment lists only STOP/REV, the running code includes STOP_PARTIAL too.

Our own, simpler program for choosing the stair height.

The mamba lab method (31 Oct): tries STEP from 1 to 10 median ranges in 0.1 steps without GMM, scored by Σ(DH weight × R) in 100 % and 50/50 modes.

⚖ Not the same as the author's calibrator: no Gaussian mix, the lower bound is exactly 1 median. The note “no forward test needed” is a premise, not proof.

Entered late — the robot still runs the trade by the program's plan.

The helper ties an account position to an open model trade of the same side (late entry); levels come from the model. Without a matching model the “no-signal” position gets no automatic exit.

⚖ The two-M5 binding limit was lifted on 29 Oct: it cut off an entry made 13 minutes late.

Protection against “got lucky once”.

When dozens of rules are tested, one will look “good” by chance. Holm correction and comparison with random sets filter such finds out.

How far the width must pull back before we believe in the turn.

DH min/max detector parameters: k — the k·STEP move back needed to confirm, min_bars — the minimum run length. Typically k = 2 and 3 candles.

⚖ A short run may give no signal though the detector flips state. In cryptotrader /dynstep k·STEP uses the chain's STEP at that candle.

An almost confident candle — fell just short.

A candle with a body of 65 %+ of its range: highlighted on the chart but, unlike a Reference candle (≥ 70 %), does not start a trend.

Different “traffic lights” deciding whether to let a trade in.

A — after a min; B — after a min in the barrier; C12/C48 — within 12/48 candles of a min; D — s_abs in the barrier; E — with the breakout; “−” — control/opposite. D4 (lab) — S min confirmed in the barrier, S still in it.

Enter when both quietness lines have just “woken up”.

A filter frozen on 3 Nov: S min confirmed in the barrier, S was there at confirmation and left; M also left after its barrier min, and the M max is not yet confirmed. STEP M = 2.5 × S.

⚖ Not the plain rule A of the DH filter and not the H1 READY status; awaits an independent Forward test.

Enter when the trend was just born and the market is waking up.

Enter an s01 segment if a fresh with-trend golden M5 exists (reference ≤ 3 / ≤ 6 candles ago) and DH S is at a bottom (D4). History: +0.644 / +0.457 R; forward-tested since 2 Nov.

A pause between floors: wait to see where price goes next.

Finding the next segment after TARGET or STOP: deviations in whole STEPs accumulate in U and D (U + D ≤ 2), keys shift, new S0 = K ∓ 3·STEP.

⚖ “New stop” (mamba) uses two S0 from candle bodies instead of one anchor.

Draw a sloped line through prices and enter when price breaks it.

The owner's method: short — a body closes below a line of rising lows, long — above a line of falling highs; TF 1h or 4h, points refined on 5m. Rules not frozen, never measured.

Testing a rule on days that didn't exist when it was invented.

A rule is frozen at a date and only new trades after it are counted. The only honest estimate, since history has already been “peeked at”.

A “trend” that died right after it started.

A golden run of ≤ 2 candles (27 % on M5). The tab ignores runs < 3 candles, research counts them — numbers are not directly comparable.

How deep we fell before rising again.

A fall of the cumulative result from its peak. In Core v2: chain_dd on cumulative PnL, intra_dd — the worst close from entry within a trade, max_dd — the larger of the two.

⚖ In cryptotrader's “Own system” drawdown is a share of the 10 000 USDT capital (10–20 % allowed); in the first pass — on closed trades in R, without intra-trade.

Candles are split into two piles: calm and wild.

A mix of two normal distributions over the window's high − low ranges: component A — calm candles (smaller mean μA), B — wide ones (μB). Not consecutive market regimes.

⚖ Not to be confused with the DH profile (B0…B8) (B0…B8 over channel width).

A ruler from “very quiet” (B0) to “very wild” (B8).

From 200–2000 past line values: R = max − min, Bk = min + k·R/8, buffer R/16. Defines min/max zones.

The staircase we use right now; old ones are kept in the archive.

The chain configuration currently applied to an instrument in s01 (STEP, start). Applying or reverting a STEP creates a new version; old ones go to the archive.

Price dropped 3 stairs from the top — it changed its mind; start a floor the other way.

A close pullback of 3·STEP from the run's close extreme. The opposite segment opens on the same close with S0 = the extreme.

Either split the win along the way, or hold all of it to the end.

STEP fit (stepfit) modes: 50/50 follows the trade's partial exits, 100 % ignores PARTIAL and holds until TARGET, STOP or the opposite KEY. Same geometry.

⚖ stepfit “100 %” is the opposite of logic B, where the whole size closes at PARTIAL.

A switch: the robot sleeps, watches, works or rehearses.

Off — does not read the account; watch — reads and notifies; auto — closes by the rules; dry-run — writes test orders without sending them.

⚖ A mode in the spec does not mean it is enabled on the remote server.

We look where price ended the candle, not at the twitches inside.

All segment levels and trend lines are checked on the candle close; wicks only “test” a level.

A bar on the chart: where price opened, closed and how far it swung.

Price over a time slice: open, high, low, close. The body is open–close; wicks reach high and low.

A trade the person made on their own, without a hint.

A trader's manual entry (price, STOP, size, reason) without a model signal. Counted in risk and the day stop; the desk does not watch its STOP, the 0.5 STEP buffer is not applied.

⚖ Not the same as a “no-signal” position the helper finds on the account (Late entry and “no signal”).

An attempt to invent our own trading system without others' tools.

A cryptotrader track since 8 Oct: a system from scratch for Bybit futures, 10 coins, a 10 000 USDT test, 10–20 % drawdown, a 10–40 % yearly goal with return/drawdown ≥ 1. Paused since 13 Oct.

A trade “as the program would have made it”.

A trade the mamba desk rebuilds from signals: ENTRY/TRADE_ENTRY with admission and partial TRADE exits. Real by default, technical on request.

A “floor” of the price staircase, 8 stairs high. Climb the floor — take the profit.

A corridor 8·STEP high from point S0 with levels KEY (3), STOP (2, with buffer 1.5), PARTIAL (6), TARGET (8). Close-only decisions. Segments form a chain.

⚖ Core v2 (cryptotrader) and “New stop” (mamba) differ on STOP after PARTIAL, TARGET exit price and aiming after STOP/TARGET; same exit in 60–73 % of segments.

Read more →

The market squeezed almost to its narrowest in a long time.

A DH-Levels flag: widthS ≤ rolling minimum × (1 + tolerance/100), default tolerance 20 %, minimum over the last 500 values. A research mark, not a direction signal.

⚖ In the author's DH (mamba) a squeeze is the Barrier B0…B1 + R/16 on the profile, not closeness to the minimum.

Looking for the moment the market goes quiet before a jump.

A research plan (v2.1, 8 Nov): an asset-state indicator “asleep / ready” from DH and STEP; a breakout is a wick leaving the squeezed DH channel, examined on 5m underneath. No calculations yet.

⚖ Breakouts here are counted by wicks, while Andrey's trend and segments use close only.

Names of entry variants in one particular study.

Candidates of the Fable study (5 Nov, RTS/MIX): A2 — a confirmed M min from the low zone along the large chain; B3 — a small-chain ENTRY along the active large chain; B6 — the same plus a recent low M min; C11 — a control golden M5 start against the large chain.

⚖ Fable B3/B6 are not the forward rule B3/B6: do not carry names across studies.

Price jumped a whole floor — we mark it but don't trade it.

If a gap jumps past both KEY and TARGET, the segment stays in the chain without a trade and the next one is glued with S0 = its TARGET (8·STEP step).

The moment a trend ends.

A close beyond both lines (golden mode) or beyond level 2 (plain). The trend closes on the previous candle.

We only look at those who survived to today — and the picture looks better than it was.

Coins are picked as liquid today (top-15 on the snapshot date), so past returns are inflated by “survivors”. Noted in the first pass and in H2.

A board: which market is “ready to wake up” and which is resting.

An H1 scanner (S 50 / M 250): READY — S bounces 2·STEP off a barrier low and M near its bottom; PAUSE — max M; RIPENING — draft. Picks instruments for the watch list.

Unlucky a few times in a row — enough for today.

A failure-counter limit after which entries are blocked until the end of the day. TARGET or a profitable KEY resets it, STOP or KEY without PARTIAL adds 1.

How much time fits in one candle: 5 minutes, an hour, a day.

The duration of one candle. mamba runs on M5 (H1 built from closed M5 by Moscow hour); cryptotrader — 5m…1w.

We enter for real only if half the target is still far enough.

Technical — every segment trade. Real — only if at least 2·STEP remain to PARTIAL at the entry close.

Labels for “how the floor ended”: reached the ceiling, turned back, hit the stop, or still going.

Profiling labels: TARGET_EXT — after TARGET the next entry is the same way at least ext_threshold STEP away, TARGET_EXT_REV — the other way; PARTIAL_REV / REV — reversal after PARTIAL or without; STOP_PARTIAL / STOP — stop with or without PARTIAL; ACTIVE — a segment with no exit yet.

⚖ Core v2 checks STOP only before PARTIAL, yet the classifier supports STOP_PARTIAL: whether it can occur depends on the engine.

One exchange working day on Moscow time.

For the desk — a Moscow calendar day including the evening session; the risk budget and percent are fixed at the start of the day.

The “plain” trend: ends at the first break of the blue line.

A two-line indicator (levels 1 and 2); the trend ends on the first close beyond level 2. Golden mode needs a close beyond both.

The safety line that crawls right behind price.

The open of the last with-trend candle; moves with each such candle.

The second, lower safety line.

The body midpoint of the last with-trend reference candle. Labelled “DH-level” on the tab — unrelated to the DH indicator.

⚖ The name “DH-level” matches DH by name only.

First we check where price opened after the jump, then where it closed.

Two processing phases of a candle after a Gap: virtual close = open, real close = the candle close. The virtual candle is not added to OHLC.

⚖ In cryptotrader a virtual_close on open ≠ previous close “does not prove a missing bar”; full candle volume is only known at real_close.

A piece of the chart where price went one way.

The interval from a trend's start to its break. For phase studies activity is measured separately as the share of in-trend candles over 48 candles.

⚖ The geometric gap section is a different object; there is no single activity threshold.

A check whether the market became too wild or too calm for our stair.

The author's module compares the current range profile with the one STEP was chosen on and returns use_saved_step or rescan_step. Re-scanning by it was tested — no better than a fixed STEP.

Trade small floors only in the direction the big floor is going.

Check of 3 Oct: a 5m/15m entry passes only if the active confirmed 4H segment points the same way; 4H state is taken strictly known_at(4H) < known_at(entry). No universal improvement; BCH 15m gave +6.57 USDT on 83 trades (PF 1.007).

⚖ mamba has no predicate named “HTF filter”; it has higher TFs (candles built from M5) and a separate weekly grid_filter.

Enter after a run of failures, or right after the previous trade.

Profiling filters: K — at least K bad segments in a row before entry; N — entry no later than N candles after the previous exit; K&N — both at once.

⚖ K here is a counter, not the price KEY nor the DH multiplier k. K-short / K-long appear in the sections without their own definition.

A fee for holding a position — paid by the more crowded side.

A periodic payment between longs and shorts of a Bybit perpetual, usually every 8 h. In backtests: direction × rate × 5m open price at the payment time, only for positions opened before it.

⚖ Many S0 checks leave funding out — the reports say so.

We check whether trends repeat “on a schedule”.

Trend regularity measures: C1 — start→start of neighbouring trends (= C3a + C3b), C2 — low→low / high→high of confirmed price extremes, C3a — trend length start→end, C3b — pause end→start of the next. Regularity is tested against memoryless models and shuffled candles.

⚖ C2 is postponed (#88). The C3b “pause” is not the PAUSE status in DH 1H.

How often we reach the ceiling, hit the stop or turn around.

Shares of completed trades with TARGET_HIT, STOP_HIT + GAP_STOP, REVERSAL and with PARTIAL. They need not sum to 100 %: PARTIAL can coexist with the final outcome.

The smallest step price can move.

The contract's minimum price increment. STEP is made a multiple of it (0.125 on crypto), which changes the calibrator's choice. Roubles per point = step_price / tick_size.

Corridor width not in money but as a percent of price.

Absolute DH channel width divided by close × 100. Comparable across instruments with different prices. On non-positive close mamba returns 0.

⚖ In cryptotrader DH-Levels everything is built on % (the compressed flag); in mamba the main series is the absolute s_abs (DH width), and the DH (%) panel on s01 was removed (#71).

A “textbook” floor and an “almost textbook” one.

Segment quality classes (classify_quality): ETALON — S0 and entry deviations ≤ 1/3 STEP, EXTENDED — ≤ 2/6 STEP.

⚖ The file header (30 Jun) has absolute thresholds 0.020/0.045, the code scales them by STEP.

The reference point used to find where the first staircase starts.

The close of candle 71 of the window: the start S0 is the extreme of the first 70 candles farthest from it (by body). High → SHORT, Low → LONG; on a tie — High.

⚖ In dh_signals.py (12 Oct) the “anchor” is a DH width minimum — a different meaning.

The first “official” program that builds segments.

The author's segment core (21 Sep 2026) with owner's fixes #107–#118 (13 Oct). P&L in points. Available on /s0/ via ?e=core.

A “quietness meter”: shows whether the market is squeezed or running wild.

Andrey Maltsev's indicator: channel width from candle open and body midpoint, B0…B8 profile, barrier, min/max signals with a k·STEP threshold. No direction; an entry filter.

⚖ The abbreviation is never expanded in the projects. Two versions share the name: DH-Levels (Sept) and the author's DH signals (since 19 Oct). Windows differ: 45/70/120, 25/120, 50/250 (H1).

Read more →

The two “anchor” prices of each candle DH is built from.

1) The candle points used by DH: open and body midpoint. 2) The early cryptotrader version (Sept 2026): three windows, width in %, a “compressed” flag.

Price climbed 3 stairs — the move is confirmed, we enter.

Level S0 ± 3·STEP. A close at or beyond it activates the segment (entry). The opposite KEY — a 3·STEP pullback from the close extreme — is a reversal.

Long bets on a rise, short bets on a fall.

Segment direction: LONG has TARGET above S0, SHORT below. LONG result = exit − entry, SHORT = entry − exit.

⚖ In “Golden trend” SHORT is also the label of a short run (Spike run), unrelated to direction.

A mark “it was quietest (or loudest) here” — placed once that is already clear.

A DH width extreme confirmed by a k·STEP move back; not a price extreme. idx is the extreme candle, confirmed_idx the confirming one; the signal is usable only from the latter.

A “quiet” period gives way to a “noisy” one.

States of the S→M sequence: an M min after an S min opens a min segment (star), an M max resets it and opens a max segment. Unrelated to S0 price segments.

Too little history yet to trust the quietness ruler.

NO_PROFILE — no DH profile yet; is_unstable — the profile exists but has fewer than 500 observations.

A robot partner that closes the trade for the person on time.

Paper logs the model's exit to a journal; the helper may send a broker order instead. In stage 1 the helper only reduces or closes a position — never opens or flips it.

At stair 6 we pocket half of the win.

Level S0 ± 6·STEP: close 50 % of the position, once per segment.

How much we made or lost.

A trade's result: LONG = exit − entry, SHORT = entry − exit, with partial closes taken into account. Money needs the point value.

⚖ Units differ: Core v2 — points, aimcore — % and R, the mamba desk — roubles, cryptotrader 4H/volume checks — USDT on a 1000 notional per trade, the first pass — R.

Profit counted in stairs. “+2R” — we made two stairs.

Trade result in its segment's STEP: R = points / STEP. Makes instruments comparable.

⚖ In #143 some results are in STEP M rather than STEP S.

How many times a typical win is bigger than a typical loss.

The actual ratio of average win to average loss. In Core v2 abs(avg_win / avg_loss), with no losses — a placeholder 999.

⚖ This is actual, not the planned target/stop. In cryptotrader's RR study (#202) the goal is TARGET in > 40 % of trades with RR ≥ 1:3.

How many times more than usual was traded on the breakout.

Volume of the closed signal candle / the mean of the 20 previous candles of the same TF. Groups: LOW < 0.75, NORMAL 0.75–1.5, HIGH ≥ 1.5; the filter is HIGH, fixed before the run. Result of 4 Oct: no universal boost.

The width of the “corridor” price fitted into over the last candles.

max(hi) − min(lo) over the window, where lo/hi are min/max of candle open and body midpoint. Rises when price leaves the channel, falls when the extreme drops out of the window.

The “floor” — the point where a segment starts.

1) The anchor point of a segment — its start; all levels are measured from it. 2) The name of the author's whole segment system (/s0/ page, “S0 · Segments”).

⚖ One word, two meanings: the point and the system. Read from context.

Our app that draws segments on Russian futures.

mamba's trading platform on Moscow Exchange futures (M5): segment chains, DH filter, the Lab. Site s01.smart-trend.ru.

The grade the calibrator uses to pick the best stair.

avgPnL × (0.5 + 0.5·WR) × (1 − 0.5·min(1, maxDD / (10·median))). No fees, in points — grows with STEP.

⚖ In the original Core v2 score = total_pnl·(0.5 + 0.5·WR)·dd_penalty, per trade — divided by completed trades; protocol v2 (2 Oct) replaced total with avg_pnl, and score_per_trade = score.

We study which floors came before good ones to guess the good ones.

A lab section: segment outcome types and the combinations before them, used to select entries. Tests K, N, K&N filters on history halves and neighbouring STEPs on the author's core.

The height of one stair used to measure price moves.

The scale unit of the segment system in price points. All segment levels are multiples of STEP; R = points / STEP. Typically ≈ one median candle range.

⚖ How to choose STEP is unsettled: the author's calibrator, mamba “STEP fit” and the owner's manual values disagree.

Read more →

Don't take a stair smaller than a usual candle.

A lower bound on STEP at the median candle range. The /step-map.html checkbox shades candidates below the median (like criterion a1); “scan from median” rebuilds the grid with the lower bound = median_range.

⚖ Over 2021–2026 a1 shrinks the loss vs the author (5m −156 → −53 R) but turns positive only on 4h/1w.

A stair measured not in dollars but as a percent of price.

STEP as a share of price: $500 at BTC 100 000 is 0.5 %. In the “Squeeze → breakout” plan the base unit is STEP / price; in percent the 5-year spread of quarterly STEP narrows (BTC 19.7× → 13.2×), but jumps of up to 6–8× between quarters remain.

⚖ On MOEX (mamba) STEP is nearly constant in points, because the index price changes little.

A small and a big stair for the same market.

Two scales per instrument: S — the working chain STEP, M — the best large-range calibrator candidate or Andrey's STEP. M ≈ 2.4–2.7 × S (RTS 110/267, MIX 250/675).

Price fell back near the floor — we were wrong, exit with a small loss.

Level S0 ± 2·STEP; fires with a 0.5·STEP buffer, i.e. at S0 ± 1.5·STEP. Trade risk ≈ 1.5·STEP.

⚖ Core v2: no stop after PARTIAL; “New stop”: closes the rest and then sets two S0 from candle bodies.

How confidently the result differs from zero rather than from chance.

Mean result / standard deviation × √n. Trend and hypothesis reports show it next to the mean (e.g. t +2.45 by day); the H2 pass threshold is t ≥ 2.

Stair 8 — the “ceiling”, take everything.

Level S0 ± 8·STEP: close the rest, end of segment. Then aiming for the next one.

⚖ Exit price: candle close in Core v2, the TARGET level itself in “New stop”.

Each week we pick the stair from the past and honestly see what happens next.

STEP is computed from past candles only (weekly, on Monday), the next week is traded with the chosen STEP, the chain is continuous. This is how the step_prot rig compares all calibrator changes.

⚖ Differs from mamba's Forward test: there a rule is frozen once and new trades are awaited.

How often we win, and how much bigger wins are than losses.

WR — share of winning trades; PF — gross profit / gross loss (on net); bp — basis point, 0.01 % of price. The trend reports (#29) give results in bp per trade plus PF.

⚖ PF and bp appear only in cryptotrader (strategy-29: PF 0.28–1.11; volume and 4H checks); no PF in mamba. WR: Core v2 divides winners by all completed trades incl. zero ones; Filter5MSegments excludes zeros; in cryptotrader “net success” is net > 0 after costs, breakeven is not a win.