Does waiting for price, momentum and volume confirmation reduce another break of the low? · By CoinNudge Research · Method reviewed 2026-09-15 · Guide updated 2026-09-15 · Historical study updated · Observation range: – · Calculation local-bottom-confirmation-1.0 · Auto-refresh about every 86400 seconds
Has Bitcoin Bottomed—or Is This Just a Relief Rally?
Current answer: As of , using Binance BTCUSDT and ETHUSDT spot historical 1h candles: Watch the combination of price above EMA21, easing ADX and negative DI, and a higher low—not momentum recovery alone. In this exploratory BTC study, the first two conditions had a 72-hour held-low rate of 83.3% (15/18); adding a higher low gave 100.0% (10/10), versus 60.0% (9/15) for ETH under the same three conditions. BTC's three-condition rate over seven days was 90.0% (9/10). These are small, retrospectively selected samples, not future bottom-confirmation probabilities. Conditions are measured after a twelve-hour wait; held means no two later hourly closes below the fixed low minus 0.1 prior ATR. The combination warrants further study, not a validated buy signal.
Historical study updated
Explore data products · Check Market Events sample · Explore Telegram alerts
Current source-backed snapshot
Which indicators should agree before calling a possible Bitcoin base? Our added exploratory analysis combines price above EMA21, easing ADX and negative DI, then a higher low—all measured at the same twelve-hour decision point. Selected BTC groups show higher short-term held-low rates, but the small samples, weaker ETH replication and seven-day outcomes limit that conclusion. The first tables compare single conditions, combinations and what each extra condition actually excludes. This extension was selected after reviewing the original experiment, not independently validated. A held low is not a profitable trade or proof of a lasting bottom; the original tests and their contradictory evidence remain below.
Input documentation: Binance Spot candle documentation · Binance public historical data · Fidelity RSI guide · CoinNudge bottom experiment · CoinNudge methodology
Observation window: 48h new-low candidates after ≥5% drawdown; 12h decision; 72h primary outcome
Calculation cadence: Auto-refresh about every 86400 seconds
Calculation version: local-bottom-confirmation-1.0
Download current dataset: JSON · CSV Raw values retain the dataset's published precision. Free fair-use limit: 60 requests per minute per IP, shared across all Research JSON and CSV endpoints.
Does price recovery + weaker selling + a higher low identify a stronger base?
The BTC combination is worth investigating, but it is not a confirmed bottom rule. At the same twelve-hour decision point, combine a close above EMA21 with falling ADX and negative DI; then test whether a higher low adds anything. These are distinct measurements, not assumed independent evidence.
BTC full-history 72h held rate: 83.3% (15/18) for EMA21 plus easing, and 100.0% (10/10) after requiring a higher low. The latter becomes 90.0% (9/10) over seven days. ETH's same three-part condition gives 60.0% (9/15) at 72h. Small selected groups and weaker cross-asset replication prevent a general probability claim.
Price recovery
Confirmation close is above EMA21. This is a position relative to the moving average, not a requirement for several closes above it.
Downtrend easing
ADX14 and negative DI14 are both below their candidate-time readings. Falling ADX alone does not indicate a new uptrend.
A shallower second test
The minimum low in waiting hours 7–12 exceeds the minimum in hours 1–6, and that first minimum is at least L−0.1 prior ATR. All three conditions must hold at T.
What “held” means
No two consecutive future hourly closes below that fixed buffered low within 72h or 7d after T. The same complete-path survivor pool is used throughout; this is not a cycle-bottom label or trade win rate.
BTC · individual conditions versus combinations
| Simultaneous conditions | 72h held / N | 7d held / N | 72h / 7d median return | Missing condition |
|---|---|---|---|---|
| EMA21 reclaimed | 62.5% (20/32) | 46.9% (15/32) | -0.3% / -1.2% | 0 |
| ADX and −DI ease | 70.4% (19/27) | 55.6% (15/27) | -0.3% / -0.3% | 0 |
| EMA21 + ADX/−DI easing | 83.3% (15/18) | 61.1% (11/18) | 1.4% / -0.8% | 0 |
| EMA21 + easing + higher low | 100.0% (10/10) | 90.0% (9/10) | 2.2% / 0.7% | 0 |
| EMA21 + easing + VWAP | 82.4% (14/17) | 58.8% (10/17) | 1.2% / -1.3% | 0 |
| EMA21 + easing + MACD | 83.3% (15/18) | 61.1% (11/18) | 1.4% / -0.8% | 0 |
| EMA21 + easing + turnover | 100.0% (3/3) | 100.0% (3/3) | -1.3% / -0.1% | 0 |
Does each additional condition improve the comparison?
| Simultaneous conditions | 72h held / N | 7d held / N | Added condition absent: 72h held / N | Missing condition |
|---|---|---|---|---|
| Within EMA21: add ADX/−DI easing | 83.3% (15/18) | 61.1% (11/18) | 35.7% (5/14) | 0 |
| Within easing: add EMA21 | 83.3% (15/18) | 61.1% (11/18) | 44.4% (4/9) | 0 |
| Within EMA21 + easing: add higher low | 100.0% (10/10) | 90.0% (9/10) | 62.5% (5/8) | 0 |
Read both branches of the conditional comparison. For example, “add a higher low” compares events with versus without it only inside the EMA21-plus-easing group. Comparing a nested combination with its whole parent group is not an independent control test. Adding MACD or turnover is also disclosed above, including redundant and tiny selections.
ETH · same rules, separate observations
| Simultaneous conditions | 72h held / N | 7d held / N | 72h / 7d median return | Missing condition |
|---|---|---|---|---|
| EMA21 reclaimed | 59.5% (25/42) | 50.0% (21/42) | 0.3% / 0.6% | 0 |
| ADX and −DI ease | 55.6% (25/45) | 40.0% (18/45) | 0.3% / 1.5% | 0 |
| EMA21 + ADX/−DI easing | 57.7% (15/26) | 42.3% (11/26) | 0.3% / 1.4% | 0 |
| EMA21 + easing + higher low | 60.0% (9/15) | 40.0% (6/15) | -0.1% / 1.3% | 0 |
| EMA21 + easing + VWAP | 62.5% (15/24) | 45.8% (11/24) | 0.4% / 1.4% | 0 |
| EMA21 + easing + MACD | 60.0% (15/25) | 44.0% (11/25) | 0.3% / 1.5% | 0 |
| EMA21 + easing + turnover | 100.0% (2/2) | 50.0% (1/2) | 3.6% / -4.0% | 0 |
Recent BTC and ETH checks, development periods and ETH incremental evidence
BTCUSDT · later historical period
| Simultaneous conditions | 72h held / N | 7d held / N | 72h / 7d median return | Missing condition |
|---|---|---|---|---|
| EMA21 reclaimed | 60.0% (3/5) | 40.0% (2/5) | 0.7% / 3.8% | 0 |
| ADX and −DI ease | 40.0% (2/5) | 20.0% (1/5) | -3.4% / -1.8% | 0 |
| EMA21 + ADX/−DI easing | 100.0% (2/2) | 50.0% (1/2) | 2.9% / -2.4% | 0 |
| EMA21 + easing + higher low | 100.0% (1/1) | 100.0% (1/1) | 3.2% / 3.8% | 0 |
| EMA21 + easing + VWAP | 100.0% (2/2) | 50.0% (1/2) | 2.9% / -2.4% | 0 |
| EMA21 + easing + MACD | 100.0% (2/2) | 50.0% (1/2) | 2.9% / -2.4% | 0 |
| EMA21 + easing + turnover | — (0/0) | — (0/0) | — / — | 0 |
BTCUSDT · development period
| Simultaneous conditions | 72h held / N | 7d held / N | 72h / 7d median return | Missing condition |
|---|---|---|---|---|
| EMA21 reclaimed | 63.0% (17/27) | 48.1% (13/27) | -1.3% / -1.3% | 0 |
| ADX and −DI ease | 76.2% (16/21) | 61.9% (13/21) | -0.3% / -0.3% | 0 |
| EMA21 + ADX/−DI easing | 81.3% (13/16) | 62.5% (10/16) | 0.4% / -0.8% | 0 |
| EMA21 + easing + higher low | 100.0% (9/9) | 88.9% (8/9) | 1.9% / -0.1% | 0 |
| EMA21 + easing + VWAP | 80.0% (12/15) | 60.0% (9/15) | -0.3% / -1.3% | 0 |
| EMA21 + easing + MACD | 81.3% (13/16) | 62.5% (10/16) | 0.4% / -0.8% | 0 |
| EMA21 + easing + turnover | 100.0% (3/3) | 100.0% (3/3) | -1.3% / -0.1% | 0 |
ETHUSDT · later historical period
| Simultaneous conditions | 72h held / N | 7d held / N | 72h / 7d median return | Missing condition |
|---|---|---|---|---|
| EMA21 reclaimed | 71.4% (5/7) | 28.6% (2/7) | -0.1% / -2.3% | 0 |
| ADX and −DI ease | 61.5% (8/13) | 30.8% (4/13) | -0.3% / 1.6% | 0 |
| EMA21 + ADX/−DI easing | 83.3% (5/6) | 33.3% (2/6) | 0.1% / -1.5% | 0 |
| EMA21 + easing + higher low | 66.7% (2/3) | 0.0% (0/3) | -0.3% / -2.3% | 0 |
| EMA21 + easing + VWAP | 100.0% (5/5) | 40.0% (2/5) | 0.3% / -0.7% | 0 |
| EMA21 + easing + MACD | 83.3% (5/6) | 33.3% (2/6) | 0.1% / -1.5% | 0 |
| EMA21 + easing + turnover | 100.0% (1/1) | 0.0% (0/1) | 3.8% / -9.4% | 0 |
ETHUSDT · development period
| Simultaneous conditions | 72h held / N | 7d held / N | 72h / 7d median return | Missing condition |
|---|---|---|---|---|
| EMA21 reclaimed | 57.1% (20/35) | 54.3% (19/35) | 0.4% / 1.5% | 0 |
| ADX and −DI ease | 53.1% (17/32) | 43.8% (14/32) | 0.7% / 1.4% | 0 |
| EMA21 + ADX/−DI easing | 50.0% (10/20) | 45.0% (9/20) | 0.4% / 1.7% | 0 |
| EMA21 + easing + higher low | 58.3% (7/12) | 50.0% (6/12) | 2.8% / 1.4% | 0 |
| EMA21 + easing + VWAP | 52.6% (10/19) | 47.4% (9/19) | 0.4% / 1.5% | 0 |
| EMA21 + easing + MACD | 52.6% (10/19) | 47.4% (9/19) | 0.4% / 1.9% | 0 |
| EMA21 + easing + turnover | 100.0% (1/1) | 100.0% (1/1) | 3.4% / 1.5% | 0 |
ETH · conditional additions, full history
| Simultaneous conditions | 72h held / N | 7d held / N | Added condition absent: 72h held / N | Missing condition |
|---|---|---|---|---|
| Within EMA21: add ADX/−DI easing | 57.7% (15/26) | 42.3% (11/26) | 62.5% (10/16) | 0 |
| Within easing: add EMA21 | 57.7% (15/26) | 42.3% (11/26) | 52.6% (10/19) | 0 |
| Within EMA21 + easing: add higher low | 60.0% (9/15) | 40.0% (6/15) | 54.5% (6/11) | 0 |
Added after inspecting the original results on September 15, 2026. This extension is hypothesis-generating, not part of the original thirteen predefined tests. Recent periods overlap full history and are retrospective checks, not independent forward validation. All percentages are descriptive; no adjusted significance or confidence interval is claimed for these added combinations. A tiny group with no failures does not establish 100% future reliability.
Reproduce from combination_checks in the JSON/CSV exports, using the original observations' flags and the recorded calendar boundaries. The full formulas and seven selections plus three conditional comparisons are listed in contract.combination_extension. This daily update uses existing archived inputs, not a new per-visitor experiment.
A higher low suggests stabilization—not a confirmed bottom
The original tests below explain why the added combination needs separate scrutiny. They do not overturn the exploratory results above: a single feature can be weak while a selected combination looks better. Neither establishes a durable bottom. Both analyses use observations that survived the same twelve-hour wait; neither buys at a low chosen afterward.
Reading the two table types: the combination tables above show held-low rates (higher means fewer rebreaks). The original tables below show rebreak rates (lower means fewer rebreaks). For the same group and horizon, these add to 100%; never compare them as though they were the same metric. Neither is a trading win rate.
1 · A higher low looked useful in historical BTC
Subsequent 72-hour rebreak: 39.4% (13/33) with the higher-low condition, versus 68.3% (28/41) without it. That is a 28.9-percentage-point historical difference—not a guaranteed improvement on the next trade.
“Higher low” here means the lowest low in waiting hours 7–12 exceeds that in hours 1–6, while the first block stays above the buffered original low. It is not a retrospectively chosen double bottom.
2 · Adding MACD did not add a second confirmation
The full-history higher-low groups contained 33 BTC observations and 44 ETH observations. Check the conditional table below: an added MACD condition with no excluded observations cannot separate stronger from weaker candidates.
In this release, the higher-low and higher-low-plus-MACD groups are identical. Two green indicators therefore do not amount to two independent pieces of evidence. MACD “improvement” means a rising histogram, not necessarily a bullish crossover.
3 · Holding the low was not the same as making money
For the 33 BTC higher-low observations, median price change during the wait was 1.3% from the candidate close; price at confirmation was a median 2.5% above the original low.
After confirmation, median reference return was 0.3% over 72 hours and -1.0% over seven days, before fees or execution effects. Waiting did not establish a profitable entry rule.
How should you use these findings?
Call it a possible base—not a confirmed bottom. Keep the original low fixed, distinguish a brief wick from two completed closes below the buffered low, and check whether extra indicators actually change the sample.
These are monitoring distinctions, not a tested trading strategy. Neither a held low nor a rebound alone proves that another decline is over. The recent-period and ETH checks below are why we do not turn the favorable BTC headline into a buy signal.
Why the favorable BTC percentage is not a bottom-confirmation rule
| Independent reading | With higher low | Without higher low | What this establishes |
|---|---|---|---|
| BTC · full history | 39.4% (13/33) | 68.3% (28/41) | Favorable historical association, not causation |
| BTC · later historical period | 83.3% (5/6) | 62.5% (5/8) | Small sample; the apparent advantage reverses |
| ETH · full history | 50.0% (22/44) | 49.2% (30/61) | The BTC advantage does not replicate |
Read the denominator before the percentage. A confirmed rebreak means two hourly closes below the original candidate low minus 0.1 prior ATR during the NEXT 72 hours after T. One minus this rate is not the chance of a profitable trade. These correlated, observational comparisons have not been jointly matched or adjusted for multiple testing. The later BTC period is a retrospective split, not an untouched prospective test.
Which observations actually enter the experiment?
Binance spot BTCUSDT is primary; ETHUSDT is separate. Research window 2021-09-01 00:00:00 UTC to 2026-09-15 00:00:00 UTC, end exclusive. The original hourly input file includes the preceding warm-up month. No funding, liquidation or OI history has been invented to fill this period.
Among 137 complete BTC paths, 57 failed during the wait and 6 were not above the fixed low at T. Separately, 2 retained paths had gaps and 1 were pending or crossed the observation boundary. None is counted as a successful bottom.
1 · Candidate, known then
A new 48-hour low and at least 5% closing-price drawdown from the previous seven-day high. Freeze L at the candidate low. This is not a trough chosen with hindsight.
2 · A common decision time
Wait twelve completed hours. Compare price, momentum and activity only at T, then measure later outcomes. Report the waiting price separately.
3 · What counts as failure?
Two future hourly closes below L−0.1 prior ATR. One wick is different. A later reclaim does not erase the earlier rebreak.
4 · What counts as evidence?
A difference that can be reproduced, survives cross-checks and adds more than a correlated indicator. A small favorable group alone is not enough.
Which single indicators and combinations changed the rebreak rate?
| Condition / comparison | With: failures / N | Without: failures / N | Difference · pp | 95% interval · pp | Evidence limit |
|---|---|---|---|---|---|
| Two-block higher-low proxy | 39.4% · 13/33 | 68.3% · 28/41 | -28.9 | -50.5 to -7.6 | Unadjusted · noncausal |
| RSI rises and recovers above 30 | 55.6% · 40/72 | 50.0% · 1/2 | 5.6 | Withheld | Small sample · exploratory |
| MACD histogram improves | 54.2% · 39/72 | 100.0% · 2/2 | -45.8 | Withheld | Small sample · exploratory |
| Price/RSI two-point divergence proxy | 53.8% · 7/13 | 55.7% · 34/61 | -1.9 | Withheld | Small sample · exploratory |
| At least 1.5× sustained turnover | 41.7% · 5/12 | 58.1% · 36/62 | -16.4 | Withheld | Small sample · exploratory |
| At least 55% valid active-buy share | 0.0% · 0/1 | 56.2% · 41/73 | -56.2 | Withheld | Small sample · exploratory |
| Close above EMA21 | 37.5% · 12/32 | 69.0% · 29/42 | -31.5 | -54.2 to -8.2 | Unadjusted · noncausal |
| Close above UTC-session VWAP | 48.1% · 26/54 | 75.0% · 15/20 | -26.9 | -50.0 to -3.2 | Unadjusted · noncausal |
| ADX and negative DI both ease | 29.6% · 8/27 | 70.2% · 33/47 | -40.6 | -61.6 to -17.4 | Unadjusted · noncausal |
| Higher low + improving MACD | 39.4% · 13/33 | 68.3% · 28/41 | -28.9 | -50.5 to -7.6 | Unadjusted · noncausal |
| Improving MACD + sustained turnover | 36.4% · 4/11 | 58.7% · 37/63 | -22.4 | Withheld | Small sample · exploratory |
| Higher low + active buying | 0.0% · 0/1 | 56.2% · 41/73 | -56.2 | Withheld | Small sample · exploratory |
| Higher low + MACD + sustained turnover | 28.6% · 2/7 | 58.2% · 39/67 | -29.6 | Withheld | Small sample · exploratory |
All nine single conditions and four combinations are shown, including weak, redundant and tiny groups. “Withheld” means insufficient group size or valid resamples, not a zero-width interval. The predeclared main interval is a 14-calendar-day circular block bootstrap with 2,000 paired draws, at least 20 events in each group and 1,900 finite draws.
Does an extra indicator actually add anything?
These comparisons begin inside an already-qualified component group. A zero negative group means the added condition cannot discriminate in this sample; a small group is not proof of a better strategy.
| Condition / comparison | With: failures / N | Without: failures / N | Difference · pp | 95% interval · pp | Evidence limit |
|---|---|---|---|---|---|
| Higher low + improving MACD Within price_hold=true | 39.4% · 13/33 | — · 0/0 | — | Withheld | No comparison group |
| Improving MACD + sustained turnover Within macd_repair=true | 36.4% · 4/11 | 57.4% · 35/61 | -21.0 | Withheld | Small sample · exploratory |
| Higher low + MACD + sustained turnover Within price_macd=true | 28.6% · 2/7 | 42.3% · 11/26 | -13.7 | Withheld | Small sample · exploratory |
ETH incremental comparisons
| Condition / comparison | With: failures / N | Without: failures / N | Difference · pp | 95% interval · pp | Evidence limit |
|---|---|---|---|---|---|
| Higher low + improving MACD Within price_hold=true | 50.0% · 22/44 | — · 0/0 | — | Withheld | No comparison group |
| Improving MACD + sustained turnover Within macd_repair=true | 38.9% · 7/18 | 50.6% · 43/85 | -11.7 | Withheld | Small sample · exploratory |
| Higher low + MACD + sustained turnover Within price_macd=true | 25.0% · 1/4 | 52.5% · 21/40 | -27.5 | Withheld | Small sample · exploratory |
Price-based conditions can reflect the same rebound. If higher-low events and higher-low-plus-MACD events are identical, they are the same evidence, not two independent confirmations. The three-way condition also requires sufficient samples before its apparently better percentage is meaningful.
What did waiting cost, and what happened over different horizons?
All future returns use T close as a reference, not a guaranteed fill. The first two columns distinguish waiting from the candidate close versus measuring from the candidate low, which may not have been executable. MFE/MAE are median maximum favorable/adverse price excursions, not strategy gains or stop losses.
| Condition present | N | Wait from t0 / from L | 24h / 72h / 7d rebreak | 72h median return | 72h MFE / MAE | Absent-condition rebound |
|---|---|---|---|---|---|---|
| Two-block higher-low proxy | 33 | 1.3% / 2.5% | 15.2% / 39.4% / 48.5% | 0.3% | 2.6% / -3.0% | 35/41 |
| RSI rises and recovers above 30 | 72 | 1.2% / 1.9% | 23.6% / 55.6% / 63.9% | -0.3% | 2.2% / -3.1% | 1/2 |
| MACD histogram improves | 72 | 1.2% / 1.9% | 22.2% / 54.2% / 62.5% | -0.6% | 2.1% / -3.2% | 2/2 |
| Price/RSI two-point divergence proxy | 13 | 1.7% / 2.1% | 38.5% / 53.8% / 61.5% | -1.7% | 2.1% / -3.6% | 53/61 |
| At least 1.5× sustained turnover | 12 | 1.9% / 2.6% | 16.7% / 41.7% / 50.0% | 0.3% | 3.0% / -3.8% | 54/62 |
| At least 55% valid active-buy share | 1 | 1.0% / 1.1% | 0.0% / 0.0% / 0.0% | 2.4% | 2.5% / -1.2% | 65/73 |
| Close above EMA21 | 32 | 2.5% / 2.9% | 9.4% / 37.5% / 53.1% | -0.3% | 2.6% / -3.2% | 34/42 |
| Close above UTC-session VWAP | 54 | 1.6% / 2.1% | 20.4% / 48.1% / 57.4% | -0.3% | 2.5% / -3.0% | 16/20 |
| ADX and negative DI both ease | 27 | 2.0% / 2.5% | 3.7% / 29.6% / 44.4% | -0.3% | 3.1% / -3.0% | 39/47 |
| Higher low + improving MACD | 33 | 1.3% / 2.5% | 15.2% / 39.4% / 48.5% | 0.3% | 2.6% / -3.0% | 35/41 |
| Improving MACD + sustained turnover | 11 | 2.0% / 2.5% | 9.1% / 36.4% / 45.5% | 0.1% | 2.0% / -3.7% | 55/63 |
| Higher low + active buying | 1 | 1.0% / 1.1% | 0.0% / 0.0% / 0.0% | 2.4% | 2.5% / -1.2% | 65/73 |
| Higher low + MACD + sustained turnover | 7 | 2.6% / 3.6% | 14.3% / 28.6% / 42.9% | -1.4% | 1.1% / -3.9% | 59/67 |
Rebound means at least one later close reached L+2 prior ATR in 72h, even if the path also rebroke later. It measures an opportunity missed by a condition, not missed profitable trades. Horizon rates and return medians are descriptive; the displayed bootstrap interval applies only to the 72h failure-rate difference.
Do recent BTC history and ETH support the same interpretation?
The later historical period is 2025-09-01 to 2026-09-15 UTC, excluding 180h paths crossing its edges. This retrospective split was not an untouched external test. Development and forward groups remain separate; forward begins 2026-09-16 and requires full paths.
BTC · later historical period
| Condition / comparison | With: failures / N | Without: failures / N | Difference · pp | 95% interval · pp | Evidence limit |
|---|---|---|---|---|---|
| Two-block higher-low proxy | 83.3% · 5/6 | 62.5% · 5/8 | 20.8 | Withheld | Small sample · exploratory |
| RSI rises and recovers above 30 | 71.4% · 10/14 | — · 0/0 | — | Withheld | No comparison group |
| MACD histogram improves | 69.2% · 9/13 | 100.0% · 1/1 | -30.8 | Withheld | Small sample · exploratory |
| Price/RSI two-point divergence proxy | 100.0% · 3/3 | 63.6% · 7/11 | 36.4 | Withheld | Small sample · exploratory |
| At least 1.5× sustained turnover | 100.0% · 2/2 | 66.7% · 8/12 | 33.3 | Withheld | Small sample · exploratory |
| At least 55% valid active-buy share | — · 0/0 | 71.4% · 10/14 | — | Withheld | No comparison group |
| Close above EMA21 | 40.0% · 2/5 | 88.9% · 8/9 | -48.9 | Withheld | Small sample · exploratory |
| Close above UTC-session VWAP | 50.0% · 4/8 | 100.0% · 6/6 | -50.0 | Withheld | Small sample · exploratory |
| ADX and negative DI both ease | 60.0% · 3/5 | 77.8% · 7/9 | -17.8 | Withheld | Small sample · exploratory |
| Higher low + improving MACD | 83.3% · 5/6 | 62.5% · 5/8 | 20.8 | Withheld | Small sample · exploratory |
| Improving MACD + sustained turnover | 100.0% · 2/2 | 66.7% · 8/12 | 33.3 | Withheld | Small sample · exploratory |
| Higher low + active buying | — · 0/0 | 71.4% · 10/14 | — | Withheld | No comparison group |
| Higher low + MACD + sustained turnover | 100.0% · 1/1 | 69.2% · 9/13 | 30.8 | Withheld | Small sample · exploratory |
ETH · full history, not pooled with BTC
| Condition / comparison | With: failures / N | Without: failures / N | Difference · pp | 95% interval · pp | Evidence limit |
|---|---|---|---|---|---|
| Two-block higher-low proxy | 50.0% · 22/44 | 49.2% · 30/61 | 0.8 | -17.9 to 19.6 | Unadjusted · noncausal |
| RSI rises and recovers above 30 | 50.0% · 50/100 | 40.0% · 2/5 | 10.0 | Withheld | Small sample · exploratory |
| MACD histogram improves | 48.5% · 50/103 | 100.0% · 2/2 | -51.5 | Withheld | Small sample · exploratory |
| Price/RSI two-point divergence proxy | 47.1% · 8/17 | 50.0% · 44/88 | -2.9 | Withheld | Small sample · exploratory |
| At least 1.5× sustained turnover | 38.9% · 7/18 | 51.7% · 45/87 | -12.8 | Withheld | Small sample · exploratory |
| At least 55% valid active-buy share | 40.0% · 2/5 | 50.0% · 50/100 | -10.0 | Withheld | Small sample · exploratory |
| Close above EMA21 | 40.5% · 17/42 | 55.6% · 35/63 | -15.1 | -33.8 to 5.2 | Unadjusted · noncausal |
| Close above UTC-session VWAP | 47.7% · 31/65 | 52.5% · 21/40 | -4.8 | -25.2 to 16.1 | Unadjusted · noncausal |
| ADX and negative DI both ease | 44.4% · 20/45 | 53.3% · 32/60 | -8.9 | -28.5 to 10.7 | Unadjusted · noncausal |
| Higher low + improving MACD | 50.0% · 22/44 | 49.2% · 30/61 | 0.8 | -17.9 to 19.6 | Unadjusted · noncausal |
| Improving MACD + sustained turnover | 38.9% · 7/18 | 51.7% · 45/87 | -12.8 | Withheld | Small sample · exploratory |
| Higher low + active buying | 40.0% · 2/5 | 50.0% · 50/100 | -10.0 | Withheld | Small sample · exploratory |
| Higher low + MACD + sustained turnover | 25.0% · 1/4 | 50.5% · 51/101 | -25.5 | Withheld | Small sample · exploratory |
ETH later period and both development periods
ETH · later period
| Condition / comparison | With: failures / N | Without: failures / N | Difference · pp | 95% interval · pp | Evidence limit |
|---|---|---|---|---|---|
| Two-block higher-low proxy | 55.6% · 5/9 | 53.8% · 7/13 | 1.7 | Withheld | Small sample · exploratory |
| RSI rises and recovers above 30 | 57.1% · 12/21 | 0.0% · 0/1 | 57.1 | Withheld | Small sample · exploratory |
| MACD histogram improves | 52.4% · 11/21 | 100.0% · 1/1 | -47.6 | Withheld | Small sample · exploratory |
| Price/RSI two-point divergence proxy | 40.0% · 2/5 | 58.8% · 10/17 | -18.8 | Withheld | Small sample · exploratory |
| At least 1.5× sustained turnover | 75.0% · 3/4 | 50.0% · 9/18 | 25.0 | Withheld | Small sample · exploratory |
| At least 55% valid active-buy share | 33.3% · 1/3 | 57.9% · 11/19 | -24.6 | Withheld | Small sample · exploratory |
| Close above EMA21 | 28.6% · 2/7 | 66.7% · 10/15 | -38.1 | Withheld | Small sample · exploratory |
| Close above UTC-session VWAP | 64.3% · 9/14 | 37.5% · 3/8 | 26.8 | Withheld | Small sample · exploratory |
| ADX and negative DI both ease | 38.5% · 5/13 | 77.8% · 7/9 | -39.3 | Withheld | Small sample · exploratory |
| Higher low + improving MACD | 55.6% · 5/9 | 53.8% · 7/13 | 1.7 | Withheld | Small sample · exploratory |
| Improving MACD + sustained turnover | 75.0% · 3/4 | 50.0% · 9/18 | 25.0 | Withheld | Small sample · exploratory |
| Higher low + active buying | 33.3% · 1/3 | 57.9% · 11/19 | -24.6 | Withheld | Small sample · exploratory |
| Higher low + MACD + sustained turnover | 100.0% · 1/1 | 52.4% · 11/21 | 47.6 | Withheld | Small sample · exploratory |
BTC · development
| Condition / comparison | With: failures / N | Without: failures / N | Difference · pp | 95% interval · pp | Evidence limit |
|---|---|---|---|---|---|
| Two-block higher-low proxy | 30.8% · 8/26 | 69.7% · 23/33 | -38.9 | -62.1 to -15.1 | Unadjusted · noncausal |
| RSI rises and recovers above 30 | 52.6% · 30/57 | 50.0% · 1/2 | 2.6 | Withheld | Small sample · exploratory |
| MACD histogram improves | 51.7% · 30/58 | 100.0% · 1/1 | -48.3 | Withheld | Small sample · exploratory |
| Price/RSI two-point divergence proxy | 44.4% · 4/9 | 54.0% · 27/50 | -9.6 | Withheld | Small sample · exploratory |
| At least 1.5× sustained turnover | 30.0% · 3/10 | 57.1% · 28/49 | -27.1 | Withheld | Small sample · exploratory |
| At least 55% valid active-buy share | — · 0/0 | 52.5% · 31/59 | — | Withheld | No comparison group |
| Close above EMA21 | 37.0% · 10/27 | 65.6% · 21/32 | -28.6 | -53.1 to -3.6 | Unadjusted · noncausal |
| Close above UTC-session VWAP | 48.9% · 22/45 | 64.3% · 9/14 | -15.4 | Withheld | Small sample · exploratory |
| ADX and negative DI both ease | 23.8% · 5/21 | 68.4% · 26/38 | -44.6 | -67.0 to -18.7 | Unadjusted · noncausal |
| Higher low + improving MACD | 30.8% · 8/26 | 69.7% · 23/33 | -38.9 | -62.1 to -15.1 | Unadjusted · noncausal |
| Improving MACD + sustained turnover | 22.2% · 2/9 | 58.0% · 29/50 | -35.8 | Withheld | Small sample · exploratory |
| Higher low + active buying | — · 0/0 | 52.5% · 31/59 | — | Withheld | No comparison group |
| Higher low + MACD + sustained turnover | 16.7% · 1/6 | 56.6% · 30/53 | -39.9 | Withheld | Small sample · exploratory |
ETH · development
| Condition / comparison | With: failures / N | Without: failures / N | Difference · pp | 95% interval · pp | Evidence limit |
|---|---|---|---|---|---|
| Two-block higher-low proxy | 48.6% · 17/35 | 47.9% · 23/48 | 0.7 | -21.3 to 21.3 | Unadjusted · noncausal |
| RSI rises and recovers above 30 | 48.1% · 38/79 | 50.0% · 2/4 | -1.9 | Withheld | Small sample · exploratory |
| MACD histogram improves | 47.6% · 39/82 | 100.0% · 1/1 | -52.4 | Withheld | Small sample · exploratory |
| Price/RSI two-point divergence proxy | 50.0% · 6/12 | 47.9% · 34/71 | 2.1 | Withheld | Small sample · exploratory |
| At least 1.5× sustained turnover | 28.6% · 4/14 | 52.2% · 36/69 | -23.6 | Withheld | Small sample · exploratory |
| At least 55% valid active-buy share | 50.0% · 1/2 | 48.1% · 39/81 | 1.9 | Withheld | Small sample · exploratory |
| Close above EMA21 | 42.9% · 15/35 | 52.1% · 25/48 | -9.2 | -30.8 to 12.6 | Unadjusted · noncausal |
| Close above UTC-session VWAP | 43.1% · 22/51 | 56.3% · 18/32 | -13.1 | -35.2 to 8.4 | Unadjusted · noncausal |
| ADX and negative DI both ease | 46.9% · 15/32 | 49.0% · 25/51 | -2.1 | -24.9 to 20.0 | Unadjusted · noncausal |
| Higher low + improving MACD | 48.6% · 17/35 | 47.9% · 23/48 | 0.7 | -21.3 to 21.3 | Unadjusted · noncausal |
| Improving MACD + sustained turnover | 28.6% · 4/14 | 52.2% · 36/69 | -23.6 | Withheld | Small sample · exploratory |
| Higher low + active buying | 50.0% · 1/2 | 48.1% · 39/81 | 1.9 | Withheld | Small sample · exploratory |
| Higher low + MACD + sustained turnover | 0.0% · 0/3 | 50.0% · 40/80 | -50.0 | Withheld | Small sample · exploratory |
Could distance from the low explain the result?
Yes: a price already farther from the fixed barrier has more room before it can rebreak. The following are separate marginal strata, not joint matching or causal adjustment. Empty/sparse cells remain explicit. The full export includes all conditions and both assets.
Distance at T in prior ATR units
| Bin | Condition | With: failures / N | Without: failures / N |
|---|---|---|---|
| [0,1) | Two-block higher-low proxy | 100.0% · 1/1 | 100.0% · 11/11 |
| [0,1) | Close above EMA21 | — · 0/0 | 100.0% · 12/12 |
| [0,1) | ADX and negative DI both ease | — · 0/0 | 100.0% · 12/12 |
| [0,1) | Higher low + MACD + sustained turnover | — · 0/0 | 100.0% · 12/12 |
| [1,2) | Two-block higher-low proxy | 66.7% · 4/6 | 75.0% · 9/12 |
| [1,2) | Close above EMA21 | 0.0% · 0/1 | 76.5% · 13/17 |
| [1,2) | ADX and negative DI both ease | 50.0% · 2/4 | 78.6% · 11/14 |
| [1,2) | Higher low + MACD + sustained turnover | — · 0/0 | 72.2% · 13/18 |
| [2,4) | Two-block higher-low proxy | 40.0% · 6/15 | 50.0% · 6/12 |
| [2,4) | Close above EMA21 | 47.1% · 8/17 | 40.0% · 4/10 |
| [2,4) | ADX and negative DI both ease | 30.8% · 4/13 | 57.1% · 8/14 |
| [2,4) | Higher low + MACD + sustained turnover | 100.0% · 1/1 | 42.3% · 11/26 |
| [4,inf) | Two-block higher-low proxy | 18.2% · 2/11 | 33.3% · 2/6 |
| [4,inf) | Close above EMA21 | 28.6% · 4/14 | 0.0% · 0/3 |
| [4,inf) | ADX and negative DI both ease | 20.0% · 2/10 | 28.6% · 2/7 |
| [4,inf) | Higher low + MACD + sustained turnover | 16.7% · 1/6 | 27.3% · 3/11 |
Prior drawdown, percent
| Bin | Condition | With: failures / N | Without: failures / N |
|---|---|---|---|
| [5,10) | Two-block higher-low proxy | 38.5% · 10/26 | 67.6% · 25/37 |
| [5,10) | Close above EMA21 | 42.3% · 11/26 | 64.9% · 24/37 |
| [5,10) | ADX and negative DI both ease | 22.2% · 4/18 | 68.9% · 31/45 |
| [5,10) | Higher low + MACD + sustained turnover | 28.6% · 2/7 | 58.9% · 33/56 |
| [10,20) | Two-block higher-low proxy | 42.9% · 3/7 | 66.7% · 2/3 |
| [10,20) | Close above EMA21 | 16.7% · 1/6 | 100.0% · 4/4 |
| [10,20) | ADX and negative DI both ease | 44.4% · 4/9 | 100.0% · 1/1 |
| [10,20) | Higher low + MACD + sustained turnover | — · 0/0 | 50.0% · 5/10 |
| [20,inf) | Two-block higher-low proxy | — · 0/0 | 100.0% · 1/1 |
| [20,inf) | Close above EMA21 | — · 0/0 | 100.0% · 1/1 |
| [20,inf) | ADX and negative DI both ease | — · 0/0 | 100.0% · 1/1 |
| [20,inf) | Higher low + MACD + sustained turnover | — · 0/0 | 100.0% · 1/1 |
Prior ATR/close, percent
| Bin | Condition | With: failures / N | Without: failures / N |
|---|---|---|---|
| [0,1) | Two-block higher-low proxy | 36.4% · 8/22 | 69.4% · 25/36 |
| [0,1) | Close above EMA21 | 45.8% · 11/24 | 64.7% · 22/34 |
| [0,1) | ADX and negative DI both ease | 26.3% · 5/19 | 71.8% · 28/39 |
| [0,1) | Higher low + MACD + sustained turnover | 33.3% · 2/6 | 59.6% · 31/52 |
| [1,inf) | Two-block higher-low proxy | 45.5% · 5/11 | 60.0% · 3/5 |
| [1,inf) | Close above EMA21 | 12.5% · 1/8 | 87.5% · 7/8 |
| [1,inf) | ADX and negative DI both ease | 37.5% · 3/8 | 62.5% · 5/8 |
| [1,inf) | Higher low + MACD + sustained turnover | 0.0% · 0/1 | 53.3% · 8/15 |
Sign of prior seven-day return
| Bin | Condition | With: failures / N | Without: failures / N |
|---|---|---|---|
| [-inf,0) | Two-block higher-low proxy | 50.0% · 9/18 | 65.2% · 15/23 |
| [-inf,0) | Close above EMA21 | 47.4% · 9/19 | 68.2% · 15/22 |
| [-inf,0) | ADX and negative DI both ease | 31.3% · 5/16 | 76.0% · 19/25 |
| [-inf,0) | Higher low + MACD + sustained turnover | 50.0% · 1/2 | 59.0% · 23/39 |
| [0,inf) | Two-block higher-low proxy | 26.7% · 4/15 | 72.2% · 13/18 |
| [0,inf) | Close above EMA21 | 23.1% · 3/13 | 70.0% · 14/20 |
| [0,inf) | ADX and negative DI both ease | 27.3% · 3/11 | 63.6% · 14/22 |
| [0,inf) | Higher low + MACD + sustained turnover | 20.0% · 1/5 | 57.1% · 16/28 |
What do actual held and failed paths look like?
Cases are selected mechanically nearest their class median 72h return, earliest event on ties. They illustrate different outcomes rather than estimate their probability.
Held without a later rebreak
At T: price hold true, MACD repair true, sustained turnover false. Later 72h: rebreak false, reference return 1.9%. The later outcome was not known at T.
Rebroke and later reclaimed
At T: price hold false, MACD repair true, sustained turnover false. Later 72h: rebreak true, reference return -1.1%. The later outcome was not known at T.
Rebroke without a reclaim
At T: price hold true, MACD repair true, sustained turnover false. Later 72h: rebreak true, reference return -4.4%. The later outcome was not known at T.
Can you reproduce the experiment from the actual inputs?
| Market | Present / expected 1h bars | Missing / duplicates | Valid flow bars |
|---|---|---|---|
| BTCUSDT | 44157 / 44160 | 3 / 0 | 44156 |
| ETHUSDT | 44157 / 44160 | 3 / 0 | 44156 |
The downloadable inputs include hourly open/high/low/close, base volume v, quote volume qv, taker_buy_qv and taker_buy_valid. The research export includes observations, raw_event_ledger, incremental, strata, funnels and coverage_items. Observations expose reference_low, atr_prior, trigger/decision times, raw features, condition flags and per-horizon statuses. Null is unavailable, never zero selling or zero return. See contract.rules for exact conditions and contract.seed for reproducible random draws.
Release 20260915-b4e7dbec2a00
Input SHA-256 8875ea3b255387b25fed97f9864b0f2e63f6cd4b386b66f5972270c9d44d4241
Calculation local-bottom-confirmation-1.0
The full input file is larger than the summary. The verifier checks candidate selection, spacing, frozen levels, timing, outcomes, raw feature formulas, condition flags, comparison counts and rates; optional bootstrap verifies 72h intervals. It does not validate statistical suitability, legal rights or profitable execution. Historical reconstruction is not original point-in-time alert delivery.
Local low candidate study, not cycle-bottom prediction. Survivor conditioning and observational groups; no joint matched controls, fees, fills or guaranteed performance.
How to read this page
- A candidate is identified using only data available then—not chosen because it later became the lowest point.
- The main endpoint is a confirmed low break within 72 hours after the twelve-hour decision point.
- Check BTC recent history and ETH before interpreting a favorable full-period percentage.
- An extra indicator adds no information if it selects exactly the same events.
- One minus the rebreak rate is not a profitable-trade probability.
What can this page tell you quickly?
- Best for
- Traders evaluating whether to wait for a local-low confirmation, and researchers testing that belief.
- What you receive
- Original event evidence, full candidate exclusions, single and combined conditions, cross-checks and downloadable inputs.
- What it cannot answer
- Whether the current price is the final cycle bottom, or whether a particular trade will make money.
- Observation schedule
- One complete-UTC-day research release; no per-visitor upstream backtest.
- Decision boundary
- Use supported alerts as monitoring context; this experimental protocol is not a promised Telegram signal.
How can we test a bottom without choosing the bottom afterward?
We start with a measurable decline, not a visually satisfying trough selected after a rally. A completed hourly candle qualifies when its low is strictly below every low in the preceding 48 hours and its close is at least 5% below the highest high of the preceding 168 hours. Neither reference window includes the candidate candle. These fixed thresholds describe a specific local-low question; they are not optimized parameters or universal market-bottom definitions.
The earliest qualifying candidate is retained. Other qualifying closes during the next 180 hours are recorded but excluded with the retained observation's ID. This spacing prevents the same hourly decline from being counted as dozens of independent opportunities and accommodates a twelve-hour wait plus seven days of follow-up. It also misses some distinct moves within that interval; the suppressed ledger lets a researcher inspect this trade-off.
At that first close we freeze the candle's low L and the previous candle's Wilder ATR14, A. Later lows cannot move this reference. We do not replace an inconvenient candidate with a lower price found later. A small buffer b=0.1A distinguishes a defined break from a marginal touch. This is a local barrier experiment, not a claim that L is the actual bottom.
Verify: Binance hourly candle fields · Full bottom-study contract and ledger
When is confirmation actually knowable?
Every comparison is made twelve completed hours after the candidate close, at T. A signal based on those twelve hours cannot be treated as information known at the original low. All later reference returns therefore start at T, while the price change during the wait is reported separately. There is no simulated entry fill at the candidate low.
The main pool includes only candidates whose T close is above L and which did not already show two consecutive closes below L-b during the wait. This conditioning is material: many declining paths have already failed before the comparison begins. The funnel publishes those early failures and candidates not recovered at T. The resulting percentages apply to survivors, not every oversold market or every new low.
To keep 24-hour, 72-hour and 168-hour comparisons on the same events, the main pool requires the complete 180-hour path from the trigger. Pending and gap-affected events remain in the export but do not become zero returns or successful bottoms. A newly matured 72-hour result can remain outside this common pool until the seven-day path is complete.
What do higher lows, RSI and MACD actually confirm here?
Our price-hold test asks whether the lowest price in hours 7–12 is above the lowest price in hours 1–6, with the first block no lower than L-b. This is a two-block higher-low proxy. It does not inspect every swing or prove a classic double bottom. A sharp recovery can satisfy it without a clean retest; the name and raw block minima make that limitation explicit.
RSI repair means RSI14 at T exceeds its trigger value and is above 30. It need not cross 30 from below. MACD repair means the 12/26/9 histogram improves between the trigger and T; it can remain negative. These are deliberately different concepts from an RSI buy command or MACD crossover. If nearly every survivor has improving momentum, that condition has little ability to discriminate within this survivor pool.
The divergence test is also narrowly defined: the trigger close is below the lowest close of the preceding 48 hours, but trigger RSI is higher than RSI at that earlier closing low. Ties use the earliest hour. This is a two-point divergence proxy, not a confirmed pivot divergence. A true pivot algorithm requiring candles on both sides would only become knowable later and would need its own timing protocol.
ADX easing requires both ADX14 and negative DI to decline from the trigger to T. It describes less intense downside trend evidence, not guaranteed buying pressure. EMA21 and UTC-session VWAP are separate location tests. The full comparison includes unfavorable or sparse results rather than listing these indicators as established confirmation tools.
Verify: Fidelity RSI definition and trend limitations · Exact study indicator rules
Does adding volume or active buying make confirmation stronger?
Sustained volume means the mean hourly quote turnover over the twelve-hour wait is at least 1.5 times the mean over the 24 hours before the trigger. Quote turnover is a measure of trading activity, not new capital entering the market. The active-buy test instead uses valid taker-buy quote turnover divided by total quote turnover across all twelve waiting hours, with a 55% threshold. Missing or invalid flow data makes that feature unavailable.
The original predefined experiment tests four combinations: price hold plus MACD, MACD plus turnover, price hold plus active buying, and price/MACD/turnover together. Its intervals require at least 20 observations per group and sufficient finite bootstrap draws. Separately, the post-hoc extension above tests seven selections centered on EMA21 and ADX/negative-DI easing, with three conditional additions. It reports descriptive counts, not inferential intervals. A stricter selection can look better simply because very few observations remain.
The incremental test asks a more useful question than comparing unrelated percentages: among events already showing a price hold, does MACD separate their later outcomes? Among MACD-repair events, does volume separate them? Among price-plus-MACD events, does adding volume help? If the added condition is present in every base event, there is no negative group and no measurable extra discrimination in that sample. This is an informative null result, not a reason to invent a combined confidence score.
Verify: Incremental comparison rows · Related active-buy data research
Does the result survive a different period, asset or distance from the low?
BTC is the primary study; ETH repeats the same rules separately. We also split history at September 1, 2025 and exclude any 180-hour event path crossing a period boundary. The later historical period ends before September 15, 2026; forward observations start September 16. This is a retrospective chronological check, not proof that nobody had inspected those historical markets before the protocol was written.
A confirming price already farther above L is mechanically less likely to cross the fixed barrier again. Price-based indicators may partly restate this distance rather than add independent information. We show separate strata for distance in prior ATR units, prior drawdown, prior volatility and the sign of the preceding seven-day return. These are marginal descriptive checks, not jointly matched controls, and some cells are tiny.
The principal 72-hour difference interval uses 2, 000 circular resamples of 14 consecutive UTC calendar days, including days with no event. Both groups share each draw. The interval is withheld below 20 observations per group or 1, 900 finite draws. It does not adjust for testing multiple correlated conditions. A favorable full-history interval therefore cannot override a weak recent period, failed ETH replication or an untestably sparse combination.
Verify: Study contract, seeds and period boundaries · CoinNudge methodology
How can you use this evidence without treating it as a buy signal?
For an individual trader, first separate an observed candidate from an established outcome. Write down the fixed low, what evidence is available now, what would invalidate the idea and how much price has already moved while you waited. A condition absent from the chart does not mean the market cannot rally; absent-condition rebound counts show that missed-opportunity trade-off.
CoinNudge's Telegram service can help you monitor its published price and technical conditions. Review the actual alert list and examples before choosing a plan. This twelve-hour bottom protocol is research, not a newly promised alert category, automatic trade or guaranteed bottom warning.
For a professional researcher, save the immutable release and hashed hourly inputs, then reconstruct candidates, exclusions, features and later outcomes. Keep future labels out of model inputs. If you need a joint matched study or a different timeframe, design it separately rather than silently tuning this published protocol. Public study downloads do not imply that every historical field is included in a paid data package; inspect product coverage and sample delivery first.
Verify: Preview supported Telegram alerts · Compare subscription plans · Research data products · Inspect data samples
What is measured, and what is not?
| Measured claim | Evidence on this page | Boundary |
|---|---|---|
| A candidate is not selected because it became a later low. | Preceding 48h/168h inputs, frozen L and a complete candidate ledger with 180h spacing. | The rule is one local-low definition, not all chart bottoms. |
| Combining indicators need not add evidence. | Public component-conditioned incremental comparisons show when an added condition has no negative group. | No joint matching or independent predictive advantage is established. |
| Waiting changes both the population and reference price. | Early exclusions, T=t 0+12h, wait-price change and T-based returns are public. | These are reference-price observations without execution costs. |
| The primary result is reproducible. | Frozen hourly CSV hash, immutable JSON, observation values, summaries and a verification example. | Retrospective history is not original locally observed PIT alert evidence. |
Take the next step with this evidence.
Reading about has bitcoin bottomed—or is this just a relief rally?? Choose ongoing notifications or a dataset you can inspect in your own research workflow.
Receive qualifying events in Telegram.
Use the ready-made alert checklist, message explanations and daily briefing. Check the supported categories and coins before subscribing; every Research page is not a separate alert category.
Check the alert list · View alert plans
Inspect the data before choosing a plan.
Market Events v1 connects recorded events and comparison rows with context, later outcomes and quality flags. Paid access provides customer API keys and daily JSON, CSV and Parquet files.
Check Market Events sample · Explore the full data catalog · Compare data plans
Researcher and Research Desk currently include Market Events v1 only, for its specified pairs and published history. This page's full dataset is not automatically included. Other data families require coverage, date, field and source-use confirmation before purchase. Telegram Pro and data access are separate subscriptions.
Method and data boundary
local-bottom-confirmation-1.0. Binance BTCUSDT and ETHUSDT spot 1h historical replay from 2021-09-01 with 31-day warm-up. Candidate low strictly below prior 48h minimum and close at least 5% below prior 168h highest high; 200 contiguous preceding hours. First candidate retained, next 180h suppressed with blocking ID. Freeze L=trigger low and b=.1 prior Wilder ATR14. At T=t 0+12h, require close>L and no two consecutive waiting closes<L-b; full 180h contiguous path for all comparison horizons. Rebreak means two consecutive closes<L-b strictly after T in 24/72/168h; main 72h. Thirteen single/combined conditions are declared in contract.rules, not tuned on results. RSI14/ADX14/ATR14 Wilder, EMA21 SMA seed, MACD12/26/9; price hold is a two-block low proxy, divergence is a two-point closing-low proxy, not future-confirmed pivots. Invalid flow/VWAP ->null. Fourteen UTC-calendar-day circular block bootstrap, 2000 shared draws, each group>=20 and>=1900 finite draws, linear 2.5/97.5 percentiles for 72h difference only. xorshift 32 shifts 13, 17, 5; seed first 4 SHA256 bytes big-endian of version|symbol|period|test|14, zero->1; block start state%days. Chronological split 2025-09-01; validation ends 2026-09-15; forward begins 2026-09-16;180h paths cannot cross split. Incremental comparisons condition on a component already being true. Marginal distance/drawdown/volatility/trend strata are descriptive, not joint matching. No causal or multiple-testing-adjusted significance claim. Supplement added 2026-09-15: bottom-combinations-exploratory-1.0, selected after inspecting original results. Seven selections combine EMA21, ADX/negative-DI easing, higher low, VWAP, MACD and turnover; three conditional additions assess retained versus excluded events inside a parent group. All use original eligibility and period boundaries, three-valued AND, held counts/N for72h and168h, and reference-return medians. No inferential intervals or adjusted significance for this post-hoc extension. Full rules in contract.combination_extension and row summaries in combination_checks.
A survivor-conditioned, exploratory historical study of local lows in two Binance spot markets. Recent samples are small; correlated indicators need not add evidence. No joint matched design, execution simulation, leverage model or point-in-time alert-delivery claim.
Sources and verification
- Binance Spot candle documentationHourly price, volume and taker-buy field definitions; not a trading conclusion.
- Binance public historical dataHistorical archive formats and timestamp conventions.
- Fidelity RSI guideIndicator definition and strong-trend limitation; not evidence for our crypto experiment.
- CoinNudge bottom experimentOwn calculated events, exclusions, feature rules, outcomes and uncertainty.
- CoinNudge methodologyGeneral provenance, missingness and research boundaries.
Frequently asked questions
How can I tell whether Bitcoin has bottomed?
Look for agreement between price recovery above EMA21, easing ADX and negative DI, and a higher low. Our post-hoc BTC combination analysis shows promising short-term held-low proportions, but small samples and weaker ETH replication do not establish a reliable bottom rule. Read the current 72-hour and seven-day counts together. All conditions are evaluated after a twelve-hour wait; no indicator proves the final low.
Is a higher low proof of a Bitcoin bottom?
No. Our two-block proxy compares lows in hours 1–6 with hours 7–12 after a candidate. It describes a narrow observed path, not a classic double-bottom guarantee. A favorable full-history rate must also survive cross-checks.
Does RSI below 30 mean it is time to buy Bitcoin?
Not in this experiment. RSI repair is measured twelve hours after a candidate, and oversold conditions can persist. The public table compares actual subsequent outcomes; it does not make RSI an automatic instruction.
Does bullish divergence confirm the bottom?
Not automatically. This experiment tests a two-point price/RSI proxy defined from data already known at the candidate close. It does not claim a pivot-based divergence confirmed by future candles. Sparse samples do not support a precise success probability.
Is using three indicators better than using one?
Not automatically. The added analysis tests EMA21 recovery plus ADX/negative-DI easing, then asks whether a higher low improves selection within that group. Conditional tables compare retained and excluded observations; ETH and seven-day outcomes limit the favorable BTC result. This was selected after inspecting the original tests, so it is exploratory, not independent validation. Adding MACD to an identical group adds no screening.
What does breaking the bottom mean here?
Two consecutive hourly closes below the frozen candidate low minus 0.1 prior ATR after the twelve-hour decision point. A wick, a single close and a later reclaim are different events. The primary horizon is 72 hours, with 24 hours and seven days reported separately.
Does a low rebreak rate mean a profitable strategy?
No. Avoiding the fixed low does not mean price finishes above your entry, and a rebreak can be followed by recovery. Fees, fills, position size and exits are not simulated. Reference returns and waiting changes are not executed profits.
Can I download the experiment's actual data?
Yes. JSON/CSV include the event ledger, conditions, outcomes, original incremental comparisons and strata, plus combination_checks summaries. contract.combination_extension describes the post-hoc selections. The page links a fixed-release JSON, hashed hourly inputs and a verification example for the original experiment. Record the release ID; the added combinations can be recounted from observation flags and published period boundaries.
Does this include funding, liquidations and open interest?
No. Those dimensions do not have the same aligned five-year coverage in this study. They are not synthetically filled or used to support these percentages. Separate market-context research has its own sources and limits.
Will CoinNudge send these bottom confirmations to Telegram?
This article is an experiment, not a promise of a new alert category. See the current Telegram alert list for delivered services. Public research, personal alerts and paid data subscriptions have distinct scopes.
Keep the evidence—not a promise of the bottom
Use supported Telegram alerts to reduce chart watching, or inspect the data catalog for reproducible research. Check actual delivery scope before subscribing.
Explore supported alerts