Three backends left, as the operator specified: OpenCL, the CPU DLL,
and pure MQL5. CDirectMLMy was a two-tier wrapper (GPU via
WarriorDML.dll, CPU via WarriorCPU.dll) whose name only ever named the
tier being removed here; the CPU DLL tier - the one actually used on
the training machine (no OpenCL, no DirectML) - is untouched.
AI/NeuronDirectML.mqh -> AI/ComputeDll.mqh: dropped the DML_* #import
block and COMPUTE_TIER_GPU (checked first that nothing persists the
enum value and only one external site reads .Tier() - safe), collapsed
every tier==CPU?CPU_x():DML_x() ternary to a straight CPU_x() call.
Renamed CDirectMLMy->CComputeDll, InitDirectML()->InitComputeDll(),
member directml/DirectML->computeDll/ComputeDll across every AI/ file
that touched a neuron/net backend plus Topology.mqh/OnlineLearning.mqh.
NetBuild.mqh's InitComputeDll also lost the dead D3D12 error-code
switch and the now-impossible GPU-tier log branch.
Verified via per-file brace-balance diff against HEAD and a whole-repo
grep for every removed symbol (CDirectMLMy/InitDirectML/
COMPUTE_TIER_GPU/DML_*) - the only surviving hit is an intentional
historical-note comment in the new file's header.
DirectML\WarriorDML.cpp/.h and its build scripts are now orphaned C++
source, left in place pending an operator decision. Architecture docs
(AI_NETWORK.md, Warrior_EA_System_Overview.md, etc.) still describe the
4-backend/GPU-tier shape and are not updated in this pass.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Last surviving compile-time feature switch in the codebase - the same pattern
already killed for the MARKET build and DirectML tier (02766b5): one build,
configured at runtime like every other module (inputs + getters/setters, set
in ConfigureAISignal during OnInit), not a second code path that only existed
if someone remembered to define a macro before compiling.
Replaced with `input bool ExportFeaturesOnly = false` (Variables/Inputs.mqh)
and a plain m_exportFeaturesOnly member + setter, matching AutoTuneIndicators'
exact shape. Four call sites converted from #ifdef to a runtime read of the
same variable:
- Warrior_EA.mq5 OnTick() - reads the input directly (this check has to
stand before any per-signal object exists)
- Topology.mqh's config-lock skip and ExportFeatureMatrix() call - read
m_exportFeaturesOnly, now set by ConfigureAISignal before InitIndicators()
runs (same init-order guarantee AutoTuneIndicators already relies on)
- ExportFeatureMatrix()/ExportRawRates() declarations - always compiled now,
called conditionally instead of not existing as symbols
No change to what the flag does when off (the state of every build that
exists today, since the macro was never defined anywhere in-repo) or when on;
only how it's set. Verified: WARRIOR_EXPORT_FEATURES fully gone from every
#ifdef/#endif in the tree; brace and ifdef/endif counts balance in every
touched file; ConfigureAISignal runs before StepInitIndicators in OnInit's
linear init chain, so the flag reaches InitNeuralNetwork() in time.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
The //| box blocks were excluded from 0b06f8e and 5efdb48 and were what
remained: 160 of them ran to 10+ lines, the longest to 88. Compressed to their
leading topic sentences - 5 lines for a function header, 8 for a file header -
keeping the box format and the standard MQL5 name/author lines verbatim.
Verified at the BYTE level this time, across every in-scope file: the list of
non-comment lines is byte-identical to HEAD and braces balance. The first check
compared a locale-decoded 'git show' against a UTF-8 read and flagged 25 files
that had not changed at all - every BOM and every non-ASCII line mismatched.
47,696 -> 40,665 lines in scope; comment share 38% -> 26%.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Same pass as 0b06f8e, applied file by file: comment runs of 4+ lines compressed
to their leading topic sentences, capped at 4 lines, whole sentences only.
Warning sentences (NEVER / MUST / trap / would-have) survive the budget.
Every file was checked the same way before committing: the list of non-comment
lines is byte-identical to HEAD, and braces balance. No code was touched.
Panel/, Enumerations/ and the already-terse System headers needed little or
nothing - PooledGate, TradeChecks, BinomialStats and Random came through with
no blocks over the threshold at all.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Reverts a863796 on the operator's call - "unnecessary complexity". It was
right about the mechanism and wrong about the priority: it re-cut the classes
for a case the measured verdict never reaches (SP500 H4 reads "both sides" at
the derived geometry), while the drift that IS happening affects every chart
and every era. Recoverable from a863796 if a one-sided book ever becomes real.
Two pieces of it survive, both independent of the exit idea:
The drift verdict keeps reading m_winLongCache/m_winShortCache rather than the
collapsed label pair. That line reports always-long vs always-short win rates,
which is what the win caches hold - each side scored on its own barriers,
published before the collapse. The label pair carries only the side touched
first, so it undercounted long wins by the both-won-goes-to-short share. There
are zero both-won bars at any geometry with target >= stop, so this changes no
number today; it changes the wrong number to the right one.
And the .cfg gains nothing and loses nothing: the two appended ints go away
again, and they were the last fields, so a .cfg written by yesterday's build
still reads correctly - the loader simply stops before them.
WHAT THE REVERT MAKES ROOM FOR. The operator's actual requirement is that the
model reproduce the label distribution the scan measured, and nothing in the
pipeline ties it to that. The loss trains on a rebalanced sample and the
abstain rate is owned by a margin threshold fitted on EDGE, so the call rate
and the label prior can drift arbitrarily far apart - and did, invisibly:
at era 1350 the models call Buy on 20-28% and Sell on 22-32% of bars against
a scan-measured 2.1% and 4.8%. Roughly a 10x over-call, and not one line in
the journal said so.
The era line now carries it:
CALIBRATION calls vs true rate Buy 28% vs 2% (14.0x) Sell 32% vs 5% (6.4x)
Neutral 40% vs 93% (0.4x)
Reported as a ratio because that is the readable number - 1.0x is calibrated.
This is deliberately a measurement and not yet a correction: matching the
label rate would put coverage near 7%, below the ensemble gate's own 12.4%
coverage floor, so calibration and the gate are in direct conflict and which
one yields is the operator's call, not mine.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
User request: "when an asymmetry is noticed in a market (like sp500 upward
drift) ... it does not need to predict shorts, but exit points. a sell signal
needs to be preceded by a buy so that it can say I predict we must close that
long."
Until now a LONG_ONLY verdict only BLOCKED short entries. The network went on
being trained to predict them - a third of its output capacity spent learning
an answer the direction policy guarantees it can never act on, while the
question the book actually faces (when to get out of the long) was never
asked. The two are not the same event: "a short pays" needs price to travel
the SHORT's target before the SHORT's stop, and at any geometry where reward
!= risk that is a different bar from "this long hits its stop first". The exit
is the second one.
So on a one-sided book TripleBarrierLabel re-cuts all three classes around the
only position the book can hold: Buy = it reaches its target, Sell = it
reaches its STOP first, Neutral = the horizon expired with it still open. Both
come off the allowed side's own barriers, which the walk already computed -
this reads longLost where it used to read shortWon, so it costs nothing.
Label lifespan and the timeout flag follow the allowed side too, so the
overlap correction is sized on the window this label actually spans.
DECIDED ONCE, AT ERA 0, AND PINNED. m_exitTargetSide goes in the .cfg beside
the derived geometry under the same doctrine and for the same reason: it
decides what Buy and Sell MEAN, and a target that moved mid-run would retrain
a fitted model against something it never saw. A .cfg from before this ends
early and reads 0/0 - "not decided, symmetric" - which is exactly what every
existing model was trained as, so nothing needs migrating. The weights
fingerprint keys on the INPUT only (explicit Long only / Short only); under
Intelligent the measured verdict must never reach a filename, or the model is
orphaned the moment more history downloads.
THE DRIFT VERDICT HAD TO MOVE OFF THE LABELS FIRST, and it turns out it was
measuring the wrong thing anyway. It counted m_labelCacheBuy/Sell and called
them "always-long vs always-short win rate", but the label pair is the
COLLAPSED first-touch verdict: a bar where both sides reached their target
carries only the side touched first, so long wins were undercounted by the
both-won-goes-to-short share. m_winLongCache/m_winShortCache are the actual
per-side win rates, published before the collapse, and that is what it reads
now. Necessary as well as more correct - deriving the verdict from labels the
verdict shapes is a feedback loop, since Sell-as-exit is near complementary
to Buy and would close the very gap that produced it. The gap's SE now leans
conservative rather than anti-conservative for the same reason.
LIVE. The retargeted class is wired to close the position, or training it
would be pointless: CheckClosePosition's "never vote-exit a certified
position" rule keeps governing symmetric books and gains a one-sided
exception, and the replay reads the identical rule through one
LiveVoteExitThreshold() so certified and traded cannot describe different
policies. Armed only when the operator picks a close threshold
(Signal_ThresholdClose ships Disabled) AND the model's own pin says its
blocked-side class means "close" - a model trained symmetric never fires it,
whatever the verdict has since become. This does trade a different game from
the one the win-rate certificate grades; the era's EXIT-POLICY REPLAY line
already reports expectancy in R for exactly this case and says so in words.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Every AI signal repeated the same five-line InitIndicators override that
did nothing but call InitNeuralNetwork. The cause was an access mismatch,
not a design: CExpertSignalCustom declares InitIndicators public, the AI
base redeclared it PROTECTED, and each subclass had to redeclare it
public to be reachable by CExpert. Worse, the base's own override does a
different job entirely - it creates the OHLC/ZigZag feature indicators -
and InitNeuralNetwork called it back scope-qualified to stop the virtual
dispatch landing in the subclass. Two jobs, one virtual name, and a
recursion trap held off by a scope qualifier.
The feature-indicator step is now InitFeatureIndicators() (protected,
non-virtual, named for what it does) and the AI base carries the single
public InitIndicators override. CONV/HYBRID/LSTM/PAI/META drop their
copies and are now purely identity plus topology, which is the classic
signal file's shape.
Comment pass on ExpertSignalAIBase.mqh, -100 lines with every constant
and every measured number kept. Three claims in the tier block were
stale and inverted - it named CalibratedConfidenceMagnitude() as the
tiering input where the code deliberately uses the RAW magnitude, and it
described the signal DB as re-ranking each tier when ApplyPatternWeight
declines the DB from the end of era 1. Also dropped a paragraph whose
subject was a previous version of the comment, and moved two notes down
onto the constants they document (CONV_COMPRESSION_DIVISOR was 16 lines
and three unrelated defines away from its own text).
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
The MECHANISM was already stdlib and is untouched: ThresholdOpen() ->
m_threshold_open, tested as `m_direction >= m_threshold_open` exactly as
CExpertSignal does it. What was wrong was the presentation. Both inputs
were preset ENUMS labelled "Min confidence to open/close (%)", which
names the wrong quantity - m_direction is a WEIGHTED MEAN OF PATTERN
WEIGHTS, not a probability, and nothing in this path is a confidence.
They are now plain ints named the way the MQL5 wizard names them:
input int Signal_ThresholdOpen = 25; // [0...100]
input int Signal_ThresholdClose = 101; // [0...100, 101 = never]
Values are exactly what shipped, so behaviour is unchanged. 101 rather
than the library's default of 100 for close: a weighted mean of pattern
weights cannot REACH 101, which is how the shipped config disables the
vote exit, and quietly lowering it to 100 would re-arm a live exit route
as a side effect of a naming change.
VOTE_CLOSE_PRESETS is deleted (its only user is gone). PERCENTAGE_PRESETS
stays - MinRecall genuinely is a percentage.
** ACTION NEEDED ON DEPLOYED CHARTS: the inputs are RENAMED, so saved
.set files no longer match and charts fall back to the defaults above.
Those defaults are the current shipped values, so a chart on 25/Disabled
needs nothing; a tuned one does.
Comment cleanup in the same pass, and this part was not cosmetic - three
blocks documented mechanisms that no longer exist:
- the AI early-exit route (deleted in 38a12a2) described as live and
still firing every bar;
- the m_lastNonNeutralSignal alternation gate (removed 2026-08-01)
described as consuming the AI's vote;
- 16 lines of VOTE_CLOSE_PRESETS documentation orphaned by that enum's
deletion, ending with "see that enum's note directly above" pointing
at nothing.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
MQL5's MathRand() is the 15-bit MSVC LCG - 32768 distinct values and
the lattice structure that shape of generator has. Two places here
actually lean on randomness and both were hurt by it:
WEIGHT INIT. Six He/LeCun-uniform sites drew
((MathRand()+1)/32768.0 - 0.5) * 2 * scale, so a first dense layer of
~250k weights had only 32768 possible values and thousands of
connections started byte-identical. Breaking that symmetry is the whole
job of random init.
SHUFFLING. ShuffleRandomIndex() already had to splice TWO MathRand()
draws to reach 30 bits, and its own comment documented the residual
modulo bias it still carried. HQRndUniformI() is rejection-sampled and
exactly uniform, so the splice and the bias note both go.
CHighQualityRand is L'Ecuyer's combined multiplicative congruential
generator - two differenced streams, 31-bit output, period ~2.3e18 -
and it ships with the terminal.
AND A BUG THE MIGRATION EXPOSED. The three MathSrand(GetTickCount())
calls sit immediately before "build a fresh topology", once per model.
GetTickCount() steps in ~15.6 ms on Windows and an ensemble builds every
member inside one OnInit, so members could be handed the SAME seed and
draw the SAME weights wherever their shapes coincide - and members that
start identical are not an ensemble. WarriorRandSeed() takes a salt (the
model id) plus a never-reset call counter, so a collision is impossible
rather than merely unlikely, while the tick keeps the run itself
genuinely unrepeatable the way those call sites asked for.
Seeds are masked positive rather than trusted: HQRndSeed computes
s % (M-1) + 1 and MQL5's % keeps the sign, so a negative seed leaves the
generator in a state its own assertions reject. GetTickCount() is a uint
and goes negative as an int after ~24 days of uptime - a fault that
would surface as "training is broken" on a long-running terminal and
nowhere else.
The indicator tuner's 52 draws move across too: its random search is
where sample quality earns its keep.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
InitNeuralNetwork() was 673 lines and the fingerprint assembly - the
single most audited block in the file, since its hash decides when a
trained model may be resumed and when it starts again from era 0 - sat
in the middle of it with no name of its own.
BuildModelFingerprint() is now that block, moved line for line. Its
header states the two rules the per-field notes have been repeating one
at a time for months: measured quantities never enter (the .cfg carries
those, adopt-don't-compare), and new fields append conditionally so
shipping one does not re-key models that never use the feature.
The assembly is byte-identical - verified by diffing every `fp =`/`fp +=`
line against HEAD, which differ only by the new call site. So no
existing .nnw/.cfg re-keys and nothing retrains.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
The codebase had THREE conventions for the same statistic. AltData took a
true median; the barrier horizon and the derived input window took the
upper of the two middle values; the MI terciles and the barrier stop
ladder used nearest-rank indexing. All four now go through MathMedian /
MathQuantile, which is R's type 7 and the library's one answer.
System\AltData.mqh column median -> MathMedian (exact, no change)
AIBase\Labels.mqh swing median -> MathMedian
leg-range med -> MathMedian
stop ladder -> MathQuantile, read in one call
AIBase\Topology.mqh window median -> MathMedian
AIBase\AutoTune.mqh MI terciles -> MathQuantile + MathMin/MathMax
Signals\SignalSessionFilter DST last Sunday-> CDateTime::DaysInMonth()
gaps[]/legs[] change from int to double so MathMedian can read them; the
values are bar counts either way.
VALUES MOVE. Even-sample medians shift by half a bin and the quantile
reads interpolate, so the barrier geometry and the derived input window
can land on different rungs - re-keying fingerprints and forcing a
retrain. Accepted deliberately: stdlib consistency was the ask, and three
private conventions for one statistic is what it buys out.
Two YAGNI finds fell out of the ladder rewrite. MathQuantile sorts its own
copy, so DeriveBarrierGeometry no longer sorts up[]/dn[] in place - which
means upUnsorted[], a full array copy kept only to undo that sort, is
gone. ArraySort(up) had no consumer needing order at all; it was pure
work. The library call also gets a failure guard the hand-rolled indexing
never needed but the ladder read does.
Verified while here: Math\Stat\Math.mqh's MathAbs/MathMax/MathSqrt/MathPow
and friends are ARRAY overloads, not scalar redefinitions, so pulling it
into the translation unit shadows no builtin.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
EstimatedInSampleBars() returned raw bars (11372 on SP500 H4) and every derived
capacity decision spent that: first-layer width, conv filters, LSTM hidden size.
But triple-barrier labels overlap - mean lifespan 9.4 bars - so the label cache line
on the same run already reports those bars are worth ~1210 independent observations.
Sizing a network against RAW bars while grading it against EFFECTIVE ones is two
subsystems disagreeing about one sample, and it disagreed in the dangerous direction
because the capacity side was the optimistic one: the warning's "roughly 1.1 weights
per training bar" is nearer 11 per independent observation.
EffectiveSampleSize() has existed since 2026-08-17 and is applied at eight sites, all
of them statistics. This adds the ninth, in the one place that decides how many
parameters get fitted. Applied inside EstimatedInSampleBars() rather than at the call
sites, because that function exists precisely so the three stages spend one budget.
SELF-ENABLING AND THEREFORE INERT WHERE IT MATTERS MOST, which is why this is two
changes and not one. MeanLabelLifespan() is 1.0 until a label cache has measured
something, so on a model's first build - before any label exists - the deflation is
correctly the identity: an unmeasured overlap must not invent a shrink. A fresh
attach constructs a fresh object, so its counters are zero too; only a mid-session
weights reset carries real evidence into a rebuild. That is deliberately safe (no
attach can now re-derive a narrower topology and discard trained weights) but it
would have left the first build - the case you most want the truth for - quoting the
flattering figure. So ReportDetectability now restates capacity against the effective
sample at the first moment L is real, for the topology already pinned. It re-sizes
nothing; it reports what was bought. Placed ABOVE that function's break-even guard on
purpose - a degenerate geometry is exactly when you want to know the net is
over-parameterised, and "it only fires for sane configs" is how the 2026-08-18
IS-error stop managed never to fire at all.
The warning also names its basis now (independent observations and L, or an explicit
"overlap NOT YET MEASURED, this is an UPPER BOUND"), so a flattering number can never
again read as a measured one.
Also factors FirstLayerFanIn() out of ComputeFirstLayerWidth so the capacity REPORT
charges for exactly what the capacity DECISION charged for - same reason
RequiredHorizonBars was factored out after the 2026-08-17 divergence - and makes
MeanLabelLifespan()/EffectiveSampleSize() const so the const budget path can call them.
Verified: no recursion (EstimatedInSampleBars -> EffectiveSampleSize ->
EstimatedInSampleBarsRaw, which computes from Bars() alone); both new StringFormat
sites hand-counted (basis 3/3 and 1/1, CAPACITY 10 specifiers / 10 arguments).
NOT COMPILED - user compiles in MetaEditor.
The schedule became part of the label's meaning (3e467f9): the same
chart trains a different target under Friday-23:45 than under
everyday-22:00. Without this token a schedule change silently resumed
weights fitted to the other target - the stale-enum-wrong-target family.
Active schedule -> "|CUT:day@hour:minute" in the fingerprint; disabled
schedule appends nothing (pre-change no-schedule models stay
byte-identical). Default-Friday charts re-key exactly once, at this
change - deliberate: their weights were trained on weekend-blind labels
and are confounded anyway.
NOT COMPILED - user compiles in MetaEditor.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
NOT COMPILED - user compiles.
Root cause of the Neutral collapse. Logit adjustment (Menon et al. 2020) makes a
classifier Bayes-optimal for BALANCED error by subsidising rare classes. It was
wired here when Neutral was the DOMINANT class - the "big move up / big move down
/ nothing much" era, where the correction pulled the model off the majority.
The triple-barrier relabel (b4a704d) inverted the distribution. The barriers are
now the EA's own SL/TP, so ~89% of bars RESOLVE and only timeouts are Neutral.
Measured on SP500 H4, from the EA's own log:
measured priors Buy 48.26% Sell 41.13% Neutral 10.61%
log-prior spread 1.52 | tau 1.00 CAPPED to 0.79
Neutral became the RAREST class, so the correction started subsidising it - by
tau*(log pB - log pN) = 1.20 logits. With no directional edge to overcome that
(direction is closed at best-of-999, p=1.0000), the model took the free lunch:
OOS recall Buy:1% Sell:0% Neutral:100%
OOS raw out spread avg 0.9993 (softmax saturated, near one-hot)
dW/W bn1 0.000% bn3 2.0% bn5 6.5% (input weight block frozen; head twitching)
The anti-collapse mechanism was the collapse. The recall gate needs >=40% on all
three classes, so nothing could ever deploy and the plateau ladder burned eras.
Present in both runs today (b6b5 froze bn1 by era ~719, 17ae by ~169), so it
predates this week's work.
FIX: the correction now spans the DECIDABLE classes only, Buy against Sell,
centred on their midpoint, with Neutral pinned at offset 0. Neutral is the
ABSTAIN outcome and abstention already has a better owner - m_dirConfThreshold,
refitted every era on the held-out calibration band against a coverage floor and
the measured break-even. Subsidising the abstain class does that job twice and
spends the whole correction suppressing the only decisions that can pay.
What still gets corrected is real: a trending symbol resolves more long barriers
than short, and uncorrected the model inherits that as a standing directional
bias. Here it is log(0.4826)-log(0.4113) = 0.16, so the offsets are tiny - the
correct answer, not a broken one. The two traded classes were already balanced;
the old spread of 1.52 only ever described how rare a timeout is.
Everything is derived from the measured distribution, as requested - offsets from
the priors, cap from the resulting spread. tau itself is deliberately NOT fitted:
tuning it against the same data that selects the checkpoint would add another
search dimension to a project that has been burned by exactly that. tau=1 is the
theory value and the cap (now ~9.5x looser at spread 0.16) will rarely bind.
Log line now reports both spreads and, when the abstain class is the rarest, says
how much the old form would have boosted it. Fingerprint |LA:<tau> -> |LA:<tau>:BS
so models trained under the all-three form re-key instead of resuming.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
User's call before deploy: "I would rather avoid lagging so the NN finds
accurate patterns." Correct instinct, and it picks the conservative variant.
110b384 deduplicated the WHOLE window, so every distinct reading survived at one
slot. The flaw is which slot: it depends on where the calendar-day boundary falls
inside that particular window, and on H4 that boundary cycles through ~6 phases.
A dense layer holds a separate weight per (slot, feature), so a given lag would
have landed on a different coordinate from one window to the next - turning a
stable lagged input into a moving one.
Now it blanks only bars carrying a BYTE-IDENTICAL copy of the anchor's reading
and stops at the first bar that differs. An as-of lookup into a daily file is a
step function in time, so those copies are exactly the contiguous run of bars
sharing the anchor's calendar day. Everything older keeps its natural replicated
run, in the same slots it always occupied - whatever the net learned to read
there, it still reads there.
Why the anchor's reading is the right one to isolate: the window's newest slot IS
the bar being predicted (BuildFeatureWindow's final iteration lands on r, and
pass 3 grades that same index), so it is the reading contemporaneous with the
decision - and the only one the alt screens ever validated. They measured the
CURRENT reading's MI against forward range and never tested lags, so the lagged
content is unproven, which is a reason to leave it undisturbed rather than a
licence to rearrange it.
What is still fixed: the anchor's reading reaches the first layer on one
coordinate instead of once per bar of its day, removing the ~16x gradient
upweight for the validated signal. And this is IDENTICAL to full dedup exactly
where replication was worst - on M15/H1 the whole window sits inside one calendar
day, so the anchor's run is the whole window - and a no-op on D1, where the bar
before the anchor is already a different day and the loop breaks immediately.
The two differ only on middle timeframes, and there this is the safe side.
Fingerprint |ALTW:1 -> |ALTW:2 so nothing trained under the hour-old full-dedup
semantics can silently resume under these.
Compile-verified: 0 errors, 0 warnings.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Measured on the live SP500 D1 export (6073 rows, 13 features, 5888 simulated
16-bar windows):
distinct values per feature per window : 1.7 - 2.7 of 16 slots
variance in the first 13 PCs : 96.5 - 97.0%
components for 95% / 99% : 12 / 17-19
effective rank (entropy) : ~11.5
208 inputs carrying about 12 dimensions. Only 6 of the 13 features move daily
(VIX complex, USD, the rates trio); 5 are weekly (COT, EIA, output gap) and 2
monthly (CPI, unemployment). The lookup is as-of by bar open time into a DAILY
file, so bars sharing a calendar day are byte-identical by construction.
The cost is NOT overfitting capacity - collinear copies span ~12 directions,
not 208, so an earlier claim that this wasted 26% of the model overstated it.
It is GRADIENT WEIGHTING. Batch norm standardizes each of the 208 coordinates
independently; that rescales the copies without decorrelating them, so one
factor arrives on 16 unit-variance coordinates, each weight takes a full-size
step, and the factor's aggregate coefficient moves ~16x faster than a per-bar
price feature's. The network was biased toward the external block by a factor
of the window length - and pointing the wrong way, since these features cleared
only a marginal incremental screen while price is the base signal.
Zeroed at WINDOW ASSEMBLY, not in BufferTempData: that output is cached PER BAR
and a bar sits at slot 15 of one window and slot 0 of the next, so a
slot-dependent value there would poison the cache or force a recompute per slot.
The cache keeps true values; only this window's copies are cleared. Width
contract untouched - same count, same positions - so conv/LSTM/HYBRID keep their
bar-major rectangle unchanged and the block arrives at the newest bar, which for
the LSTM is the final timestep. Zero-variance coordinates are safe through batch
norm (divisor is MathMax(MathSqrt(var + BN_EPSILON), BN_MIN_STD)).
Fingerprint gains |ALTW:1 when alt data is on. Same width and same .cfg, so
nothing else would have caught a model trained under the replicated layout
resuming under this one. Conditional append per the existing rule: configs
without alt data keep their fingerprints and their trained models.
NOT the concat branch. CNet is a strictly linear stack (CLayerDescription has no
input-source field; NetBuild wires i to i+1 and stores layer L's weights on
L-1), so a real two-tower model needs a new multi-input layer type across
WarriorCPU, WarriorDML and the OpenCL kernels plus an .nnw format change - the
highest-risk change in this repo, in the code that produced the transposed dense
gradient, the Adam second-moment bug and the reversed LSTM window. This captures
the part of that idea the measurement actually supports, at no engine risk.
Compile-verified: 0 errors, 0 warnings.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
TOPOLOGY - reverts the two constants and drops CausalHiddenLayerFloor.
The MQL5 article's 30%-per-layer cut and floor of 20 are load-bearing on ITS
first-layer width of 1000 (1000->300->90->27 needs a floor to stop). This
codebase MEASURES that width, and on the live SP500 H4 config it is 16 units -
already floored, with the budget printing "11360 estimated in-sample bars
cannot support a 800-wide input ... roughly 1.1 weights per training bar -
expect overfitting". At 16 units a floor of 20 makes lastHidden >=
m_initialNeuronsCount, so ComputeHiddenLayerCount returns on its first branch
and the width taper - the only part derived from this symbol's data - became
dead code on all four ensemble members, with depth (2 -> 4) set entirely by
counting feature domains. ComputeLayerWidths had already rejected this exact
pair of constants in its own comment.
The causal floor's premise does not hold either: layers are not inference
steps. The "1 layer linear / 2 nonlinear / 3 multi-connected" result is
Lippmann 1987 and is about hard-threshold units; with sigmoid/ReLU, Cybenko
1989 and Hornik 1991 give universal approximation from a single hidden layer.
Depth buys parameter efficiency for compositional functions, not reasoning
hops. ForceHiddenLayers remains for measuring depth directly.
RANKING SLICE - the backfill no longer reads the window it is judged on.
The deployed checkpoint is CHOSEN as the best-scoring era on the OOS window,
so win rates measured back over it are selection-inflated, and the backfill
was writing exactly those into the table filter weights rank on: the
selection set consumed twice, beside a deploy gate that applies a Sidak
correction for that effect. The newest RANK_SLICE_PCT_OF_OOS (20%) of the OOS
window, plus a label-horizon purge, is now reserved and graded by nothing -
not pass 3, not checkpoint selection, not the gate. The backfill reads only
that. The gate keeps ~80% of its measurement (power goes as the square root,
so ~10% of a sigma), and the slice is the newest data, which is the regime
about to be traded. RankSliceBars returns 0 when no honest slice fits and the
backfill then REFUSES and says so, rather than falling back to the scoring
window and looking like a success.
SHRINKAGE - per-tier win rates are shrunk toward the filter's own pooled rate
by MIN_TRADES_FOR_WIN_RATE pseudo-trades before becoming weights. The raw
ratio at the minimum sample count carries a ~15pp standard error, so a tier
that went 8-2 was handed weight 80 and outranked a tier measured over
hundreds of calls at 55 - the ranking was being driven by which small tier got
lucky. Opt-in per call site (priorWeight 0 keeps the raw behaviour).
Compile-verified: 0 errors, 0 warnings.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Four defects in 64c5dd5/1a05e63, found by review + a baseline compile.
Goals 1-8 of that session are unchanged; this makes 6 and 8 actually reachable.
1. HEAD DID NOT COMPILE - 6 errors. CControlPanel::Minimize/Maximize were
declared `virtual bool ... override`, but CAppDialog declares both as
`virtual void` (Controls\Dialog.mqh). errors 265 + 404 on each, plus 151
on `bool ok = CAppDialog::Minimize()`. Return type is void now; there was
never a success flag to forward. Verified: 0 errors, 0 warnings.
2. THE BACKFILL COULD NEVER ADVANCE, and neither could the OOS continual
simulation (that one has been dead since it was written). Both are armed
at the instant convergence is declared, and both advance only from inside
Train(), one chunk per call. But ScheduleTrainingIfNeeded's only per-tick
ArmStudyEvent site sits in the `else` of a branch taken whenever
m_trainingComplete is set and m_trainRunActive is clear - which is exactly
the state FinalizeTrainRun() leaves behind one line before they are armed.
Train() was never called again, so the walks sat at their start index
forever: no "simulation complete" line, and not one row written to the DB
this feature exists to fill. Only a manual Resume/Retrain unstuck them.
Both flags now keep the model schedulable.
3. IN AI_HYBRID - the mode this ships in - the backfill was never even armed.
Ensemble members deploy at Train() ENTRY and return immediately (so no era
is wasted), which skips the era-end block the backfill was started from.
All four members were a no-op for a second, independent reason. Armed on
the ensemble deploy path too, from m_resumeBars/m_resumeOosCutoff.
4. RE-RUNS DUPLICATED ROWS. RegisterSignal inserts unconditionally - no key,
no duplicate check - and m_dbBackfillDone is in-memory, so every later
attach that retrained to convergence wrote a second full set of rows for
the same bars. The ranking would count one bar once per model that ever
deployed, weighting superseded opinions as heavily as the live one. A
.dbfill marker stamps the deployed era; written only on completion (an
interrupted walk redoes itself rather than ranking a partial window) and
deleted with the other sidecars on reset-weights.
Also: WarmBlocking's timeout was silent, which restored the exact silent
pin failure it was added to prevent - it now says so in the journal, and
returns true for "no reference pairs to wait for" so the warning stays rare
enough to be read.
Not addressed, needs a decision: the backfill scores the OOS window with the
checkpoint that was SELECTED as best on that same window, then writes those
win rates into the table filter weights rank on - the selection set consumed
twice, undiscounted, while the deploy gate right next to it applies a
family-wise correction for exactly that effect. The rows are also simulated
triple-barrier outcomes at today's spread sharing a table with realised
fills. The completion log line now states both plainly.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Four user-reported/requested items, one root cause chain:
1) DEAD CONTROL PANEL in AI_HYBRID mode. All members posted custom event
id 1 and handled id 1001, and CExpertCustom broadcasts every chart
event to every filter - so each posted event ran a train chunk in ALL
N members (N*N chunks per round) and the chart thread never idled
long enough to deliver clicks/drags. profiling.csv: 99.45% of time in
OnChartEventHandler. Fix: per-instance study-event ids
(STUDY_EVENT_ID_BASE + construction order, offset above the Controls
library's ON_* codes - id 1 was also ON_DBL_CLICK, so panel
double-clicks fired training chunks). ArmStudyEvent() is the single
post site; lost-event watchdog replaces the accidental
sibling-clears-my-flag rescue.
2) WARM-UP DUPLICATION. The auto-tune sweep is deterministic over
identical features/labels, and it ends in the full MI diagnostic
suite, which the MI-share gate never intercepted on the sweep path -
four members ran four identical ~36s sweep+report blocks. First
member publishes outcome (g_ensembleChartTuneDone/Installed/Settings);
the rest apply it and skip both.
3) DEINIT STRANDED PANEL+ARROWS (user repro 18:52). Root cause from the
log: the 4,500ms budget runs from MetaTrader's stop REQUEST - a heavy
autosave in flight ate it, OnDeinit got ~430ms and died in the first
member's arrow persist ("Abnormal termination" 432ms in). Fix: early
visible-UI sweep (native prefix deletes for status/panel/dialog)
right after ClearStatusLabel, and a fast path for still-training
models - their arrows are re-rendered every era, so they get one bulk
purge instead of scan+atomic-write in the death window.
4) ENSEMBLE FEATURES (user requests): era BARRIER - members advance era
by era together; a member ahead of the slowest still-training member
declines Train() calls and its chunk budget is donated
(TRAIN_TIME_BUDGET_MS = 120/activeTrainers, UI headroom constant).
COMBINED-VOTE OOS SCORE - each member's pass-3 scan contributes its
adjusted per-bar decision (0.0 on abstain) to a shared row buffer;
the last member to finish the era scores the averaged vote vs the
mirrored Min_Vote_Open against the same target-before-stop outcomes
members grade themselves on, publishing an "Ensemble vote" line on
the aggregated panel. Member headlines now carry their lifetime win
rate with break-even.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
- PeriodMA/MA_Type/PeriodRSI: input -> const seeds (closing the set: every
indicator parameter is now tuner-owned)
- Variables\TunedPeriods.mqh: chart-level tuned-period state. A gated
install writes TunedPeriods_{SYM}_{TF}.cfg; next attach reads it BEFORE
the DB fingerprint and classic-signal config, so classic votes, DB key,
and tuner seeds always describe the same indicators regardless of
classic/AI/hybrid use. Restart-grained adoption by design (no mid-run
handle churn); new periods re-key the signal DB (semantics rule).
- EnableAltData input in AI Input Features (consumption gate only;
collection keeps running); |ALT DB-fingerprint token; opt-out on an
alt-trained model correctly starts fresh via the width compare.
- Defaults: all four classic votes OFF (AI-first; WARRIOR_MARKET_BUILD
branches collapsed with the marketplace pivot), order-flow/Wyckoff NN
features OFF (alt data is the default information diet; toggles stay).
Compiles 0 errors / 0 warnings.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
The 18 inputs added 2026-08-08 (when the tuner defaulted off and the
values needed an operator path) become compile-time aliases of their own
defaults - same names, zero consumer churn, byte-identical values. The
tuner is now the only path by which these values move: it defaults ON
(the 08-08 off-flip was measured against the direction target's flat
landscape; the objective is now RANGE, which has signal), searches from
the seeds under the Sidak family-wise gate, and persists winners in the
.nnw beside the weights. ADP fingerprint token retired (deviation now
impossible by construction; tuned values were never its job).
Menu shrinks 102 -> 84 inputs. Compiles 0 errors / 0 warnings.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
The user is right that no special combination logic is needed: the AI
signals are ordinary voting filters, and the aggregate already has
union semantics - abstaining filters do not dilute the average, so an
ensemble chart trades whenever ANY deployed member clears the vote
threshold and disagreeing members net out. What the ensemble preset
actually adds:
- AI_CHOICE value 4 renamed AI_CONVLSTM (the name says the front-end);
enum VALUES stable, CSignalHYBRID class and State\HYBRID\ folder kept,
so saved configs and trained models keep their identity.
- New AI_HYBRID = 6: enables PAI+CONV+LSTM+CONVLSTM together on one
chart - replaces four separate charts of the same symbol. Each member
trains and self-gates independently; only certified members ever vote.
- |ENS1 fingerprint token on every member, so an ensemble member's
weight files can never collide with a solo model of identical
settings on another chart of the same symbol (the duplicate-chart
guard would otherwise correctly fight over one .nnw).
- Private default AIType = AI_HYBRID: one D1 drop now yields every
topology's gate verdict for that symbol.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
User direction (2026-08-15): back to predicting swing turns, D1 charts,
fractals over ZigZag pivots (their call - balances classes, matches the
reference library target, and a 5-bar fractal confirms 2 bars after its
extreme so labels resolve nearly to the present with no repaint embargo).
- TRAINING_TARGET enum + TrainingTarget input: TARGET_BARRIER (Market
default - existing models keep their meaning and fingerprints) or
TARGET_FRACTAL (private default).
- FractalDirectionLabel (Labels.mqh): per-bar 3-class label = direction
from the bar close to the next confirmed strict 5-bar fractal extreme,
costs charged in the same bid-series convention as the barrier label,
Neutral when the move cannot clear max(2 spreads, 0.10 ATR) or on an
outside bar (both-extreme bars are unorderable within OHLC).
- The barrier walk still runs in full: measured SL/TP geometry, the
expectancy scan, excursion caches and the era gate all keep scoring
what a trade at the EA's own stop/target actually collected - only the
TRAINING label changes. NOT the pre-b4a704d "is this bar the pivot"
form; that target's 31:1 imbalance stays retired.
- Fingerprint token |TGT:FRA1 so switching targets trains a separate
model; AI_META unaffected (guarded setter).
- Private defaults: AIType back to AI_HYBRID (direction topology needed)
+ TrainingTarget=TARGET_FRACTAL = drop-on-D1-chart workflow.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
A resumed META model hot-looped pass 1 (0->100% scan oscillation, silent for
3 minutes until the stall reporter fired) because EVERY window failed at the
first AD/Wyckoff feature: the init-time param adoption called
ReInitADIndicators unconditionally, destroying five freshly-calculating
indicator instances to recreate them with BYTE-IDENTICAL params (verified by
parsing the .nnw header - the MI tuner had kept the configured settings), at
process start, on a box with 1 GB free of 31. The replacements sat cold for
6+ minutes while full-history resweeps starved the indicator threads harder.
- AdoptIndicatorParams: installs a loaded param set into the tuner and
rebuilds handles ONLY when the set actually differs from what the live
indicators run. Both call sites (resume init + panel reload) use it.
- Resumed models get the same 3 warm-up passes as fresh ones. The skip was
the shared root cause of the cold-ATR (ba13eef), cold-AD (2026-08-11) and
this incident - custom indicators recompute from scratch every process
start regardless of what the .nnw proves.
- Cold-sweep backoff: a pass-1 sweep in which every window failed on a
TRANSIENT cause arms a 5s era-start pause instead of an immediate
full-history resweep, so the retry loop stops consuming the CPU/memory the
warming indicators need. The stall reporter names the backoff branch.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
The NN now has a target that is not per-bar direction (closed, best-of-999
p=1.0000): P(win | this journaled candidate, at the EA's own SL/TP, net of
cost). One net for all 52 pattern-sides, AIType=AI_META.
- NetForward.mqh: the host-side softmax+CE gradient generalized total==3 ->
2||3 on both backprop paths; a 2-class softmax IS a logistic head, and no
compute backend changes.
- SignalMETA.mqh (new): corpus loaded read-only from the LARGEST signal DB on
disk (decoupled from the config fingerprint that burned four S1 runs); the
GMT->server offset is measured PER ROW against entryPrice vs bar open
(DST-immune, histogram logged); a window-span regime filter drops the
pre-2017 daily-backfill rows; 31-feature setup descriptor appended at the
input (26 one-hot + side + tanh netVote + SL/TP ATR + spread/ATR).
- Training.mqh: candidate-queued pass 1, binary-target pass 2, per-candidate
calibration (2.5) and OOS (3) walks. Counter mapping win->Buy / loss->Sell
lets checkpoint selection, the edge floor, the plateau ladder and the
family-wise deploy gate run UNCHANGED: precision reads as win rate among
traded candidates, chance as the base win rate, recalls as sensitivity/
specificity. Era-end META line: coverage x (p - break-even) vs the null.
- Labels are the side-conditional triple-barrier win caches - never the DB's
stop-and-reverse outcome. Logit adjustment deliberately skipped (~40% base
rate). Live inference + online learning guarded off until S3.
- Fingerprint: conditional |TGT:META1; State\META\ folder + 2-output filename
slot keep meta models fully separate from direction models.
Compiles clean (0 errors, 0 warnings). S2 run = attach a chart with
AIType=AI_META; S3 wires the votes via the per-side hooks.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
The reference-pair set was re-discovered from Market Watch on every
build, so adding or removing a terminal symbol silently changed what a
trained model's six cross-asset features meant - the last open
train/serve parity gap from the 2026-08-11 audit. The set a model's
FIRST successful build actually used is now stamped into its .cfg
(append-and-length-guard, adopt-don't-compare - the derived-barrier
pattern) and every later build constructs the panel from exactly that
list; a pinned pair that is temporarily unavailable is skipped, never
substituted.
Also warms SymbolSelect/SeriesInfo for every reference symbol at
InitNeuralNetwork, so the terminal's ~minute of async cross-symbol
download starts at init instead of when the first Build() trips over
an unselected symbol - the source of the startup 'only 0 usable
reference pairs' console failures.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
On a CFD whose base and quote currency match (SP500 -> USD/USD) the FX
encoding degenerated: base and quote strength were the SAME series twice
and the divergence feature collapsed to the symbol's own 20-bar return.
Index mode re-encodes the six slots: denomination-currency strength
(fast/slow), a risk-proxy currency's strength (JPY by fixed preference
order - deterministic across rebuilds), and divergence as own move minus
what the denomination alone implies. FX-pair symbols are untouched.
Fingerprint gains :IDX2 for base==quote symbols only, so index models
trained under the degenerate encoding re-key while FX models keep their
filenames. FORCES RETRAIN on index/CFD charts.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Completes the 2026-08-09 training audit. FORCES A RETRAIN of every
Wyckoff-enabled config (N1 re-keys the fingerprint), and BOTH DLLs must be
redeployed alongside the .ex5 - they carry new exports.
F4 - mini-batch accumulation, TRAIN_BATCH_SIZE=32. Training was pure online
SGD (one weight update per bar), which is the mechanical source of the
era-to-era whipsaw every downstream guard was built to cope with. The O(n^2)
outer product is native - AccumulateWeightGrad / AccumulateWeightGradConv /
AccumulateBufferInto in Network.cl, WarriorCPU and WarriorDML - while the
optimizer step is host-side MQL5 shared by all tiers (ApplyAccumToBlock), so
there is one Adam/SGD implementation instead of four that can drift.
- the LSTM needs no outer-product kernel (WeightsGradient already holds the
sample's full dW) but could NOT simply be left un-zeroed between samples:
CPU_LSTMSeqBackward/DML_LSTMSeqBackward memset it on entry. Hence a
separate accumulator plus an elementwise add.
- batch-norm gamma/beta accumulate in host arrays, not new BatchOptions
slots - BN_OPT_STRIDE is baked into every persisted .nnw.
- scoped to pass 2; online learning keeps immediate updates. Every save /
checkpoint / scoring boundary flushes, scaling by the real sample count.
- degrades to per-sample updates (one log line) on a tier that cannot
accumulate, so old devices and DLL-free builds are unaffected.
- verified offline: DirectML/batch_accum_check.cpp drives the real exports
against an independent reference; at B=1 the accumulator matches the
shipped unbatched kernel's own gradient to 1.1e-16. Math only - the
in-situ check remains the per-layer dW/W report on a real era.
F6 - ComputeFirstLayerWidth budgeted against the RAW input width even where a
conv/LSTM front end had already reduced it, so an LSTM's dense stack was
charged for 1,280 inputs when it receives 64. Confirmed from the deployed
.cfg files: CONV, LSTM and HYBRID were all pinned at the 16-unit floor. Now
budgeted against the front-end output and capped at it (never fan out), with
the derivation reordered so both stages settle first.
N1 - EventCode/EventPhase/StructuralPhase are signed categoricals packing
direction and Wyckoff stage into one scalar across a sign discontinuity. Split
into direction + [0,1] magnitude, the same convention the base OHLC block uses.
Information-preserving; 13 readings now occupy 16 inputs.
Compiled clean (0 errors, 0 warnings); both DLLs rebuilt.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
AutoTuneIndicators now defaults to FALSE, and the 33 AD/Wyckoff parameters
it used to search are now inputs.
WHY THE DEFAULT FLIPPED - not because the search is broken. It is correct,
and its own Sidak gate is what proves it: 324 candidates per model on
SP500 H1, "no improvement" on all four topologies (0.00236 -> 0.00236 on
the AD configs, 0.00370 -> 0.00370 on PAI), winner rejected at p=1.0000.
It cannot do better here by construction - it ranks candidates by MARGINAL
MI, and the headline MI is 0.00370 nats against a shuffled null of
0.00379 +/- 0.00061 (p=0.4975), so every candidate is a noise draw and the
maximum over N of them is noise too. The cost is 45-56 min per model in
one synchronous call with no yield, and it was the amplifier for the
handle leak fixed in 33f106d. The EA's own report says it plainest: "no
per-feature indicator retuning will help."
THE INPUT STAYS. TuneIndicatorsByFilter is one function of twelve in
AIBase/AutoTune.mqh; the other eleven are the MI/lag/excursion/geometry
diagnostics that produced every verdict this project relies on, and they
run regardless of this flag. Removing the input invites removing the file.
WHY THE INPUTS WERE NEEDED. All 33 were literals in CADIndicatorTuner's
constructor with no input of any kind, while MA/RSI/MACD/Ichimoku have had
their periods exposed from the start. On the AD configs those indicators
contribute 28 of 64 features per bar. Survivable while the tuner searched
them; indefensible with it off, where they would freeze at values nobody
chose.
CONSOLIDATED 33 -> 18. volClimax/volHigh/rangeClimax/rangeSignificant/
stVolRatio/atr were duplicated verbatim across CumulativeDelta, Wyckoff
Events, Failed Structure and Bar Inversion - the same constants restated
3-4 times. One concept, one input. They are SEEDS: each fans out to the
indicator's own struct field, so with the tuner on it retains full
per-indicator freedom to move them apart. Same contract as PeriodMA.
NO RETRAIN. Every default is byte-identical to the literal it replaces,
and the fingerprint's new ADP token is appended ONLY on deviation
(MACD/Ichimoku/BN/XA convention), gated on the AD features being enabled.
At defaults the token is absent, so every model on disk keeps its filename
and stays loadable. Without that guard, merely EXPOSING these parameters
would have re-keyed every config and forced a from-scratch retrain of all
four topologies for a change that alters no number anywhere.
All-or-nothing rather than per-input, so the token can never encode a
partial picture of what the features were built from.
Compiles clean: 0 errors, 0 warnings.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
BuildFeatureWindow() replaces eight hand-rolled copies of the same loop
and feeds the window OLDEST BAR FIRST. Every copy fed it newest-first,
because MQL5 timeseries indices run backwards and `r + b` with b ascending
walks into the past.
Harmless for PAI and CONV - a dense layer learns a weight per position
either way, a conv learns time-mirrored kernels. Not harmless for the
recurrent stacks:
- LSTM_SeqStepForward reads `inputs + t*Iw`, so step t is block t.
- It writes output[] only when t == steps-1: the visible output IS the
last hidden state.
- c_t = f*c_{t-1} + i*g decays toward the start of the sequence.
lstm_seq_flowcheck.cpp measured block 0's influence on the output at
1.2e-2 of block T-1's, at the shipped forget bias of 1.0.
So the bar being PREDICTED sat at the far end of the decay and the output
was handed to the OLDEST bar in the window - the exact inverse of what the
window is for. ~80x backwards on LSTM and HYBRID, on all three tiers
(OpenCL kernel, CPU DLL, pure-MQL5 inference), which is why it never
surfaced as a backend discrepancy.
This does not create edge - the MI diagnostics read at the noise floor
(p=0.4975) with a working positive control. It makes the one hypothesis
those diagnostics explicitly do NOT cover testable: they are marginal and
per-bar, and state they "cannot rule out one that only exists in
combination or across time". The sequence model is the instrument for
across-time structure and it has been crippled, so that hypothesis has
never been honestly tested.
Fingerprint gets an unconditional |WIN:2 - the vector keeps its shape and
its features, so a stale .nnw would load cleanly and run a model fitted to
one ordering against the other, silently. Re-keying every config is the
point, not collateral damage. FORCES A FULL RETRAIN.
Also: the now-relative bar caches are re-keyed on the two live paths.
EnsureBarCachesCapacity() was only ever called from training paths, but
once m_trainingComplete is set ScheduleTrainingIfNeeded() routes every bar
to RefreshConvergedSignal() and Train() is never re-entered - so nothing
cleared the feature cache again for the life of the process. A chart that
trained to convergence kept replaying the rows computed for the last
training era's bar grid: the live signal froze at its convergence-time
value, and OnlineLearnStep() backpropped those stale features against
freshly resolved labels. Backtests were never affected (an inference-only
process never allocates the arrays, so every read recomputes).
Compiles clean: 0 errors, 0 warnings.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Three enums left the Inputs tab. They were three things a user had to pick and,
in the tester, three more axes for a genetic optimization to overfit.
Entry_Multiplier is pinned to MARKET. Its pending modes place the entry at a
LEVEL while the rest of the pipeline measures from the bar open - the exact
mismatch that manufactured the +0.097 R "retail fade" result later retracted as
a fill artifact. This codebase's fill model cannot honestly simulate a pending
entry, so it is no longer offered.
SL_Mode/TP_Mode become a STARTING pair. ReportBarrierGeometryScan now ADOPTS its
winner instead of printing "set SL_Mode/TP_Mode to X and retrain":
- only when it clears the family-wise gate from 04ee2e1 (beat the null of the
MAXIMUM, not merely the incumbent). This is why that gate had to land first:
without it, removing the inputs would hand a noise-picked geometry direct
control over the training target with no human in the loop - strictly worse
than the input it replaced. On SP500 H1 today it does NOT clear (p=0.1463),
so 2:6 is what you get - now chosen by measurement rather than assumed.
- only at m_eraCount == 0. Relabelling a partly-trained net moves the target
out from under weights already fitted to the old one.
THE GEOMETRY LEFT THE WEIGHTS-FILENAME HASH, because it is now measured. Same
rule that moved the horizon and the derived topology values out: a filename
keyed on a measured quantity changes the moment the measurement does - a few
more bars shift which pairing wins - and the EA then looks for a file that does
not exist, starts from era 0 and orphans a trained model silently. It is PINNED
IN THE .cfg instead: appended at the end (the only backward-safe change),
length-guarded like the 2026-07-30 derived pair, and ADOPTED on load rather than
compared, so a trained model keeps the barriers it actually learned and never
re-measures.
Two traps closed while wiring it, neither of which announces itself:
- m_barrierHorizonResolved latches the horizon ONCE PER PROCESS. Adopting 2:8
(wants ~192 bars) after it settled for 2:6 (128) would label the new target
against the old ceiling - the truncation fixed in 168422f, where every model
learned "target within 128 bars" while the EA holds to SL/TP. It lands in
Neutral, not in the timeout counter watching for it. Unlatched on adoption,
along with the label cache the old barriers filled.
- the .cfg adopt runs at init, before the horizon latches and before any label
is computed, so a resumed model has its pinned pair in place first. Verified,
not assumed.
FORCES A FULL RETRAIN: the fingerprint change orphans every existing .nnw.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Adds spread/ATR and the spread change ratio as network inputs (EnableSpreadFeature,
default on). Spread is the one microstructure channel that is both FX-available and
genuinely historical in the Strategy Tester - "during testing, the spread is not modeled
but is taken from historical data" - so unlike swap, signed tick flow or depth of market it
is something a backtest can honestly validate.
What it encodes, stated precisely because the raw measurement overstates it.
research/test_spread.py found spr/atr the strongest single feature in this codebase, on 5
of 8 instrument/geometry cells at 2-4x any volume feature. But the barrier LABEL charges
the spread inside its own barriers, so a wide-spread bar is mechanically likelier to
resolve as a loss and the feature would partly be predicting its own cost model. Relabelling
at zero cost and re-measuring the identical feature showed 20-40% of it WAS that tautology
and the majority was not (XAUUSD retained 97%). What survives is a volatility-regime
reading: spread is near-fixed while ATR is not, so the ratio runs high exactly when
realised volatility is below its own ATR estimate, which genuinely predicts whether
ATR-scaled barriers get reached. It is UNSIGNED - Neutral-vs-directional only, never a side.
Also fixes a stale-index bug I introduced with the cross-asset panel and had just repeated
in the spread series. Both cached on length alone:
if(m_crossAsset.Bars() >= bars) return true;
MQL5 series indices are relative to NOW, so one new closed candle shifts every index by
one. Keyed only on length, the panel keeps serving its index 0 as a bar that is no longer
the newest, and every cross-asset value is read one bar out of step with the price features
sitting beside it in the same vector - silently, with no error and no shape change. This is
the same class of defect as the dtStudied watermark behind the zero-direction backtests.
Both now carry a datetime anchor on m_Time.GetData(0), the same invalidation key the
label/feature bar caches already use.
And a performance fix that fell out of it: with correct invalidation the panel rebuilds on
every new bar, and RefreshConvergedSignal runs per bar - which in the tester would mean one
full multi-symbol resample per simulated bar at training depth. Inference only reads bars
0..m_historyBars-1 plus the panel's own slow window, so it now requests exactly that. The
cache check is >=, so a deeper panel left from training still satisfies it.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
The block fed exactly one number: (v[i] - v[i-1]) / v[i-1]. That is the first difference,
and it cannot express three things that matter - the LEVEL relative to a baseline (two
dead bars and two frantic bars both read ~0 change), and the two volume-vs-range
interactions, where heavy participation that went NOWHERE (absorption) and heavy
participation that travelled (continuation) mean opposite things and currently collapse
onto the same value.
research/test_volume.py measures each candidate's mutual information with the triple-
barrier label across 3 instruments x 2 geometries, against a BLOCK-permutation null -
blocks sized to the barrier horizon, because adjacent labels share almost their entire
outcome window and a free shuffle yields a null so tight that everything looks
significant. Finite-sample MI bias (~7/n here) is reported alongside rather than
subtracted, since the permutation null already absorbs it.
Result: volLevel beats the shipped change ratio outright on 4 of 6 cells (EURUSD 2:3
+0.000118 excess at p=0.006, USDJPY 1:2 +0.000284 at p=0.002); absorption is the single
strongest reading anywhere in the sweep at EURUSD 1:2 (+0.000404, p=0.002) though it is
null on XAUUSD; vol x range clears on 4 of 6. The shipped change ratio is itself
significant on 5 of 6, so it stays.
Kept OUT: a session-relative z-score against the same hour-of-day's own recent history.
It was the weakest candidate - null on both EURUSD cells - and it is the only one needing
per-hour rolling bookkeeping in MQL5. Not worth the state for a reading that did not
survive its own null on the primary instrument.
Magnitudes, stated plainly because they are the point: the excess MI is ~2e-4 nats against
a label entropy near 1.05. That is under a tenth of one percent of the label's
uncertainty. It is real, it repeats across instruments, and it is nowhere near an edge -
this is worth having because it costs one 50-bar loop, not because it changes the answer.
Prior work stands: the whole single-series feature family measured at the noise floor.
m_neuronsCount is already in the fingerprint, so the width change re-keys existing caches
by itself, which is correct - the input vector genuinely changed shape.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Two defects surfaced by research/test_classic.py, both verified fixed by re-running the
transcription against 178k bars of EURUSD H1.
CSignalMA model 1 could never fire. For any recursive average - and MA_TYPE_EMA is the
shipped default - MA(i) = a*Close(i) + (1-a)*MA(i+1), so
DiffMA(i) = a * (Close(i) - MA(i+1))
DiffCloseMA(i) = (1-a) * (Close(i) - MA(i+1))
are positive multiples of one quantity and always share a sign. Model 1 asks for a close
BELOW a RISING average, which is precisely the combination that identity forbids: 0.000%
of bars, either direction, any symbol. The MQL5 standard library this was ported from
defaults to MODE_SMA, where the two are merely correlated - the bug arrived with the EMA
default, not with the port. Reading the slope one bar back (DiffMAPrev) breaks the tie for
every MA type while keeping the model's stated meaning. Now fires on 7.92% of bars.
CSignalIchimoku model 11 fired on 27% of bars at weight 100. Sanyaku is three standing
STATES conjoined with no transition term, so it held across long stretches - and being
last in the if-chain at the top weight, the module's highest-conviction reading was also
its most common one, overwriting all eight event models below it on a quarter of all bars.
The old comment rejected an event form because "demanding all three flip on the same bar
would fire almost never" - true, but that is not the alternative. Kouten is the TURN: the
ALIGNMENT transitions, and only one role need change for it to. Testing !Sanyaku(idx+1)
fires once per aligned stretch. Now 2.17%, in line with Kumo breakout (2.4%) and the
strong TK cross (1.1%). DataReady() extended one bar deeper to cover the lookback.
Neither pattern showed edge before or after; this is about the models meaning what they
say and the vote not being dominated by a constant.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
The bottleneck on this project has never been the modelling - it is that
every hypothesis costs a compile, a deploy, an attach and a log read, and
answers exactly one question. Days have gone into questions that are
seconds of arithmetic once the data is in hand.
Adds a RESEARCH-ONLY build, gated behind WARRIOR_EXPORT_FEATURES and
never compiled into a shipped binary, which writes two things to
Common\Files\Warrior_EA\Research\ and then does nothing at all:
<symbol>_<tf>_features.csv - one row per bar: index, time, OHLC, ATR,
and the m_neuronsCount feature values. Exactly what the network sees.
The raw bars ride along on purpose: with OHLC and ATR offline, every
barrier geometry, horizon and in-trade target is recomputable without
MetaTrader in the loop.
<symbol>_<tf>_rates.csv - raw OHLCV across a grid of 8 symbols x 5
timeframes. The 26 engineered features only exist for the attached
chart (indicator handles bind to PERIOD_CURRENT); raw rates do not, so
ONE attach yields the whole research grid. The bar time also makes
session/hour/day-of-week derivable - the only inputs in play that are
not a transform of the same OHLCV series.
Safety, because this binary gets attached to a chart on a LIVE ACCOUNT to
reach real history:
- OnTick returns immediately, so Expert.OnTick() - the entire trading
path - is unreachable regardless of the AlgoTrading toggle, the
signal state or the inputs. Structurally incapable of sending an
order, not merely unlikely to.
- No config lock. It never trains and never saves a model, so it has
nothing to protect against a concurrent chart - and taking the lock
would make it refuse to start exactly when the config it wants to
read is already open, which is when it is most useful.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
The corrected geometry scan exposed something bigger than the geometry
question it was asked. Every pairing from 2:6 upward came back CLAMPED -
including 2:6, the SHIPPED configuration.
First-passage time for a driftless walk leaving [-m,+k] goes as m*k, and
the measured swing median here is ~12 bars at m*k=1, so 2:6 wants ~144
bars and 3:10 wants ~360. The ladder stopped at 128. A clamped label
stops meaning "does the target come before the stop" and quietly becomes
"...within 128 bars", while the deployed EA holds until SL or TP with no
bar limit. So the target the models have been trained on all along was
not the strategy the EA executes, and the trades it silently reclassified
as Neutral were the SLOW WINNERS - precisely the ones a 1:3 barrier
exists to capture. Timeout share stayed ~0% throughout, which is why this
never showed up: the truncation lands in Neutral, not in the timeout
counter that was watching for it.
Ladder extended to 384 (12..128, 192, 256, 384) so every selectable
geometry gets an honest horizon. Cost is one embargo of at most 384 bars
out of ~38k.
Second fix, same class of error as the H(Y) one: the scan's "best
eligible" was 2:2, a 1:1 barrier, against a shipped Min_Risk_Reward_Ratio
of 1:2. Training four topologies on that target would have produced a
model whose every setup is rejected at the door - the exact failure
behind four consecutive Market rejections for "no trading operations".
Sub-minRR geometries are now ineligible and marked [<minRR], printed
rather than hidden.
Also drops the dense-depth tag from the display name ("Perceptron 3L" ->
"Perceptron"). Depth is derived, so it names nothing a user chose; the
config tag [PAI-0be2] already disambiguates concurrent charts and does it
for every input rather than one. Full topology still logged by "config -".
Compiles 0 errors / 0 warnings, standard and Market. Build tag
horizon-384-v1. Changes the LABEL for every geometry, so the next scan
supersedes the previous numbers - and a retrain is required before any
model trained under the truncated target means anything.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
- Moves CLayer neuron construction to AI/Impl/Layer.mqh to keep Network.mqh clean
- Unifies four previously duplicated architecture initialisation blocks (MLP/CONV/LSTM/HYBRID) into a single shared function
- Eliminates risk of behavioural drift where one architecture missed a setter, causing mismatched feature sets or targets
