bd2003e217e6be8fcebb1c636294bc97c8b6b515
6 Commits
| Author | SHA1 | Message | Date | |
|---|---|---|---|---|
 Brummel
|
ff4a1b3d4a |
feat(0092): run-from-blueprint infrastructure — restructure + topology_hash + shared seam (Tasks 1-3)
Cycle-1 infrastructure for #165 (World/C21). Behaviour-preserving + workspace green; the CLI arm + tests (Tasks 4-5) follow. Task 1 — restructured stage1_r_graph into stage1_signal() (the serializable signal leg: SMA-cross -> Bias, a `price` input-role + a `bias` output) + wrap_stage1r(signal, ...) (broker/sinks/exec/cost around a nested signal), with stage1_r_graph() = wrap_stage1r(stage1_signal(...)). DEVIATION from the plan's "callers untouched" claim, verified behaviour-preserving: nesting the signal prefixes its param-space names (stage1_signal.fast.length), which broke the sweep's bare .axis("fast.length"); fixed exactly as the #137 stop-knob machinery does — FAST_LENGTH_SUFFIX/SLOW_LENGTH_SUFFIX suffix-resolution + stage1_r_friendly_name arms mapping the prefixed names back to the bare manifest names. All observable behaviour (manifest names, member keys, metrics, traces) byte-identical; the flat graph and every metric unchanged (full suite green, incl. sweep/walkforward/MC). Task 2 — RunManifest.topology_hash: Option<String> (Tier-1 optional, serde default/skip — old manifests byte-identical, #156), threaded None into all 24 literal sites across 7 crates. Also fixed the aura-registry RunManifestRead read-mirror to carry topology_hash (it was hardcoding None on load) — else a stored hash would silently drop on the registry round-trip once Task 3 stores a real one. Task 3 — the shared run_signal_stage1r seam (hash the signal, wrap, compile with params, bootstrap, run, manifest with params from param_space + topology_hash) + the sha256_hex topology_hash helper (sha2 in aura-cli, off the frozen engine, invariant 8). run_stage1_r now carries its topology_hash too (every run becomes self-identifying, #158). The seam is unwired until Task 4 (transient #[allow(dead_code)], retired there); RED-first seam tests added. Verified: cargo test --workspace green (51 suites, 0 failures); cargo clippy --workspace --all-targets -D warnings clean. refs #165 |
||
|
Brummel
|
0a06afed29 |
feat(0078): cross-instrument generalization as a validation step
Last piece of the Inferential-validation milestone (#146): a new `aura generalize` subcommand grades how consistently a single candidate holds across a set of instruments — the across-instrument axis of the same anti-false-discovery concern as trials-deflation (#144, within one family) and plateau-over-peak (#145, within one instrument's surface). Aggregator/validator, not a third selector (user-ratified): it runs a brought candidate (a single-cell stage1-r grid — --fast/--slow/--stop- length/--stop-k pinned to single values) across an instrument list (--real SYM1,SYM2,...), collects per-instrument R-metrics, and reduces them to a worst-case floor (min over instruments) + a sign-agreement count + the per-instrument breakdown. R-only (C10 — R is the only account/instrument-agnostic unit); the worst-case min is the cross- instrument sibling of PlateauMode::Worst and satisfies the non- domination constraint by construction (a strong instrument cannot lift a min). Rejected: a cross-instrument t-stat mean/std*sqrt(M), which reintroduces the pooled mean the milestone warns against. - RunManifest gains a first-class `instrument` lineage field (serde- widened, C14/C18 — legacy lines load as None, existing manifest bytes unchanged); the compat read-mirror threads it in lockstep. - FamilyKind::CrossInstrument (C12 comparison axis); the M per-instrument runs persist as a CrossInstrument family, each member self-identifying via its stamped instrument. The aggregate is recomputable from the members (McAggregate precedent), printed live. - The generalization reduction lives registry-side beside optimize_ deflated/optimize_plateau (C9), reusing resolve_metric/metric_value/ is_r_metric; check_r_metric is the data-free pre-check the CLI runs before evaluating any instrument. - Additive: existing sweep/walkforward/mc/standalone paths stay byte- identical (C23 — instrument omitted when None via skip_serializing_if). Verified by the orchestrator: cargo test --workspace green (0 failed — the new generalization_*, check_r_metric, parse_generalize_*, and the gated cross-instrument E2E all pass; the E2E ran the real GER40/USDJPY path on this host, exit 0); cargo clippy --workspace --all-targets -D warnings clean. closes #146 |
||
|
Brummel
|
a2959050a4 |
feat(0076): deflate the sweep winner's metric for the number of trials
Adds optimize_deflated beside optimize in aura-registry: the in-sample walk-forward selection now records, on each OOS winner's manifest, how much the winner's metric is inflated by the size of the search it won — a deflated score and (R arm) an overfit probability — without changing which member wins (additive, C23; a regression test pins optimize_deflated's winner == optimize's). - R arm (sqn_normalized/expectancy_r/...): a centred moving-block reality-check. Each member's per-trade R series is mean-subtracted to impose the no-edge null, resampled via the r_bootstrap kernel (extracted to a shared resample_block, byte-identical), and the best-of-K null-max gives overfit_probability = (count(null >= raw)+1)/(n+1) and deflated_score = raw - p95(null). Deterministic given the seed (C1). - total_pips arm: a closed-form expected-max-of-K dispersion floor (inv_norm_cdf + expected_max_of_normals), no probability. The record (FamilySelection on RunManifest) rides the proven serde(default, skip_serializing_if) widening + a compat.rs mirror, so legacy runs.jsonl/families.jsonl lines load unchanged and unstamped sweep/mc/standalone manifests stay byte-identical (C14/C18). The SelectionMode enum reserves a Plateau slot for #145. CLI: walk-forward stamps both arms; runs family <id> rank surfaces a human-readable deflated/overfit line. Quality review (re-dispatched after the implement-loop's quality gate exhausted) caught a real defect: on the R arm with positive resamples but no member carrying trade_rs (a family loaded from disk — trade_rs is serde-skipped), the null was all-NEG_INFINITY, turning deflated_score into +inf. Fixed by collapsing "no usable null" (zero resamples OR no trade_rs) into one degenerate floor (deflated == raw, overfit == 1.0); pinned by optimize_deflated_no_trade_rs_floors_instead_of_infinity. Verified: cargo test --workspace and cargo clippy --workspace --all-targets -D warnings both green. Frictionless Stage-1 R; n_eff advisory and reload-time recompute deferred (spec 0076 Out of scope). closes #144 |
||
|
Brummel
|
2c43296c2c |
feat(stage1-r): bias rename + stop-rule + position-management + summarize_r (iter 1)
Iteration 1 of the Stage-1 "R-based signal quality" cycle (spec 0065): a strategy's signal quality, measured in R on its unsized bias stream, feed-forward. New discrete-trade machinery, NOT a SimBroker extension. What lands: - exposure -> bias (refs #126): the Exposure node/type/file/output-field renamed to Bias (computation unchanged: clamp(signal/scale,-1,+1); the SEMANTICS change — the output is the unsized strategy bias, sizing leaves the strategy). Scoped to the semantic core; behaviour-preserving cosmetics are deferred (see below). - stop-rule nodes (refs #119): VolStop = k*EMA(|price - prev_price|) (one fused node; the volatility that defines 1R, close-to-close — true-range ATR deferred, needs OHLC) + FixedStop (constant, the test fixture / fixed-vs-vol structural-axis sibling). - PositionManagement (refs #127): the stateful heart. Latches the entry-cycle stop distance as the FROZEN R-denominator (never re-read), marks/exits no-look-ahead like SimBroker (a position earns from the next cycle; an exit realises against the cycle's close), and emits ONE dense per-cycle R-record (C8). Stop-outs are NOT capped at -1R (a gap through the stop realises R < -1 — the honest loss tail); R is computed size-invariantly (size = (exit-entry)*dir / latched_dist cancels). The trade ledger is the rows where closed_this_cycle; the R-equity is cum_realized + unrealized; a position open at window end is the last row (open=true) — explicit, never silent MtM. - summarize_r + RMetrics (refs #129): the post-run fold (NOT an in-graph node — recorders drain post-run). E[R], win-rate, profit-factor, avg win/loss R, by-trade max R-drawdown, n_open_at_end (window-end force-close). SQN / conviction terciles / net-of-cost / RunMetrics.r are iteration 2. Design adversarially hardened before implementation (12-juror refute panel; decisions on #117). Ratified implementer deviations from the plan snippet, verified by hand: - col-4 `direction` tracks the OPEN position at cycle end (window-end synthesis; names the reopened leg on a reversal), falling back to the closed trade's dir only when flat. summarize_r does not read col 4; the R-metrics are unaffected. - .named("exposure") kept on the 4 previously-unnamed Bias nodes so the auto-derived param-path stays exposure.scale (no manifest/dir-name drift) — the unnamed nodes would otherwise flip to bias.scale. - intra-doc links [`Exposure`] -> [`Bias`] in sim_broker/latch + the sample-model test pin (cosmetic, behaviour-neutral; broken intra-doc links fail cargo doc). - the E2E drives a full bootstrapped Harness (stronger than the plan's direct-node drive — exercises the real cross-crate producer->consumer seam). Deferred (behaviour-preserving, separate cosmetic pass — NOT this iteration): RunMetrics.exposure_sign_flips / Metric::ExposureSignFlips, SimBroker/LongOnly "exposure" input ports, the "exposure" tap/trace names, and the .named("exposure") instance labels. Iteration 2: the Sizer seam, the RiskExecutor composite (+ the Veto documented-seam), summarize_r enrichment, RunMetrics.r, and the CLI/recording surface. Verification (orchestrator-run, not agent-claimed): - cargo build --workspace: clean. - cargo test --workspace: all green, 0 failed (incl. the new bias/stop_rule/ position_management unit tests, the summarize_r arithmetic tests, and the stage1_r_e2e capstone + layout guard). - cargo clippy --workspace --all-targets -- -D warnings: clean (exit 0). - Keystone RED tests pass: no_lookahead_bias_exit_realises_the_held_move, stop_out_is_not_capped_at_minus_one_r, no_gap_stop_is_exactly_minus_one_r, reversal_closes_one_leg_and_reopens, open_at_window_end_is_carried_on_the_last_row. refs #117 #119 #126 #127 #129 |
||
|
Brummel
|
d858caf67b |
chore: scrub dangling references to deleted specs/plans from sources
docs/specs and docs/plans were retired (prior commit); the source/test
comments that cited them ("spec 0050 §4.1", "spec §Testing N", "per spec",
"the spec's ...") now point at nothing. Strip every such pointer while
preserving the technical substance, the design-ledger contract refs
(C1/C11/C20/C34/C12.1/...), and the Gitea issue refs (#41).
Comment/doc edits only across 20 files — no logic change; full workspace
suite green, clippy clean.
|
||
|
Brummel
|
e17d78f24f |
feat(aura-engine): random param-sweep — RandomSpace + a Space trait + typed ParamRange
Ship the random half of the C12.1 param-sweep axis (grid landed in 0028).
A new `Space` trait (`points`/`param_specs`) generalizes `sweep`/`sweep_with_threads`
from `&GridSpace` to `&S: Space`; `GridSpace`'s trait impl forwards to its existing
inherent methods, so the grid path is behaviour-preserving (C1, every pre-existing
grid sweep test stays green). `RandomSpace` is the second `Space` producer: N seeded
uniform points over per-slot typed `ParamRange { lo, hi }` ranges (I64 inclusive
[lo,hi], F64 half-open [lo,hi)), validated in `new` against the param-space — a
non-numeric slot, a range-kind mismatch, and an empty range are the three new typed
`SweepError` variants, all caught before any run. Sampling reuses the existing
bit-stable `SplitMix64` (promoted to `pub(crate)`) as a code-path-disjoint instance
from the data-edge seed RNG — the #52/#71 World-II source-seam firewall: they share
only the `u64` type, never a path. The sweep-layer signature stays param-only
(`Fn(&[Cell]) -> RunReport`); no `Source`/stream type enters it.
Deviation from the spec's verbatim sampler, accepted as a correctness hardening:
the I64 draw guards the full-width span ([i64::MIN, i64::MAX] computes a span of
2^64 that wraps a u64 to 0) and uses `lo.wrapping_add(draw as i64)` instead of a
plain `+`. This avoids both the `% 0` divide-by-zero at the full domain and the
signed-overflow panic the literal form hits on wide ranges in debug builds, while
preserving uniform-over-[lo,hi] draws and determinism (an extra RED test,
random_points_full_range_i64_does_not_panic, pins it). Modulo bias for spans not
dividing 2^64 remains the spec's documented, accepted simplification (param search
needs no cryptographic uniformity).
Includes a public-API E2E suite (tests/random_sweep_e2e.rs) exercising the feature
exactly as a downstream researcher would — reproducibility at the report boundary,
seed-sensitivity, in-range draws, the typed NonNumericRange gate, a well-formed
empty (count==0) family, and Grid/Random interchangeability through `Space`.
Verified: cargo test --workspace green (incl. 17 new engine tests + 6 E2E);
cargo clippy --workspace --all-targets -- -D warnings clean.
closes #52
|