Run real-data members at the sidecar pip — the graph broker uses the synthetic pip while the manifest stamps the real one #232

New Issue

2026-07-10T17:14:52+02:00

Brummel commented

2026-07-10 17:14:52 +02:00

wrap_r builds the broker as SimBroker::builder(SYNTHETIC_PIP_SIZE) (0.0001) unconditionally (crates/aura-cli/src/main.rs:1269); SimBroker's pip is closure-captured and the node declares no params (crates/aura-std/src/sim_broker.rs:61), so it cannot be rebound after construction. On the real-data paths the resolved geometry-sidecar pip reaches only the manifest: manifest.broker = r_sma_broker_label(pip) (main.rs:1386 and 1430), and campaign members pass geo.pip_size to run_blueprint_member where it never touches the graph (crates/aura-cli/src/campaign_run.rs:296).

Effect on every real-data run: total_pips and max_drawdown are computed with divisor 0.0001 regardless of instrument — on GER40 (sidecar pipSize=1) inflated by 10^4 — while the recorded broker label claims the sidecar pip. The label misstates what ran. R metrics are pip-free (the RiskExecutor branch) and unaffected; within a family the distortion is uniform, so intra-family ranking survives; cross-instrument pip comparisons do not.

The doc comment at main.rs:73-80 ("The real path threads the sidecar's looked-up pip_size instead") is true only for the label. Fix home is the same scaffolding #231 retires — the pip belongs to the source binding.

the broker in the graph runs at the resolved instrument pip on real data (or pip-denominated metrics are dropped from real-data reports)
the manifest broker label matches what actually ran

`wrap_r` builds the broker as `SimBroker::builder(SYNTHETIC_PIP_SIZE)` (0.0001) unconditionally (crates/aura-cli/src/main.rs:1269); SimBroker's pip is closure-captured and the node declares no params (crates/aura-std/src/sim_broker.rs:61), so it cannot be rebound after construction. On the real-data paths the resolved geometry-sidecar pip reaches only the manifest: `manifest.broker = r_sma_broker_label(pip)` (main.rs:1386 and 1430), and campaign members pass `geo.pip_size` to `run_blueprint_member` where it never touches the graph (crates/aura-cli/src/campaign_run.rs:296). Effect on every real-data run: `total_pips` and `max_drawdown` are computed with divisor 0.0001 regardless of instrument — on GER40 (sidecar pipSize=1) inflated by 10^4 — while the recorded broker label claims the sidecar pip. The label misstates what ran. R metrics are pip-free (the RiskExecutor branch) and unaffected; within a family the distortion is uniform, so intra-family ranking survives; cross-instrument pip comparisons do not. The doc comment at main.rs:73-80 ("The real path threads the sidecar's looked-up pip_size instead") is true only for the label. Fix home is the same scaffolding #231 retires — the pip belongs to the source binding. - [ ] the broker in the graph runs at the resolved instrument pip on real data (or pip-denominated metrics are dropped from real-data reports) - [ ] the manifest broker label matches what actually ran

Brummel added this to the Real-project readiness — role-complete research loop (no freeze) milestone 2026-07-10 17:14:52 +02:00

Brummel added the bug label 2026-07-10 17:14:52 +02:00

Brummel referenced this issue

2026-07-10 17:16:35 +02:00

Bind strategy input roles to sources as authorable data — retire wrap_r's single-price weld (C20/C24) #231

Brummel commented

2026-07-10 18:29:02 +02:00

Fix direction decided (2026-07-10): thread the resolved pip into the harness scaffolding rather than dropping pip-denominated metrics from real-data reports — the manifest already stamps the resolved pip, so running the graph at it restores label truthfulness without shrinking the report.

Landed as eb5fc21 (RED: metamorphic pip-invariance test — total_pips * pip must agree across two distinct pips on identical input) + ed1e8a0 (GREEN): wrap_r gained a pip_size parameter passed to SimBroker::builder. Real-pip call sites pass the resolved sidecar pip (run_signal_r, run_blueprint_member, and the campaign trace re-run in persist_campaign_traces — the latter required so the determinism drift alarm compares re-run and nominee at the same pip); param-space/axis probes keep the synthetic pip explicitly (metrics discarded). The two golden stitched_total_pips pins rescaled by exactly the 10^4 bug factor while n_trades/expectancy_r stayed byte-identical — confirming R metrics were never pip-contaminated, as the issue predicted.

Both acceptance boxes are met on the branch: the graph broker runs at the resolved instrument pip on real data, and the manifest broker label matches what ran. Closes via commit on merge.

Fix direction decided (2026-07-10): thread the resolved pip into the harness scaffolding rather than dropping pip-denominated metrics from real-data reports — the manifest already stamps the resolved pip, so running the graph at it restores label truthfulness without shrinking the report. Landed as eb5fc21 (RED: metamorphic pip-invariance test — `total_pips * pip` must agree across two distinct pips on identical input) + ed1e8a0 (GREEN): `wrap_r` gained a `pip_size` parameter passed to `SimBroker::builder`. Real-pip call sites pass the resolved sidecar pip (`run_signal_r`, `run_blueprint_member`, and the campaign trace re-run in `persist_campaign_traces` — the latter required so the determinism drift alarm compares re-run and nominee at the same pip); param-space/axis probes keep the synthetic pip explicitly (metrics discarded). The two golden `stitched_total_pips` pins rescaled by exactly the 10^4 bug factor while `n_trades`/`expectancy_r` stayed byte-identical — confirming R metrics were never pip-contaminated, as the issue predicted. Both acceptance boxes are met on the branch: the graph broker runs at the resolved instrument pip on real data, and the manifest broker label matches what ran. Closes via commit on merge.

Brummel referenced this issue from a commit

2026-07-11 15:18:17 +02:00

test(cli): RED — real-data member pips ignore the resolved sidecar pip

Brummel referenced this issue from a commit

2026-07-11 15:18:17 +02:00

fix(cli): thread the resolved pip into wrap_r so real-data runs hit the graph broker

Brummel closed this issue

2026-07-11 15:18:17 +02:00

Brummel referenced this issue from a commit

2026-07-11 15:18:18 +02:00

audit: real-project-readiness run close — drift-clean, one low debt noted

claude referenced this issue