You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
Commit 7ec2f15
Browse filesBrowse the repository at this point in the historyBrowse files
Re-anchor documentation citations on 288f189; note the harness applied check reads timestamps only
The rebase onto 288f189 moved cited lines in the adapter boundary ADR and the
native feature parity page; the citations are carried across by line map.
The report quote page now says the harness APPLIED line is inferred from the
terminal timestamps, which is exact for a quote one interval after the mark
and not for a quote stamped at the mark itself.
Co-Authored-By: Claude Opus 5.5 (1M context) <[email protected]>
Copy file name to clipboardExpand all lines: docs/adr/0001-kernel-adapter-boundary.md
+1-1Lines changed: 1 addition & 1 deletion
Display the source diff
Display the rich diff
Original file line number
Diff line number
Diff line change
@@ -836,7 +836,7 @@ The table records, at the tree the doc wave read, what each block described and
836
836
|`PhysicalExecutionContext::account_fx``include/pineforge/execution.hpp:92`, read by `build_close_trade_with_costs``src/engine_orders.cpp:70-87` and handed by `append_open_position_report_rows``src/native_execution_consumer.cpp:8165-8186`| a closing or mark-to-market row converts at the account-currency rate its execution names -- a mark at a report point's instant, a fill at its own cursor -- and at the presented clock's rate when it names none, as before (R5 lane B-ADAPTER, item 3) | the rate is data a host of any venue supplies with the row it asks for; nothing branches on a platform, and the row's currency is the basis its owner's magnitudes are recorded in | none: no spec field; a caller names the rate per call, and a call naming none converts exactly as before |`tests/test_adapter_range_end_fx.cpp` (a stepped FX curve at seven instants around the range end, plain and aggregated charts) |
837
837
|`NativeExecutionConsumer::execute_current``src/native_execution_consumer.cpp:6780`| a current execution converts at its cursor's rate, threaded through its terms facts, inspection, preview and settlement contexts, and neither writes nor restores the host's presented clock, so its own hooks and the host after it see the frame's clock; its preview `inspect_current_execution` still converts at the presented clock (R5 lane B-ADAPTER, item 5) | an execution converting at its own instant's rate is the same for every host and venue, and the host's presented clock stays its own | none: the rate is the one the run already chose (the cursor's instant), carried as data |`tests/test_native_current_execution_rate.cpp` (kernel-only) |
838
838
| `NativeExecutionConsumer::validate_current_execution` `src/native_execution_consumer.cpp:6505` | a current execution refuses a request bound to a host roster (`BindCohort`) as `UnsupportedRequest`, before its point is taken and whatever the roster holds: it admits exactly the owners the evaluation's current shape executes (`current_shape` `src/native_order.cpp:383`: `Independent`, `BindOpening`, `BindOpenings`). Since R4 slice C it had admitted `BindCohort` as well, which the evaluation then left unevaluated, so a roster holding a live lot failed the run with "native current evaluated allowance mismatch" and an empty one answered `UnreadyOwner`; its read-only preview (`inspect_current_execution`) settled what the execution could not (R5 lane PAR-ORDERS-2) | one legality rule, the evaluation's own, decides a command before any point is taken, the same for every host and for the C surface (`strategy_native_execute_current_v1` answers through this validation); a roster read at the match is not a membership fixed at the command, which is what the documented contract names (`docs/pages/native-engine.md` "Selected exposure and current execution") | none: no spec field or hook. A refused request stays queued and fills at the next match like any market close; a host that wants a cohort's lots closed at once fixes its roster first (`BindOpenings`), as the Pine adapter's immediate closes do | `tests/test_native_current_cohort_refusal.cpp` (kernel-only: a live and an empty roster refused, then filled as queued closes; a fixed roster executing at once) , `tests/test_native_cohort_sliced_close.cpp` (its preview rows: the roster close's preview is refused, the same selection's settles) and `tests/test_adapter_live_state_equivalence.cpp` (reversals seed 3246599, which the failure used to end) |
839
-
| the chart symbol's daily partition -- `prepare_chart_day_partition` `src/engine_aux_security.cpp:354`, kept in `chart_day_partition_` `include/pineforge/engine.hpp:1218` and presented to the language runtime through `RuntimeAmbient::day_partition` `src/runtime_ambient.hpp:38` (the TradingView comment above `NativeDayPartition` in `include/pineforge/timeframe.hpp`) | on an intraday chart whose run installed its own symbol's `D` feed, the chart-level D period follows that feed's stamps: the exchange's trade-date bars, a holiday-merged day included. It governs `time("D")`, `timeframe.change("1D")`, `ta.change(time("D"))`, the default-anchored `ta.vwap` and `crosses_boundary(DAY)`. Without the feed, on a 1D chart or on another symbol's clock every rule is the nominal session day | the stamps are the venue's own data, as for the authoritative-feed partition above: another venue is driven by installing its own daily bars, and nothing branches on a platform | the policy knob is installing the feed or not, so a "which daily partition" field would restate it. Only the Pine host builds it (`prepare_chart_day_partition` `pine_strategy_host.cpp:1431`, `:1472`) and presents it for each bar it publishes (`AmbientDayPartitionScope` `pine_strategy_host.cpp:1946`). A bare host never calls the builder, so its calendar stays nominal | `tests/test_o_close_pct_day_anchor.cpp` (the partition is built from the D feed, and none without it) and `tests/test_chart_day_memo.cpp`; the tape is `lab tv` `o-cme-dayanchor-full`, recorded at the site |
839
+
| the chart symbol's daily partition -- `prepare_chart_day_partition` `src/engine_aux_security.cpp:354`, kept in `chart_day_partition_` `include/pineforge/engine.hpp:1218` and presented to the language runtime through `RuntimeAmbient::day_partition` `src/runtime_ambient.hpp:38` (the TradingView comment above `NativeDayPartition` in `include/pineforge/timeframe.hpp`) | on an intraday chart whose run installed its own symbol's `D` feed, the chart-level D period follows that feed's stamps: the exchange's trade-date bars, a holiday-merged day included. It governs `time("D")`, `timeframe.change("1D")`, `ta.change(time("D"))`, the default-anchored `ta.vwap` and `crosses_boundary(DAY)`. Without the feed, on a 1D chart or on another symbol's clock every rule is the nominal session day | the stamps are the venue's own data, as for the authoritative-feed partition above: another venue is driven by installing its own daily bars, and nothing branches on a platform | the policy knob is installing the feed or not, so a "which daily partition" field would restate it. Only the Pine host builds it (`prepare_chart_day_partition` `pine_strategy_host.cpp:1455`, `:1496`) and presents it for each bar it publishes (`AmbientDayPartitionScope` `pine_strategy_host.cpp:2023`). A bare host never calls the builder, so its calendar stays nominal | `tests/test_o_close_pct_day_anchor.cpp` (the partition is built from the D feed, and none without it) and `tests/test_chart_day_memo.cpp`; the tape is `lab tv` `o-cme-dayanchor-full`, recorded at the site |
840
840
| the bar-time accessors -- `_decompose_bar_time` and `_bar_hour` through `_bar_weekofyear``include/pineforge/engine.hpp:1066-1093`| the presented bar's stamp broken into calendar fields in UTC, memoized per stamp | a function of the stamp alone, and UTC is the engine's storage timezone | none: they read no spec field and fold into no digest. They are protected `BacktestEngine` members, part of the codegen contract. Their comment claimed that generated code reads them for Pine's bare `hour` / `minute` / `dayofweek`. It does not: the codegen transpiler emits `pine_hour(current_bar_.timestamp, syminfo_.timezone)` and its siblings on the exchange timezone (`pineforge_codegen/codegen/tables.py` at the corpus's pinned commit `f3285d79`), which are the language calendar functions above. No generated strategy in `corpus/` names a `_bar_*` accessor. The comment now says so (R5 lane H-DOCGATES) |`tests/test_chart_timezone.cpp` (`_bar_hour()` is the UTC hour) and `tests/test_intraday_rollover_chart_tz_l4a.cpp`|
841
841
| the conventional offset spellings of `timezone_accepted``include/pineforge/native_calendar.hpp:208` (its comment's "TV sign" line) |`UTC+05:30` and `GMT-4` mean what they say: five and a half hours ahead of UTC, four behind (ISO 8601's sign), where a POSIX `TZ` string of the same shape means the opposite; a POSIX rule string keeps POSIX's own sign | a zone spelling is data, and the sign is the one a venue printing `UTC+8` means; nothing branches on a platform | a sign-convention switch over a spelling whose meaning is fixed |`tests/test_native_calendar.cpp` (`resolve_civil("UTC+05:30", …)` and `resolve_civil("GMT-4", …)` against their fixed instants) |
842
842
| the grid and lot helpers -- `tick_grid_price` `include/pineforge/engine.hpp:799`, `price_grid_decimals` `include/pineforge/engine.hpp:833`, `level_on_price_grid` `include/pineforge/engine.hpp:843`, `apply_qty_step` `include/pineforge/engine.hpp:932`, `apply_exit_qty_step` `include/pineforge/engine.hpp:952` (their TradingView calibration comments sit above each) | nearest-tick and price-grid quantization and the lot-increment floors, each calibrated on TradingView tapes | **not a kernel mechanism**: no kernel or source-layer code calls any of them (`tests/test_f6_dead_kernel_members.cpp`, witness 2). The adapter runs its own copies of TradingView's per-order-kind tick rules (ruled native-only in the `price_grid` row above) and sizes its own lots; a native host declares a price grid and a quantity grid in its run spec | nothing to spell. They are rule 5's legacy, kept only because the twin-parity freeze (`scripts/check_twin_parity.py`, base `ab9714be`) holds the checks of the suites that reach them, and they leave with those checks | `tests/test_stop_tick_rounding_l4d.cpp`, `tests/test_level_grid_snap_l4d.cpp`, `tests/test_qty_step_epsilon_floor_l4b.cpp` |
0 commit comments