Another symbol's bars reach a run as a feed: kernel instrument feeds + Pine foreign request sites (XSYM-D) - #292
Merged
Merged
Conversation
… interval, and the Pine adapter's request sites of other symbols Kernel (ADR-0001 rule 3, opt-in). NativeRunSpec::instrument_feeds holds other instruments' own bars, keyed by (instrument, timeframe), each bar with its own close (close_ms) and optional named columns; a NativeSeriesSource::InstrumentFeed series names one. On each accepted input the series is handed, in feed order, every bar not yet handed over that has closed by the input's NativeInterval::last_traded_close_ms (lookahead off) or opened by its open_ms (lookahead on); gaps is the subscription contract's; bars before the first input ride on it. No calendar is guessed for the instrument; the chart side reads the input's calendar interval (resolved from the label on a raw-label partition, whose inputs carry no duration). With no feed installed every digest, delivery and run is what it was. Streams refuse a spec that installs one. Nine named refusals (NativeRunSpecError/Field, with C twins so the translation stays exhaustive); C has no route (1.0 boundary row). Adapter. register_security_eval(sec_id, symbol, tf, input_tf, lookahead_on, gaps_on, ignore_invalid_symbol) registers a site of another symbol; the host projects its installed feeds as the spec's instrument feeds, declares one InstrumentFeed series per site after its same-symbol sites (which keep their route), and runs the payload on every bar handed over in the requested context (history and TA over the context's bars, bar_index, time_close and syminfo.* the context's). gaps_on clears at the bar open, before the COOF checkpoint. An invalid symbol reads na under ignore_invalid_symbol and fails closed without it; a missing feed fails closed naming symbol and timeframe; an aggregated chart is refused. Recorded request series: key -> chart-bar open -> value, na where the tape has no row. C ABI: strategy_set_symbol_feed / _feed_column / strategy_set_symbol_facts / strategy_set_recorded_series, each behind its own PINEFORGE_HAS_*_V1 probe, reached through BacktestEngine virtuals whose kernel base answers false; a source host refuses stream_begin while any holds data, and check_c_abi_runtime.py pins that each declaration says so. Tests on synthetic bars only: test_native_instrument_feed (kernel, the report's pinned cases), test_native_instrument_feed_twin (kernel/adapter twin), test_symbol_request_data (C ABI, recorded series, stream refusal). Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
…INEFORGE_REQUESTS_ROOT The environment contract of workflow docs/xsym-requests.md: when PINEFORGE_REQUESTS_ROOT is set and <root>/<basename of the strategy dir>/ requests.json exists, the harness holds the manifest to pineforge-probe-requests/v1 (the refusals of the workflow's validateProbeRequests) and to this probe, verifies every file it names under the probe's own files/ against its sha256 (a feed against its byte count), parses the feeds (an empty or NaN volume or extra value is NaN), the recorded tapes and the symbol facts, and installs them through strategy_set_symbol_facts / _feed / _feed_column / strategy_set_recorded_series before the run. It reads nothing else: never another slug's directory, never the case-wide PINEFORGE_PINE_LIBRARIES. The run provenance records each feed's (key, sha, bars), each tape's (key, sha, rows) and the manifest's sha, as it does for the aux and native security feeds. A malformed manifest stops the run by name before any engine call; a library without the setters and the docker runner are refused. Unset, or no manifest for the probe: the engine is handed exactly what it always was. The frozen verifier names the probe's build directory BUILD / slug (pineforge-lab 3bac0b7b scripts/verify-engine-local.py:1695) and runs this script on it (:1593-1596, :1614-1617) through _run_timed (:190-193), whose subprocess.run passes no env=, so the runner's environment reaches it. Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
…nstrument-feed contract native-engine.md: the field list, the Limits paragraph and the stream-limits bullet state the new contract; a new section "Instrument feeds: another symbol's bars" gives the struct, the delivery rule, the validation table, the stream refusal and the Pine adapter's use (the symbol-key overload, the payload in the requested context, ignore_invalid_symbol, the fail-closed refusals, recorded request series, the run_strategy.py environment contract); the 1.0 C boundary table has the "instrument feeds" row its checker pins. ADR-0001: the rule-2 ruling row "Wave XSYM-D" (mechanism: another venue is driven by changing data only; knob: none -- lookahead and gaps are already the subscription's delivery rules), calibrated by XSYM-DESIGN's 203/203 merge pins and this lane's two witness reproductions; the HTF paragraph, the run-spec paragraph and the adapter's section name the new surface. mtf.md, coverage.md and README: the other-symbol surface, the recorded fundamentals store, the four C setters; the runtime export counts move 58 -> 62 (66 -> 70 declared, 109 -> 113 in total). Sentences this commit removes, each superseded by the contract it states: - ADR-0001 (5691220, lane L14-C): "request.security-style series are reachable from a bare host, as declared series of the run's own symbol" -- another symbol's bars are reachable now too, and the paragraph says so. - coverage.md (cc4a56e): "Runtime owns the security state machine, ratio / calendar aggregation, lookahead / gaps semantics, lower-TF emulation, and per-security diagnostics", "The runtime owns same-symbol security computation" and "Why not done yet: Most strategy logic in our test corpus relies on the chart symbol's bars + indicators" -- the request.security row and its section state the other-symbol surface instead. - coverage.md (9507d8f): "Would need a parallel data-ingestion path so the user can supply a CSV / Parquet of fundamentals", "Future story: Fundamentals and economic indicators would land as a generic auxiliary timeline feature" and "Fundamentals- and macro-driven strategies are a meaningful but smaller user segment" -- the recorded request series store and the pinned requests manifest are that ingestion path. Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
…'s four rows The rebase onto 1e360bd (INT27 with its round-2 fix) shifted cited lines: reanchor_doc_citations.py --ref 1e360bd plus check_doc_anchors.py --fix and four hand fixes bring every anchor back (1725/1725; the diff changes digits only). Floors, counted with ctest -N on the lane's tree under each ci_verify profile's cache definitions: kernel 292 registered, 291 run (+2: test_native_instrument_feed, test_run_strategy_requests; the WebSocket row still skips) release 720 (+4: the two above, test_native_instrument_feed_twin, test_symbol_request_data) debug/sanitizers/native registration 701/701/710 (+4 each) KERNEL_MIN_TESTS 289 -> 291, RELEASE_MIN_TESTS 716 -> 720, EXCLUDED_REGISTERED_MIN 697/697/706 -> 701/701/710; docs/ci.md and the floor citations follow. Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
…d); setters refuse a run in progress; the literal input_tf trap deleted An independent review of 1e360bd..80236d5a found, and this commit fixes: - The workflow keeps a request that was na on every chart bar as a tape with a header only (campaign/tv/xsym/decode.mjs keeps non-na rows), and pins it. strategy_set_recorded_series(key, ..., 0) erased the key, so the first read failed the run as "no recorded series is installed". n == 0 now installs a series without rows: every read is na. strategy_set_symbol_feed(..., 0) likewise installs a feed without bars, whose sites read na, instead of leaving the symbol unfed. - The four setters answer -1 while a run is in progress (a stream's warmup or realtime leg included), naming it: the source host's mutation guard is inert, so a mid-stream install changed what the run read and its digest. - strategy_set_symbol_feed_column judges the column name the kernel's way at install (a name that is not UTF-8 was accepted, then refused at begin). - register_security_eval(id, "SYM", "tf", "literal", bool[, bool]) bound the same-symbol overload (const char* -> bool outranks std::string) and read the symbol as the timeframe. That spelling is deleted; generated code passes input_tf_ (a std::string) and is unaffected. - run_strategy.py refuses by name a manifest the schema cannot walk (JSON nested past the recursion limit, a mintick past a double, an unreadable file) and a stamp an int64 cannot hold, which ctypes would wrap silently. - abi-stability.md records the four BacktestEngine virtuals added inside v19 (as N5 did inside v18); mtf.md and native-engine.md state the empty installs, the in-run refusal and that the generated clear_security() also clears a gaps_on payload's own history (a codegen limit). Tests: test_symbol_request_data 77 -> 119 checks (empty tape, empty feed, column name, in-run refusal, the deleted spelling as static_asserts); test_native_instrument_feed 10543 -> 12375 (a raw label outside the session reads the calendar's zero-traded slot); test_run_strategy_requests 13 -> 15. Fail-before on 80236d5a's headers and library: the static_asserts fail to compile; 15 runtime failures; the harness raises OverflowError. Sentences of 4076d7d this commit rewrites: mtf.md's and native-engine.md's "gaps_on reads na on a chart bar that received no new bar" now also states that the generated clear_security() resets the payload's own history, and mtf.md's "Historical runs only: a stream refuses an installed symbol feed" now also states the in-run refusal. Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
…etached-comments census 8 -> 0) The preflight's detached-comments stage (measure_detached_comments.py --check-ceiling, ceiling 0) counted the eight-line group comment above the four PINEFORGE_HAS_…_V1 probes as detached: a second block in the gap above the first probe's own doc. Its text now opens that doc, so the census of the kernel compile closure is 0 again; the lines keep their numbers. Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
![github-code-quality[bot]](https://avatars.githubusercontent.com/u/9919?s=60&v=4){kind=small}
| import math | ||
| import sys | ||
| import tempfile | ||
| import unittest |
| from pathlib import Path | ||
| from unittest import mock | ||
|
|
||
| import run_strategy as rs |
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Another symbol's bars reach a run as a feed (XSYM-D)
A strategy that reads
request.securityon a symbol other than its chart can now run against that symbol's own bars. Before this, only the run's own symbol was addressable, so these scripts were refused.Kernel
NativeRunSpec::instrument_feeds(NativeInstrumentFeed {instrument, tf, bars, close_ms, columns}) and a new series sourceInstrumentFeedkeyed by (instrument, timeframe).stream_beginrefuses a spec that has instrument feeds.Pine adapter
bar_index,time_closeandsyminfo.*belong to the other symbol.ignore_invalid_symbolturns an invalid symbol into na.C ABI and harness
PINEFORGE_HAS_…_V1probe:strategy_set_symbol_feed,strategy_set_symbol_feed_column,strategy_set_symbol_facts,strategy_set_recorded_series. They refuse while a run is in progress.run_strategy.pyinstalls a probe's pinned request data fromPINEFORGE_REQUESTS_ROOT, sha256-verified, and records it in the run provenance. With the variable unset, nothing changes.BacktestEnginevirtuals are added inside the current C++ ABI epoch;abi-stability.mdrecords that objects compiled against an earlier v19 header must be rebuilt.PF_ABI_VERSIONis unchanged.Verification
test_native_instrument_feed: 12,375 checks;test_native_instrument_feed_twin, the kernel/adapter twin: 353 checks;test_symbol_request_data: 119 checks.The population parity sweep of this exact tree is posted as
pineforge/parity.🤖 Generated with Claude Code
https://claude.ai/code/session_01LifAW9e5J6gFx7gatuxETE