pikers
/
piker
4
1
Fork 0
Code Issues 20 Pull Requests 13 Packages Projects Releases Wiki Activity
piker/ai/prompt-io/codex/20261005T041132Z_eed58566_t...

4.7 KiB
Raw Blame History

Prompt

The user requested an autonomous search of GitHub issues and local notes, code, documentation and comments for Trend Calculus research, then a rigorous Piker-native implementation in the FSP subsystem. They asked to prepare multiple high-frequency timeframe regimes for the JEV System-1 provider service, with work bounded by 00:18 Eastern. They authorized same-caliber subagents and requested, as a bonus, multiple-timeframe reversal markers and live rectangle annotations.

The user supplied these implementation resources:

They offered a Pine implementation but had not supplied it. They said the API key was still being set up and instructed us to continue offline. They clarified the workspace concern, disabled fast mode, and supplied repository rules: preserve human-owned task states, no autonomous commits or pushes, 69-column prose, Xonsh snippets, and the Python style skill.

Response summary

Implemented a bounded Numba recurrence and a four-order native FSP with provisional-tail isolation, skipped-step recovery, indexed publication, quiet sampler wakeups, sparse events and history replay. Added complete aligned multiple-timeframe aggregation, immutable provenance-rich snapshots, bounded HTTPX/Trio JEV requests and a paced worker that rejects stale or invalidated results. Added native feed/service wiring and opt-in Qt markers and regime rectangles, including actual 1/5/15-second layers and model receipt/expiry rules.

Research identifies pinned Lua/Scala sources, issue discussions, local notes and the manuscript, with explicit semantic differences. Original Lua fixtures and an independent materialized-window oracle check 21,870 exhaustive histories, 48 random recursive streams and 62 original Lua reversals. No complete Pine source was recovered.

The TypeSafe agent skill was installed and applied. No live API key was read, no model request was sent, no broker session was opened, and no Git history or human-owned task state was changed.

Verification

The combined targeted suite passed 252 tests, with one unrelated remote gap-dialog actor case deselected. The one warning is an existing Tractor dependency’s deprecated Trio Event truth test. The new SHM test includes an actual separate read-only child actor after in-place replay. This does not prove concurrent cross-process write atomicity. Ruff, diff whitespace and 69-column checks passed.

The combined suite covered test_trend, test_trend_reference, test_trend_shm, test_trend_regimes, test_jev, test_regime_feed, test_trend_graphics, test_trend_mtf_graphics, test_fsp_sync, test_fsp_momo and test_gap_overlays. The standard py313 environment and pytest with the xonsh plugin disabled were used. Socket/SHM execution required approved sandbox escalation.

The synthetic offscreen chart preview was generated and inspected. The reproducible benchmark measured roughly 8.586 million batch rows/second and 6.963 microseconds median scalar update on this host; these exclude rendering, transport and provider inference. Native subsecond sampling remains unsupported by the existing clock. Live provider quality and the user’s Pine parity remain unverified.

Files changed

  • piker/fsp/__init__.py
  • piker/fsp/_api.py
  • piker/fsp/_engine.py
  • piker/ui/_display.py
  • piker/ui/_fsp.py
  • piker/fsp/_trend.py
  • piker/fsp/_trend_fsp.py
  • piker/fsp/_regime.py
  • piker/fsp/_jev.py
  • piker/fsp/_regime_feed.py
  • piker/ui/_trend.py
  • piker/ui/_trend_mtf.py
  • tests/test_trend.py
  • tests/test_trend_shm.py
  • tests/test_trend_reference.py
  • tests/test_trend_regimes.py
  • tests/test_jev.py
  • tests/test_regime_feed.py
  • tests/test_trend_graphics.py
  • tests/test_trend_mtf_graphics.py
  • tests/_inputs/trend_calculus_lua.json
  • examples/benchmark_trend.py
  • examples/trend_regimes.py
  • examples/trend_graphics.py
  • docs/trend_calculus.rst
  • ai/prompt-io/README.md

Human edits

The human supplied the objective, deadline, research resources, provider choice, optional skill installation, graphics extension, parallel-agent authorization and offline credential constraint. Those directions materially determined the implementation and scope. No direct manual source edits were observed during this work. The patch is pending human review; acceptance is not inferred. Research, implementation, review and tests were performed by the root Codex agent with three same-caliber delegated agents. Existing unrelated working-tree artifacts were preserved.