Teach Hermes: decision received, can continue revising

Your decision is saved in the browser and exportable as JSON; nothing is uploaded to a server

Published on

July 13, 2026

User standard answers received

The 2026-07-13 teaching-decision JSON selected upstream_v1.4.1_engine and gave canonical steps plus the questions that must be asked before every run. The form below is kept for continued revision:

  • which program/runner is the base;
  • which patterns to keep;
  • which historical actions are forbidden from now on;
  • which script / .par / log is the golden example;
  • what Hermes's biggest misunderstanding about extraction is.

The exported JSON will feed the next round of local implementation / skill updates, but it does not automatically authorize executing SAS.

NoteAccepted teaching decision · source revision v2
  1. Use one unified esas_playground_v1.4.1.py; different OBSIDs, bands, regions, and needs only change the .par.
  2. First fix the OBSID + RA/Dec DS9 region combination, and make the band, point-source exclusion, and extended-source exclusion explicit.
  3. Inventory how far ODF/RPC stages have progressed; fill in field reduction only when 1.calibration / 2.filter / 2.1.filter / 3A.edetect_stack are missing.
  4. Before extraction, visually confirm the region in RA/Dec sky coordinates with an evselect image.
  5. After extraction, overlay the mos-spectra image with the exact region in RA/Dec, and deliver the xsherpa inspect.

The user checked to keep all 11 safety/QA patterns. The original decision has execution_authorized=false and did not specify any run's band/region/point-source/extended-source answers, so it constitutes no run authorization.

User teaching decision

This is your teaching record, not a web poll, and nothing is sent to a server.

Shared plan ID: initializing…

1. Choose the base most worth learning
2. Patterns that must be kept
3. Patterns forbidden from now on

Nothing selected yet; content is saved only in the current browser.

The adopted canonical method

  1. canonical engine: esas_playground_v1.4.1.py;
  2. runner pattern: a single engine + .par driving; keep the XARTATOMS clean-shell / pilot / hash / status, dropping the R4–7-specific logic;
  3. product acceptance: take the minimal common set from the M31HotISM 84 gates and XART qualification;
  4. M31CGM serves as the full-reduction and failure corpus, not as a new extraction runner;
  5. the web pages only teach/audit, never execute SAS;
  6. before every run, ask: band, region, whether to mask point sources, whether to remove extended sources, and check the ODF/RPC stage state;
  7. pre/post spatial QA both use RA/Dec sky coordinates, with the post-run xsherpa no-fit inspect delivered separately.

These method rules are adopted; each concrete run still requires per-run input and visual approval.

Manual audit checklist

Downloads

中文