[petrvecera]

Final Stand hands the player a Technology Choice at the start of the match and at the start of every wave: three technologies to pick one from - a new unit for the Barracks, a new player ability, or a passive bonus. That is the in-match progression, as opposed to the meta progression already exported in data/fs-perks.json.

This PR adds scripts/xml-to-json/fs_technologies.py, which writes data/fs-technologies.json (~466 KB).

Where the data comes from

Source	What it gives us
ReferenceAttributes.sga -> xml/attrib/instances/custom_property_container/hoff/technologies/*.xml	the pool: one list per faction plus a shared common list the game adds to all of them
ReferenceAttributes.sga -> xml/attrib/instances/upgrade/hoff/**/technology/*.xml	each technology: ui text and icon, upgrade_type tags, THRESHOLD_MIN / THRESHOLD_MAX, what it unlocks
Data.sga -> scar/hoff/hoff_technologymenu.scar	the order of the twelve picks and the bucket each one draws from

The pick schedule is game script rather than attributes, so Data.sga is now downloaded and unpacked by the workflow too (~32 MB, added to coh3-filelist.txt). It turns out to be fully deterministic:

1 unit(bucket0)  2 passive  3 ability1  4 unit1   <- first boss wave
5 passive        6 ability2 7 unit2               <- second boss wave / bronze
8 passive        9 ability3 10 unit3              <- third boss wave / silver
11 passive       12 ability4                      <- final boss wave / gold

Only the contents of each menu are randomised: a weighted draw over the technologies tagged with that bucket, filtered by thresholdMin / thresholdMax on the passive picks.

Shape of the JSON

• meta - choicesPerPick (3), maxSlots, maxOfferingCount (a technology is only ever offered once), and picks: the twelve picks with the wave they are handed out at, the category they draw from and the bucket a technology must be tagged with to appear.
• races - the pool per faction, denormalized: the common list is already merged in and a source field says which list an entry came from, so the website does not have to join anything. Per technology: category / buckets / tags, thresholdMin / thresholdMax, weight, ui, the raw custom properties, and the squad / ability / upgrade it unlocks for cross referencing sbps.json and abilities.json.

Counts: afrika_korps 74, american 85, british 75, german 76 technologies. All 785 locstring IDs resolve against data/locales/en-locstring.json, and every pick has at least four candidates for its three slots.

Two quirks are flagged in the data and documented in the README:

• hoff_technology_ability_rocket_overwatch_ger gets "enabled": false - it is in the German list, but its ui_menu is not the technologies menu, so the game never offers it.
• hoff_technology_passive_reduce_production_fuel_cost ("Munitions Focus") gets "category": null - it has no upgrade_type at all, so no bucket matches it and it can only turn up when the game tops up a menu it could not fill.

Refactor

The instance parent-chain merging that fs_perks.py used moved to a shared hoff_instances.py, so the two Final Stand scripts do not carry two copies of it. The perk export itself is unchanged: running the pre-refactor fs_perks.py and the new one against the same xml/ gives byte-identical output.

Regenerating data/fs-perks.json against the current xml/ did pick up five stale values though (WEAPON_OVERALL_ATTACK_SPEED 0.9/0.8/0.7/0.6 -> 1.1/1.2/1.3/1.4 on one perk) - pre-existing drift from a patch, not a result of the refactor.

Testing

• python scripts/xml-to-json/fs_technologies.py --game-data ./game-data against a locally unpacked Data.sga
• Both scripts run from the repo root and from scripts/xml-to-json (the way the workflow invokes them)
• fs_perks.py output diffed against the original script's output

🤖 Generated with Claude Code

[Copilot]

Pull request overview

Adds a new extractor for Final Stand (HOFF) “Technology Choices” and wires it into the GitHub Actions workflow, while refactoring shared instance-parent-chain merging logic out of fs_perks.py into a reusable helper module.

Changes:

• Add scripts/xml-to-json/fs_technologies.py to export Final Stand technology choice pools + pick schedule into data/fs-technologies.json.
• Factor shared raw instance loading/merging + UI/custom-property parsing into scripts/xml-to-json/hoff_instances.py and update fs_perks.py to use it.
• Update workflow/docs to download + unpack Data.sga, run the new extractor, and ignore the unpacked game-data/ directory.

Reviewed changes

Copilot reviewed 7 out of 9 changed files in this pull request and generated 3 comments.

Show a summary per file

File	Description
scripts/xml-to-json/hoff_instances.py	New shared helpers for loading/merging instance variants and parsing UI/custom properties.
scripts/xml-to-json/fs_technologies.py	New Final Stand technology choices exporter (attributes + Data.sga scar script parsing).
scripts/xml-to-json/fs_perks.py	Refactor to reuse shared instance utilities; perks output logic remains.
README.md	Document new fs-technologies.json export and required Data.sga unpack steps.
data/fs-perks.json	Regenerated perks output reflecting upstream data drift.
.gitignore	Ignore unpacked game-data/ directory used by the new exporter/workflow.
.github/workflows/extract-data.yaml	Workflow now downloads/unpacks Data.sga, runs fs_technologies.py, and cleans up.
.github/workflows/coh3-filelist.txt	Add Data.sga to the downloaded file list.

---

💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.