LevelXStudios/ogc-db
3
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

Aggregator, intake with deployment, catalog, first curated entries #1

Manually merged
SkyfaR merged 0 commits from night-2 into main 2026-10-07 07:16:26 +02:00
Conversation 0 Commits - Files changed -
SkyfaR commented 2026-10-07 01:17:28 +02:00
Owner
Copy link

Four pieces, each built, reviewed adversarially and fixed separately, then merged here and tested together (cargo test --workspace: all pass; clippy clean).

  • aggregator/ (ogc-db-aggregate): raw contributions and curated measurements → the complete /v1 directory and a private report.md. Size variants, builds with a change-point rule, weekly cap, MAD outliers, weighted median/p25/p75, k ≥ 3 and ≥ 2 weeks gate for crowd cells, anchor check, confidence, recommendation (≥ medium on both file systems), tiers, anomaly gate with --approve, deterministic output. Golden datasets: flood, slow poisoning, update with change-point, bimodal sizes, gen blocklist. The curated format has a schema (model/schema/curated.schema.json).
  • intake/ (ogc-intake): unix socket only, one route, validation with ogc-db-proto, PoW and replay by stamp, shuffled weekly spool, kill switches on SIGHUP. deploy/: nginx, BunkerWeb (1.6.x) and systemd settings for the VM behind BunkerWeb and the Tailscale tunnel, so the submit host keeps no client address on disk; webhook deployment of both sites; a canary that sends requests from a marker address and searches every log, spool and state file (passes locally with nginx and with BunkerWeb 1.6.15). Owner checks are listed in deploy/README.md.
  • catalog/ (ogc-db-catalog): names of published games and copies of their Steam header images, ≤ 1 request/s, cached; the site shows the copies from our own host.
  • curated/: nine LZX measurements from the Windows test runs, each traced to its source in curated/SOURCES.md.

A local dry run on the curated entries (aggregator → catalog → aggregator with names → site) builds the real site: 9 games, all with names, images and a Windows tier. public/ is not committed yet.

Four pieces, each built, reviewed adversarially and fixed separately, then merged here and tested together (`cargo test --workspace`: all pass; clippy clean). - **aggregator/** (`ogc-db-aggregate`): raw contributions and curated measurements → the complete `/v1` directory and a private `report.md`. Size variants, builds with a change-point rule, weekly cap, MAD outliers, weighted median/p25/p75, k ≥ 3 and ≥ 2 weeks gate for crowd cells, anchor check, confidence, recommendation (≥ medium on both file systems), tiers, anomaly gate with `--approve`, deterministic output. Golden datasets: flood, slow poisoning, update with change-point, bimodal sizes, gen blocklist. The curated format has a schema (`model/schema/curated.schema.json`). - **intake/** (`ogc-intake`): unix socket only, one route, validation with ogc-db-proto, PoW and replay by stamp, shuffled weekly spool, kill switches on SIGHUP. **deploy/**: nginx, BunkerWeb (1.6.x) and systemd settings for the VM behind BunkerWeb and the Tailscale tunnel, so the submit host keeps no client address on disk; webhook deployment of both sites; a canary that sends requests from a marker address and searches every log, spool and state file (passes locally with nginx and with BunkerWeb 1.6.15). Owner checks are listed in `deploy/README.md`. - **catalog/** (`ogc-db-catalog`): names of published games and copies of their Steam header images, ≤ 1 request/s, cached; the site shows the copies from our own host. - **curated/**: nine LZX measurements from the Windows test runs, each traced to its source in `curated/SOURCES.md`. A local dry run on the curated entries (aggregator → catalog → aggregator with names → site) builds the real site: 9 games, all with names, images and a Windows tier. `public/` is not committed yet.
ogc-db-aggregate groups records by game, content and size variant, follows
builds with a change-point rule, weighs weeks with a cap, drops outliers by
MAD, publishes only cells that pass the k/week gate, and holds the run for
approval when the anomaly gate trips. The curated-measurement format gets a
schema; recommendations need medium confidence on both file systems.
The intake service listens only on a unix socket, validates with
ogc-db-proto, checks proof of work and replays, and appends shuffled records
to weekly spool files. deploy/ holds nginx, BunkerWeb and systemd settings
that keep client addresses off disk, the webhook deployment of both sites,
and the canary that proves no address is stored.
ogc-db-catalog looks up the names of published games and keeps copies of
their Steam header images, politely and with a cache; the site shows the
copies from our own host, with a fallback where none exists.
Each value is traced to its source report in curated/SOURCES.md.
# Conflicts:
#	Cargo.lock
#	Cargo.toml
# Conflicts:
#	Cargo.lock
#	Cargo.toml
SkyfaR manually merged commit f8144e13b8 into main 2026-10-07 07:16:26 +02:00
Sign in to join this conversation.
Reviewers
No reviewers
Labels
Clear labels
No items
No labels
Milestone
Clear milestone
No items
No milestone
Projects
Clear projects
No items
No project
Assignees
Clear assignees
No assignees
1 participant
Notifications
Due date
The due date is invalid or out of range. Please use the format "yyyy-mm-dd".

No due date set.

Dependencies

No dependencies set.

Reference
LevelXStudios/ogc-db!1
No description provided.