screpdb is an advanced Starcraft replay reporting tool.
212.43 ms/replay · corpus: 150 replays · GitHub-hosted 2-core runner · updated automatically on merge to main
Alias list support for progamer replays (built-in, editable, importable/exportable), and automatic aliasing for local user's player names
Sophisticated command de-duping on the early game to facilitate precise build order detection and timing comparisons
See CHANGELOG.md for release notes.
⚠️ Security: On Windows, screpdb runs its worker at Low integrity — the OS confines all of screpdb's writes to a single app-data folder, so even a compromised replay/map parser cannot write elsewhere on your machine (see Security / I/O model). On macOS and Linux there is no OS sandbox yet: screpdb routes all its own I/O through in-process facades (writes confined to the app-data dir and the replays folder, no outbound network calls beyond user-initiated self-update), but these are best-effort guardrails rather than an OS boundary, so exercise judgement before running it.
Windows — recommended: install with Scoop
👉 Recommended: install with Scoop. Open PowerShell and paste these commands:
scoop install git # required by 'scoop bucket add' (skip if you already have git)
scoop bucket add screpdb https://github.com/marianogappa/screpdb
scoop install screpdbThat's it. Now run screpdb-gui (the app opens in your browser), or screpdb for the CLI.
To upgrade later, just run:
scoop update screpdb💡 Seeing an old version, or
installfails on a missing file? Your local copy of the bucket is a git clone that only refreshes onscoop update. Runscoop update(no package name) first to pull the latest manifest, thenscoop install screpdb/scoop update screpdb.
Scoop is the happy path because it downloads without a browser, so Windows won't show the "unidentified developer" / SmartScreen warning, and upgrades are one command. Don't have Scoop yet? Install it first (one line, from scoop.sh):
Set-ExecutionPolicy -ExecutionPolicy RemoteSigned -Scope CurrentUser
irm get.scoop.sh | iexPrefer a direct download? (expect a SmartScreen warning)
Grab screpdb-gui-windows-amd64.exe (the GUI; screpdb-windows-amd64.exe is the CLI) from the Releases page and double-click it.
The binaries are not code-signed, so on first launch Windows may warn you — none of these mean the binary is malicious:
- SmartScreen "Windows protected your PC". Click More info → Run anyway.
- Microsoft Defender or third-party antivirus may flag or silently quarantine the binary. Unsigned Go binaries that read files and make network requests are a known false-positive pattern. If the file vanishes from Downloads, check Defender's Protection History and restore it (or add an exclusion).
- Enterprise machines running AppLocker or Windows Defender Application Control may block it outright. There's no workaround without code signing.
The GUI binary is a windowed app with no console — if you dismiss the SmartScreen dialog it simply won't start and won't print an error. Scoop avoids all of this. You can also build from source.
💡 Want the in-app Update button to work? Put the
.exein a folder you can write to without admin rights — e.g. create%LOCALAPPDATA%\Programs\screpdb\and drop it there. screpdb can only replace its own binary when its folder is user-writable, soC:\Program Files\(needs admin) won't self-update. Otherwise the app just shows the download link instead.
The Scoop manifest lives at bucket/screpdb.json and is bumped automatically on each release.
Linux — one-line installer, or Homebrew
Install with one command (downloads the right binary, verifies it against the release's signed SHA256SUMS, drops it on your PATH):
curl -fsSL https://raw.githubusercontent.com/marianogappa/screpdb/main/install.sh | shThen run screpdb. To upgrade, re-run the same command (or use the in-app Update button).
🔍 Don't pipe scripts you haven't read.
install.shis deliberately short and dependency-free so you can audit it in under a minute — it only downloads the binary for your OS/arch, checks it against the release's signedSHA256SUMS, and copies it to~/.local/bin. To read it first, then run your local copy:curl -fsSL https://raw.githubusercontent.com/marianogappa/screpdb/main/install.sh -o screpdb-install.sh less screpdb-install.sh # audit it sh screpdb-install.sh
Prefer Homebrew / Linuxbrew?
brew install marianogappa/screpdb/screpdb # upgrade later: brew upgrade screpdbOr download the binary for your architecture from the Releases page, make it executable, and move it onto your PATH — put it in a writable folder (not a Homebrew prefix) so the in-app Update button works:
chmod +x screpdb-linux-amd64 # or screpdb-linux-arm64
mkdir -p ~/.local/bin && mv screpdb-linux-amd64 ~/.local/bin/screpdb~/.local/bin is the one-line installer's default — any writable folder on your PATH works. Binaries fetched via curl/brew carry no quarantine flag, so they just run.
💡 screpdb self-updates only when its folder is user-writable and not owned by a package manager. A binary you run from
~/Downloadsor a Homebrew prefix won't auto-update — the app falls back to showing the download command instead.
macOS — Homebrew, or one-line installer
Install with Homebrew:
brew install marianogappa/screpdb/screpdb # upgrade later: brew upgrade screpdbOr the one-line installer (verifies it against the release's signed SHA256SUMS, installs to ~/.local/bin):
curl -fsSL https://raw.githubusercontent.com/marianogappa/screpdb/main/install.sh | shWary of piping to sh? It's the same install.sh shown in the Linux section above — read it first, then run your local copy.
Then run screpdb. No Gatekeeper "unidentified developer" block with either method — brew and curl don't attach the quarantine attribute that triggers it, so the binary just runs (no notarization needed).
Prefer a direct download? (this one does hit Gatekeeper)
Download the binary for your architecture from the Releases page, then clear the quarantine flag and move it onto your PATH — a writable folder (not a Homebrew prefix) so the in-app Update button works:
chmod +x screpdb-darwin-arm64 # or screpdb-darwin-amd64
xattr -d com.apple.quarantine screpdb-darwin-arm64 # clear the browser-download quarantine
mkdir -p ~/.local/bin && mv screpdb-darwin-arm64 ~/.local/bin/screpdb(Or right-click the binary → Open to approve it once.) ~/.local/bin matches the one-line installer's default.
💡 screpdb self-updates only when its folder is user-writable and not owned by a package manager. A binary you run straight from
~/Downloadsor a Homebrew prefix won't auto-update — the app falls back to showing the download command instead.
You'll need Go 1.25.2 or later. Use make build (not a bare go build) so the embedded dashboard UI assets are rebuilt first:
git clone https://github.com/marianogappa/screpdb.git
cd screpdb
make build1. Remove the binary
| Installed with | Command |
|---|---|
| Scoop (Windows) | scoop uninstall screpdb |
| Homebrew (macOS/Linux) | brew uninstall screpdb |
| One-line installer / manual | Delete the binary you placed (e.g. ~/.local/bin/screpdb) |
2. Delete the data folder (optional — skip this to keep your data for a future reinstall).
# Windows (PowerShell)
Remove-Item -Recurse -Force "$env:LOCALAPPDATA\screpdb"
# macOS
rm -rf "$HOME/Library/Application Support/screpdb"
# Linux
rm -rf "${XDG_CONFIG_HOME:-$HOME/.config}/screpdb"CLI ingestion, MCP server, and full OpenAPI — click to expand
- CLI for ingestion onto SQLite database. No need to use UI: just ingest and query the database.
./screpdb ingest
- `-i, --input-dir`: Input directory containing replay files (default: system replay directory)
- `-s, --sqlite-path`: SQLite database file path (default: screp.db)
- `-n, --stop-after-n-reps`: Stop after processing N replay files (0 = no limit)
- `-d, --up-to-yyyy-mm-dd`: Only process files up to this date (YYYY-MM-DD format)
- `-m, --up-to-n-months`: Only process files from the last N months (0 = no limit)
- `--store-right-clicks`: Store `Right Click` commands (disabled by default to reduce command-table volume)
- `--clean`: Drop all non-dashboard tables before ingesting to start over (useful for migrations)- MCP server: point an MCP client (Claude Desktop, Claude Code, Cursor, …) at the replay database and ask questions in natural language about any game, player, matchup, build order, or event. The client's model turns your question into read-only SQL over the ingested data. The server exposes tools to run queries (
query_database), inspect the schema (get_database_schema), read StarCraft domain knowledge (get_starcraft_knowledge), and discover players and derived events (list_top_players,list_event_types).
./screpdb mcp
# Specify custom database file
./screpdb mcp -s /path/to/custom.db- Server / API:
./screpdb dashboard(also the default when run with no subcommand) starts the HTTP server and opens the dashboard UI. All UI functionality is exposed as a JSON API — OpenAPI schema available. Run it headless as an API-only server (no UI, no browser) with--headless:
./screpdb dashboard --headless -p 8000 -s /path/to/custom.db
# then: curl http://localhost:8000/api/healthEvery golden value — unit stats, build times, expert timings, detection thresholds — is generated from source and test-backed (see SPECIFICATION.md)
screpdb makes a lot of derived claims: "this is a 9 Pool", "your Spawning Pool was 6s late", "a Zealot takes 25.2s". Skeptical? Audit them.
SPECIFICATION.md documents every golden value the app relies on — unit names, build times, expert timings, costs, tech-tree rules, detection thresholds, and more. It's:
- Generated from the Go source of truth (
go generate ./...), so it can't drift from the code. - Test-backed — CI fails if any value is wrong or the file is stale.
In short: not aspirational docs that rot, but a provably-accurate description of what the app actually does.
Verifying downloads — checksums + minisign signature
Each release publishes a SHA256SUMS file and a SHA256SUMS.minisig minisign signature alongside the binaries.
Verify the checksum (Linux/macOS):
sha256sum -c SHA256SUMS --ignore-missingVerify the checksum (Windows PowerShell):
Get-FileHash screpdb-windows-amd64.exe -Algorithm SHA256
# Compare the printed hash against the line in SHA256SUMSVerify the signature (requires minisign):
minisign -Vm SHA256SUMS -P 'RWS9gPPOydPD/tR8JBOelXKhif526NoAKY18dau7QHR4dqg84QMhJ5L/'screpdb minimizes its attack surface by routing all I/O through facades and keeping dependencies small (see #135). On macOS and Linux this is a best-effort, in-process guard; on Windows a Low-integrity worker adds a real OS write boundary.
How the I/O model works — filesystem, Windows sandbox, network, self-update, enforcement
- Filesystem — all disk access goes through
internal/iofacade, which permits reads/writes only within: a single per-OS app-data directory (%LOCALAPPDATA%\screpdbon Windows,~/Library/Application Support/screpdbon macOS,$XDG_CONFIG_HOME/screpdbon Linux) that holds the SQLite database, game-asset cache, logs, crash reports, and extracted sample replays; and the configured replays folder (read replays, write "watch me" replays). A narrow, read-only exception walks up from the replays folder to find StarCraft'sCSettings.json. - Windows OS sandbox — on Windows the app splits into a Medium-integrity launcher and a Low-integrity worker (#237). The launcher marks the app-data directory Low-writable and relaunches the real worker at Low integrity; the worker keeps read-down access to replays anywhere but can only write into that one Low-labeled folder — every other write is refused by the OS, even from a compromised
screp/scmapanalyzerparser. The launcher retains self-update (it must overwrite the install.exe) and brokers the single "watch me" write into the read-only replays folder on the worker's behalf. This does not stop a compromised parser from reading private files (Low integrity can read up-level); blocking reads needs AppContainer + a broker process, a deferred "Tier 2" follow-up. - Network — the dashboard server binds to
localhostonly. The binary's outbound calls are confined to three sanctioned packages:internal/selfupdate(#212) queries GitHub Releases for self-update, verifying every byte against a minisign-signedSHA256SUMS(embedded public key) before any swap;internal/bnetfacade(#317) talks to SC:R's local web-api bridge (loopback only, path-prefixed to/web-api/) and downloads replays fromstorage.googleapis.com(allowlisted to the single path prefix/starcraft-user-uploads-prod/S1-replays/, with length +seRSmagic-byte validation on every download);internal/netfacadehouses localhost readiness probes. - Self-update — updates are always user-initiated, never automatic. Package-manager installs (Scoop on Windows, Homebrew/Linuxbrew on macOS/Linux) and non-writable install directories are detected and excluded so the updater never fights
scoop update/brew upgradeor needs elevation; those installs are pointed back at their package manager. Thecurl | shinstaller drops into a writable dir (~/.local/bin), so in-app self-update keeps working there. Self-written binaries carry no macOS quarantine xattr / Windows Mark-of-the-Web, so Gatekeeper/SmartScreen don't re-prompt after an update. - Enforcement —
TestNoDirectIOOutsideFacades(ininternal/iofacade) parses the whole module on everygo testrun and fails the build if any package reaches the filesystem or network directly instead of through the facades.internal/selfupdate,internal/bnetfacade, andinternal/winsandbox(the Windows process-spawn / integrity-labeling / broker surface) are the documented exceptions.
On macOS and Linux this is a best-effort, in-process guard, not an OS sandbox: paths handed to trusted dependencies (the SQLite driver, the screp parser, scmapanalyzer) are opened inside those libraries, and the facade only constrains screpdb's own code. On Windows the Low-integrity worker adds a real OS write boundary on top of the same facades.
The LLM that authors each change records a dated, one-line verdict on whether it could weaken the I/O rules above (see AGENTS.md); TestIOSafetyAuditPresent fails CI if the log is empty, and the enforcement test above stays the authoritative guard.
2026-09-03 OK. Dead code removal ahead of the in-memory replay library (step 1 of removing SQLite from the dashboard). Deletes the unused executeQuery/withFilteredConnection local-filter path and its raw *sql.DB query/scan Store helpers, ten unreferenced Store query methods with their sqlc queries, the orphaned sqlcgen player_summary file, test-only player-insight helpers, and the empty /api/custom/global-replay-filter/options stub (spec removed, code regenerated). Tests now discover replay ids and player keys through the public API instead of raw SQL. Pure removal: no new I/O, endpoints, queries, os/net calls, facade changes, or AlgorithmVersion bump.
Older I/O safety audit entries (click to expand)
2026-09-02 OK. Players list omnibar + player summary list tweaks. FilterOmnibar is parameterized (axes, state labels, noun, and an optional live text filter) and the players list now uses it in place of its ad-hoc filter row, wired to the existing name/onlyFivePlus/lastPlayed filter state and /api/players parameters; no query or endpoint changes. The players table adopts the games list header and hairline styles. The player summary last-games rows gain a Time column and show the player-oriented matchup instead of 1v1; the ladder stat guards against missing wins/losses (kospetrov rendered "undefined-undefined"). Go changes are removals only: the [country-flags] debug log lines in internal/dashboard/bnet_profiles.go are deleted (no behavior change) and a counter loop is modernized to range-over-int. No new I/O, endpoints, queries, os/net calls, facade changes, or AlgorithmVersion bump.
2026-09-02 OK. Player Summary tab visual polish, frontend-only. The Battle.net and Local games sections gain emoji-titled headers and labeled stat tiles (battle tag and country rows dropped; EAPM folded into the APM tile), "Also plays as" aliases render as pills and link to the local player page when one exists (checked client-side via the existing GET /api/players/{key} endpoint, no new backend code), the last-games list becomes a single table with a header row, hairline row separators and a wider composition column with early/mid/late labels, the player title spacing is normalized, and chat summary messages render full-width. No Go changes at all: no new I/O, endpoints, queries, os/net calls, facade changes, or AlgorithmVersion bump.
2026-09-02 OK. Players page rework. Removed: the player Summary tab's per-matchup cards, special pills and distinctive-outlier machinery, the Recent games tab, and the player Back button, ripping out their endpoints (playerSummaryPerMatchup/Special/Outliers, playerOutliers, playerRecentGames), builders, payload types, eleven now-orphaned sqlc/store queries (player_summary.sql entirely, the outlier and population queries in player_insight_static.sql) and their tests. Added: playerLastGames (GET /api/players/{playerKey}/last-games), which reuses the recent-games query (now also selecting team_format, LIMIT 10) and per-item populates the requesting player's result, APM/EAPM (two columns added to an existing SELECT), a disconnect flag (one new read-only query over replay_events player_dropped rows), featuring patterns and the request-time unit composition already computed for the per-game page; CountPlayerBnetGames (one COUNT over replays+players) gating the new Battle.net summary section; and lifetime games/wins/losses/disconnects plus a derived average APM decoded from the already-cached bnet profile payload (no new fetches, cache-only read path unchanged). Frontend: players list pages like the games list, tab order Summary/Hotkeys/Skill proxies/Chat summary, and the new Summary tab renders Battle.net (hidden without bnet-sourced replays) and Local games with the last ten games as rows. All reads go through existing sqlc/store paths; no os/net calls, no facade changes, no enforcement-test change, no AlgorithmVersion bump (nothing persisted changed).
2026-09-02 OK. Hotkey "Add" semantics fix, detection-layer only. Investigating a player (Howard_The_Duck in the BGH fixture) who "used groups without ever assigning" showed shift+number ("Add", hotkey type 0x02) adds the current selection TO the group, creating it when empty; every selection replay in the codebase (internal/unittags Analyze/Coordinates/MorphSelectionSizes/DetectMutaHarass and the hotkeystream extractor) had it backwards, pulling the group into the selection. All five now mutate the group and leave the selection intact, and the extractor folds Add into an assign of the group's new contents so shift+number-only players get building labels and counts. Marker and parser goldens regenerated (their diffs are exactly the retired used_hotkey_groups rows plus the corrected Add attribution); the storage fixture count drops by the retired marker's rows. Pure in-memory analysis changes: no new I/O, endpoints, queries, os/net calls, or facade changes; AlgorithmVersion stays 68 (unreleased, its re-ingest carries this too).
2026-09-02 OK. Hotkey screens polish, purely presentational. Map composite: building sprites now keep their true aspect ratio (width fit to the footprint, standing up to twice its height, anchored bottom center) instead of being stretched into the footprint box; footprint outlines use one accent colour; the circled comma-joined group badge becomes a row of small keycap-style chips drawn above the sprite so it is never covered. Frontend: short games stretch the timeline to the full lane width (ResizeObserver-driven px-per-minute floor), map crops cap at 75vh with object-fit contain, and the tab note reuses the standard workflow-section-warning styling. All drawing stays in the existing in-memory composite; no new I/O, endpoints, queries, os/net calls, or facade changes; no AlgorithmVersion bump (stored data unchanged).
2026-09-02 OK. Hotkey screens iteration, all presentation and derived-data changes. Backend: hotkey_stream unit counts now cap at Brood War's 12-unit selection limit (extract.go); the used_hotkey_groups marker, its evaluator and its summary pill are retired (never_used_hotkeys stays), with ListHotkeyGamesRateByPlayer re-pointed from that marker to players.hotkey_stream IS NOT NULL, so the never-hotkeys player pill keeps working on the same read path; the hotkey map endpoint accepts an exact-second cutoff (same validation shape as minute) so the frontend can notch its slider at building-assign moments; ComputeSignature additionally aggregates per-key assign-minute spans. Frontend: timeline drops the legend, race chips, per-row summaries and hotkey-less players, repeats the minute axis per player, and states that unit types are examples; map crops render two per row with a notched slider that picks the richest moment per minute; signature cards drop title, prose, scores and labels, and render one amber assign bar and one blue use bar per key with instant tooltips. No new I/O: same tables, same endpoints (one new optional query param), no os/net calls, no iofacade/netfacade change, no enforcement-test change; AlgorithmVersion stays 68 (unreleased on this branch, its re-ingest carries the cap + marker retirement).
2026-09-02 OK. Test coverage lift to 82% (hand-written code, scripts/coverage.sh) plus one correctness fix the new tests caught: screp names the comsat Build unit "ComSat" and the queens nest "Queens Nest", so extract.go now canonicalizes screp build names before evidence/placement matching (without it, comsat add-on evidence and comsat/queens-nest tile locations silently never matched). All other changes are tests: a synthetic-replay harness for hotkeystream.Extract, unit tests for the hotkey map composite and its handler against the ingested test corpus, direct endpoint tests for gameHotkeys/playerHotkeySignature (including a temporarily renamed-and-restored row set in the shared in-memory test DB), feature-flag and gaming-session route tests, a read-only endpoint sweep over every corpus game and player, and seeded dashboard/db Store tests for the hotkey, gaming-session, workflow filter-count and bnet-profile queries. Tests write only to per-test temp DBs and the existing test fixtures; no new I/O paths, no os/net calls, no iofacade/netfacade change, no enforcement-test change, no AlgorithmVersion bump (68 already forces the re-ingest that carries the comsat fix).
2026-09-02 OK. Hotkey intel: timeline + map on the game report, signature on the player page. players.hotkey_stream moves to wire format v2 (internal/hotkeystream: second deltas, assigns annotated with unit counts or the proven building type + build tile), extracted in the parser from the raw screp stream (hotkeystream.Extract mirrors unittags' selection replay) and stored through the existing insertPlayersBatchTx path; AlgorithmVersion 67->68. Reads added: three sqlc queries over players/replays (per-replay streams, per-player streams by key, one player+replay stream), two strict-server JSON endpoints (gameHotkeys, playerHotkeySignature) that only decode stored blobs, and one custom PNG endpoint (/api/custom/hotkeys/map) that re-reads the already-ingested replay file for terrain via the same scmapanalyzer render, disk cache path, and singleflight key as the existing /api/custom/game-assets/map, then composites sprites in memory. All file access goes through iofacade/appdata paths that were already permitted; no new roots, no writes outside the existing game-assets cache, no os/net calls, no netfacade change, no enforcement-test change.
2026-09-01 OK. Filter omnibar for the games list (issue #359). Two read-only aggregate queries complete the option counts (Store.CountWorkflowMatchupGames and CountWorkflowMapKindGames, one GROUP BY over replays each, mirroring buildMatchupClause and buildMapKindClause so a count matches its filter), which makes a zero mean "none in your corpus" instead of "unknown". The rest is frontend: a new components/FilterOmnibar.jsx replaces two <select> dropdowns and forty flat filter pills with one typed input plus a browse row of seven axis blocks, reading the same /api/games filter_options payload and writing the same mainGamesFilters state, so no new endpoint, request or query parameter is involved. Chips are grouped by filter state key rather than display axis so the rendered OR/AND grammar matches BuildWorkflowGamesListWhere. Also aligns the games-list team blocks into one CSS grid per Players cell. No writes, no detection change, no AlgorithmVersion bump, no os/net calls, no iofacade/netfacade allowlist widening, no enforcement-test change.
2026-09-01 OK. Per-option game counts for the games-list featuring filters (issue #359, omnibar groundwork). New Store.CountWorkflowFeaturingGames resolves each UI filter key to a count via workflowFeaturingCountShapeFor, which mirrors the existing workflowFeaturingExistsSQL switch so a count and its filter always agree (two new guard tests assert the two switches stay in step). Cost is a fixed three read-only aggregate queries (GROUP BY over replay_events, a payload-label breakdown, one replays COUNT) plus one COUNT(DISTINCT) per composite key, all through the existing replay-scoped store helpers; counts are corpus-wide and take no user input, so no new query parameters are interpolated. Read-only: no writes, no os/net calls, no iofacade/netfacade allowlist widening, no enforcement-test change, no AlgorithmVersion bump.
2026-09-01 OK. Replaced the two literal NUL (0x00) bytes in internal/dashboard/frontend/src/lib/compositionPill.jsx with the \x00 escape. The raw bytes made git classify the whole 251-line file as binary, so every change to it rendered as "Bin <n> -> <m> bytes" and could not be reviewed in a diff or a PR. The escape produces the identical character at runtime, so the composite (unit, spell) map keys stay byte-identical (verified: the built JS bundle hash is unchanged). Source-encoding only: no behaviour change, no os/net calls, no iofacade/netfacade allowlist widening, no enforcement-test change, no AlgorithmVersion bump.
2026-09-01 OK (net reduction: one fetch and one endpoint consumer removed). Dashboard visual-hierarchy rebalance, first tranche (issue #359). Presentation-only: an emphasis-ladder token block in styles.css, games-list column widths (Featuring no longer crops behind a "..." toggle), player-name pill fills/flags/crowns dropped in the list with the winning side rendered as weight, a single Players cell component for every player count (teams chunked into balanced rows from the already-loaded players array, replacing the separate 1v1/team/no-team-info branches), unit-composition segment fills changed to a greyscale ramp, and the replay player colour moved from the name's text colour to an adjacent swatch (which retires legendTextStyle and its black|navy|darkblue text-shadow allowlist, fixing 4 of 8 names below WCAG AA). The /api/player-colors rank palette is no longer consumed: the api.js client method and its loader are deleted, so the frontend makes one fewer request at boot; the Go endpoint is untouched and now unused. No Go changes at all, no detection change and no AlgorithmVersion bump, no new os/net calls, no iofacade/netfacade allowlist widening, no enforcement-test change.
2026-09-01 OK. Hotkey commands now stored as an encoded blob column on players (issue #357, AlgorithmVersion 67). Each player's Hotkey commands are delta-varint-encoded in memory (new internal/hotkeystream package) and written into the new players.hotkey_stream BLOB by the existing player insert; Hotkey rows are no longer written to commands_low_value, and the --skip-hotkeys CLI flag and skip_hotkeys ingest API field are removed, which narrows the API surface. All inputs are already-parsed in-memory commands; the schema change is one additive migration. No new os/net calls, endpoints, facade exemptions, allowlist widening, hosts or paths.
2026-09-01 OK. Mass-disconnect games no longer credit the replay saver a phantom win (issue #358, AlgorithmVersion 66). All changes consume data already parsed in memory: the detector reads the in-memory players/commands slices, winner clearing touches the same structs, and the two new replay_events types (player_dropped, mass_disconnect) flow through the existing worldstate emit + storage insert path with the type allowlist widened accordingly. No new os/net calls, endpoints, facade exemptions, hosts or paths.
2026-09-01 OK. Country flags fixed on Windows + country-name tooltips (issue #361). Windows ships no flag glyphs, so the dashboard now vendors the Twemoji Country Flags woff2 (committed at frontend/src/assets, sha256-pinned) into the frontend build, where the existing go:embed of frontend/build serves it from the app's own origin — no page fetches a remote asset. A canvas feature-probe activates the font only where the platform lacks native flags; tooltips resolve names via the browser's built-in Intl.DisplayNames. Presentation only: no new endpoints, os/net calls, facade exemptions, iofacade allowlist widening, hosts or paths; no AlgorithmVersion bump (no detection change).
2026-09-01 OK. Expert golden-line timings re-derived from the aurora-ID-labelled progamer corpus (issue #362). All shipped changes are data-only: Marker.Expert targets/tolerances in definitions.go, a render-time fuzzyZergExpertByLabel table, and the Build Orders tab resolving fuzzy-opener bands from the already-persisted payload label over the existing ListEarlyZergMorphsForBOTimings query. The mining pipeline itself lives in scripts/expert-mine — dev-only, never shipped, exempt from the facade guard like the rest of scripts/ — and reads the local screpharvest/scfingerprint corpus paths passed by flag, writing only under its -workdir. No new shipped os/net calls, endpoints, facade exemptions, allowlist changes, hosts or paths; no AlgorithmVersion bump (targets are presentation, not detection).
2026-08-30 OK. Battle.net profile details surfaced on the player page and the Gaming Session players tab. All of it parses payloads already cached in the bnet_profiles table by the existing fetch path: one new read-only query (ListBnetProfilePayloadsByPlayerKeys, bounded by an IN clause over the keys on screen) and a decoder for the identity, alternate-toon and matchmaking fields. Nothing here fetches, so no bridge request is issued and no rate-limit budget is spent; the avatar URLs in the payload are decoded away rather than rendered, so no page pulls a remote asset. No new endpoints, os/net calls, facade exemptions, allowlist changes, hosts or paths; no AlgorithmVersion bump (no detection change).
2026-09-03 OK. Built-in progamer profiles (internal/propack). The binary embeds a precomputed snapshot (pros.json + Liquipedia portraits) generated offline by scripts/pro-pack from the aurora-ID-labelled ladder corpus; the app reads it with embed only and makes no request to cwal.gg or Liquipedia (the script does, from a developer machine, outside the Go module's enforcement scope in scripts/). Runtime additions: featured rows on the players list, pro player keys (pro:<id>) on the player/insight/hotkey endpoints served from the embedded data, one new hand-written endpoint (GET /api/custom/pros/{proID}/photo) serving embedded bytes, and a cache-only read of bnet_profiles (aurora IDs of the user's own toons) to hide a pro's built-in profile when the user is that pro. Battle.net profile fetches for pros reuse getOrFetchBnetProfile with known (toon, gateway) pairs, so no new hosts, paths or facade exemptions; the Battle.net section now also decodes game_results from the already-cached payload (no extra requests) and appends them to a new bnet_game_results table (dashboard migration 000003) on each fetch the app already makes, so play habits (weekday/weekend, time of day in the account country's zone via the embedded time/tzdata) accumulate at zero extra bridge cost. No iofacade allowlist widening, no AlgorithmVersion bump (no detection change).
2026-08-30 OK. Money-map classification fixed and the Gaming Session view wired up. The map rule now takes the median mineral field rather than whichever field the map file stored first, and compares against a standard 1500 patch instead of a strict > 10000, so "Big Game Hunters - Remastered" stops reading as a Regular map; this reads already-parsed map data and adds no I/O. core.AlgorithmVersion 63 to 64 so map_kind is recomputed on re-ingest. New env var SCREPDB_SESSION_RECENCY widens the session window for development; it is read with os.Getenv, touches no filesystem, and cannot name a path. No new os/net calls, no facade exemptions, no allowlist widening, no new hosts or paths.
2026-08-30 OK. Manual player aliasing removed; Feature Flags and a flag-gated Gaming Session added. The player_aliases table is dropped by a new settings migration (000002) and its queries, endpoints, OpenAPI operations and UI are gone, which narrows the API surface rather than widening it. The one piece kept, the "you" players, still reads CSettings.json through the same iofacade.FindAndReadAncestorFile call as before, at the same path, read-only; the result is now held in memory instead of persisted, so this removes a DB write and adds no I/O. Feature flags live in a new settings column. Two new read-only endpoints (GET /api/custom/feature-flags, GET /api/custom/gaming-session) and one write (PUT /api/custom/feature-flags) that accepts only allowlisted flag keys; all read the local DB only. The oapi-codegen generator is now pinned instead of @latest, which stops a floating generator emitting code against the pinned runtime. No new os/net calls, no facade exemptions, no allowlist widening, no new hosts or paths.
2026-08-30 OK. Bridge connection state now watches the port instead of polling HTTP. New bnetfacade.BridgePortOpen opens and immediately closes a loopback TCP connection to an address the same isLocalAddr guard already vets; it sends zero bytes, so SC:R has nothing to forward upstream and no rate-limit budget is spent, and it is unmetered for the same reason ProbeBridge is. The monitor spends a real HTTP probe only when the port's presence flips or on a 60s floor, which is strictly fewer bridge requests than the previous unconditional 30s poll. Profile backfill is now capped at 20 players per page view (previously unbounded, and a players page could have queued 25 players x 5 gateways). One new read-only endpoint, GET /api/custom/bnet/country-codes, serves the existing profile cache and never fetches, so the UI can poll it for free; its IN clause is bounded at 200 keys. No new hosts, paths, facade exemptions or iofacade allowlist changes; no AlgorithmVersion bump (no detection change).
2026-08-30 OK. Player-pill glyph sizing, games-list Featuring column, fingerprint domain gate, and scmapanalyzer bump. All presentation and query-shape changes: CSS only for the pill adornments; one SQL predicate added to ListPlayerFingerprintVectors restricting fingerprint input to 2-human non-money games (narrows what is read, never widens it); build-order markers dropped from the games-list Featuring column. scmapanalyzer bumped to 2026-08-16 for base recognition on newer maps — an existing dependency at an existing call site, no new I/O capability — with core.AlgorithmVersion 62 to 63 so bases re-resolve on re-ingest. No new os/net calls, no facade exemptions, no iofacade allowlist widening, no new hosts or paths.
2026-08-30 OK. Fetch and cache SC:R aurora profiles (issue #329). New bnetfacade.FetchAuroraProfile goes through the already-metered BridgeGet (loopback-only, /web-api/ prefix, #319 budgets apply — no new hosts, paths, or facade exemptions) and normalizes the payload to UTF-8 before parsing. Responses are cached in a new bnet_profiles table (dashboard migration 000002) keyed on (toon, gateway) per #344, 24h TTL per Blizzard's Cache-Control max-age=86400, with the unknown-toon 200/aurora_id-0 response negative-cached so misses don't re-spend budget; failed refetches serve the stale row. One new hand-written endpoint (GET /api/custom/bnet/profile). No iofacade allowlist widening, no AlgorithmVersion bump (no detection change).
2026-08-30 OK. Two-budget rate limiter at the bnetfacade boundary (issue #319). BridgeGet and DownloadReplay now spend separate in-package budgets (token buckets with priority queues, persisted daily caps, exponential cooldown on the bridge's explicit rate-limit signal), so no caller can bypass them; ProbeBridge stays unmetered (local liveness only). New file write: bnet_budget.json (daily counters + cooldown) inside the app-data root, read/written strictly through iofacade — no allowlist widening, no new hosts or paths, no AlgorithmVersion bump (no detection change). Dashboard surfaces a requests-today meter via the existing /api/custom/bnet/status endpoint.
2026-08-29 OK. SC:R local web-api detection and connection-state UI (issue #318). Adds platform-specific port discovery (macOS: lsof, Linux: /proc/net/tcp, Windows: GetExtendedTcpTable via iphlpapi.dll) and a ProbeBridge function to bnetfacade — all loopback-only, one GET to /web-api/v1/gateway every 10s. Dashboard gains a background monitor goroutine (atomic-cached state), two new hand-written endpoints (GET/POST /api/custom/bnet/{status,toggle}), and a nav pill. The global off-switch disables all probing. No new outbound hosts (all 127.0.0.1), no iofacade allowlist widening, no AlgorithmVersion bump. Platform-specific files use os/exec (macOS lsof) and os.ReadFile (Linux /proc/net/tcp) inside the already-exempt bnetfacade package.
2026-08-29 REVIEW (deliberate, documented widening). New internal/bnetfacade package (issue #317): third sanctioned network surface added to the enforcement-test skip list. Loopback client restricted to 127.0.0.1 + /web-api/ prefix for SC:R's local bridge. Outbound client allowlisted to exactly storage.googleapis.com + /starcraft-user-uploads-prod/S1-replays/ prefix for GCS replay downloads. Downloaded replays validated (length ≥ 16 bytes + seRS magic at offset 12). Bridge payloads decoded leniently (cp949/ISO-8859-1 fallback) for Korean map titles. Host, path, and loopback restrictions all covered by tests. This widens the attack surface: the binary now makes one genuine outbound call to a third-party host, guarded by host+path allowlist at the facade boundary. golang.org/x/text promoted from indirect to direct (cp949/charmap decoders). No iofacade allowlist widening; no AlgorithmVersion bump (no detection change).
2026-08-29 OK. Derive game_source and lobby_kind from replay content at ingest (issue #346). Reads only in-memory screp replay struct fields (rep.ShieldBattery, rep.RepFormat, rep.Header.Players, rep.Header.Title). Two new replays columns via migration 000004, surfaced on the API/MCP schema. No new os/net calls, no iofacade/netfacade allowlist widening, no enforcement-test change; AlgorithmVersion bumped 61→62 to drive backfill via the existing re-ingest hint.
2026-08-27 OK. Fingerprint feature vectors extracted at ingest and stored in a new player_fingerprint_vectors table. New dependency github.com/marianogappa/scfingerprint is pure computation over the already-parsed replay (embedded model/dataset via go:embed, no filesystem or network I/O in the code paths used — Extract/FeatureVersion/ModelTag). No new os/net calls, no iofacade/netfacade allowlist widening, no enforcement-test change; AlgorithmVersion bumped 60→61 to drive vector backfill via the existing re-ingest hint.
2026-07-04 OK (net reduction in the SQL surface's capability). MCP-server modernization + dashboard headless API mode. MCP: query_database now rejects non-read-only SQL (only SELECT/WITH/EXPLAIN/PRAGMA, single statement, comment-stripped) so an MCP client can no longer mutate the corpus; corrected tool descriptions/annotations, expanded GetDatabaseSchema introspection to replay_events/player_aliases, refreshed the domain-knowledge text, added two read-only discovery tools (list_top_players, list_event_types), and bumped mcp-go v0.41.1→v0.55.1. Dashboard: new `--headless` flag serves the JSON API only (no embedded SPA, no browser-open — one fewer os call in that mode); documented 8 operational endpoints (game-assets, debug map-layout, markers definitions, sample-set load, self-update status/apply) in the OpenAPI spec, excluded from code generation, with the validator middleware deferring method-less spec paths to their hand-written handlers while still returning 405 for genuine wrong-method calls. All DB access stays through the storage/dashboard layer; no new os/net calls, no iofacade/netfacade allowlist widening, no enforcement-test change, no AlgorithmVersion bump (no detection change).
2026-07-04 OK. Zerg opener supply fix: larva morphs cancelled before the player's first Overlord are dropped from the "N Pool"/"N Hatch" count (a cancelled egg that early is provably a Drone, so it refunds a supply) — fixes e.g. a 5 Pool with a cancelled drone reading as 6 Pool. New commands.DropCancelledMorphs runs on the already-filtered stream in the parser; AlgorithmVersion 58→59 (re-ingest), SPECIFICATION.md regenerated. Reads the in-memory command slice only: no os/net calls, no iofacade/netfacade allowlist widening, no enforcement-test change.
2026-07-04 OK. Beta-exempt the catch-all residual buckets (bo_zerg_other / bo_protoss_other / bo_terran_other / opener_unresolved) so the dashboard stops flagging them "beta" — they claim whatever the named openers leave over, so there is no premise to verify. Added the keys to markers.betaExemptFeatureKeys plus a guard test that every exempt key names a live marker. Display-time curation metadata only (beta tag is computed from FeatureKey at the definitions endpoint): no detection/ingest change, no AlgorithmVersion bump, no os/net calls, no iofacade/netfacade allowlist widening, no enforcement-test change.
2026-07-04 OK. "Update available" UX polish: managed/not-writable installs now show a copyable upgrade command with a Copy button and a Changelog link, the loud (major) banner is dismissable like the quiet one, and the not-writable macOS/Linux case surfaces the `curl | sh` install-script re-run. Go change is additive-only — a new runtime.GOOS-derived `OS` field on selfupdate.Status so the frontend can pick a platform-correct command; plus README direct-download placement tips. No new os/net calls, no iofacade/netfacade allowlist widening, no enforcement-test changes; the self-update mechanism (minisign-verified, user-initiated, writable-dir/package-manager detection) is unchanged.
2026-07-03 OK. Removed the Go Report Card README badge (goreportcard.com no longer rendering it) and added a static coverage badge (81%, from scripts/coverage.sh — generated code excluded). Docs-only: no code, os/net calls, or facade allowlist change.
2026-07-03 OK. Fixed the player-outliers endpoint returning 500 (not 404) for an unknown player: GetOutlierPlayerSummary now COALESCEs the NULL-name aggregate to '' (sqlc query + regenerated sqlcgen). Query/codegen only, still through the dashboard store; no os/net calls, no facade allowlist change.
2026-07-03 OK. scmapanalyzer bumped to v0.0.0-20260702193642 (upstream copyright-only change) + Wraith Cloak timing now reports Cloaking Field completion (start+63s, drops if the game ends first), AlgorithmVersion 58. Dependency/detection only: map analysis still runs through the iofacade allowlist, no os/net calls added, no allowlist widening, no TestNoDirectIOOutsideFacades change; 193 no-cache tests across the scmapanalyzer-dependent packages pass.
2026-07-03 OK. Speedlot-timing now reports Leg Enhancement research completion (and drops when the game ends first) + "9 Pool 9 Hatch" relabel; AlgorithmVersion 57, goldens + SPECIFICATION.md regenerated. Pure detection/pattern logic: no os/net calls, no iofacade/netfacade allowlist change, no TestNoDirectIOOutsideFacades change.
2026-07-03 OK. Toolchain bump to Go 1.26.4 (go.mod go directive) plus a repo-wide gofmt pass. Tooling/formatting only: no os/net calls added, no iofacade/netfacade allowlist widening, no change to the TestNoDirectIOOutsideFacades enforcement.
2026-07-02 OK. README presentation pass, no behaviour change: collapsed the per-OS install sections, baked the measured ingestion-throughput figure into the badge, trimmed the Security/I/O-model prose, and reformatted this audit log into a code block. The TestIOSafetyAuditPresent regex was loosened to also match the new plain-date log line. Docs/test-format only: no os/net calls, no iofacade/netfacade allowlist widening, no change to the TestNoDirectIOOutsideFacades enforcement test.
2026-07-02 OK. Follow-up to #248 (PR #255): dashboard-frontend copy only — for package-manager (Homebrew/Scoop) installs the "update available" control shows the exact upgrade command (brew upgrade screpdb / scoop update screpdb) in its label, renders as a non-link (a download page is useless when the fix is a terminal command), and surfaces it via an instant hover/focus tooltip. No Go changes, no os/net calls, no iofacade/netfacade allowlist widening, no enforcement-test changes.
2026-07-02 OK. BO/marker label render fixes (issue #251) + ingestion-speed benchmark tracking (issue #249). #251 is render-only: a new markers.DecodePayloadLabel decoder resolves the persisted {"label":...} value so the games-list Featuring strip and game-detail openers show "3 Hatch Muta"/"~9 Overpool" instead of the placeholder name — reads existing payloads, no new os/net calls. #249 adds a make bench-ingest target, scripts/bench-ingest.sh / scripts/update-readme-bench.sh, and a bench-ingest.yml workflow — all shell/CI tooling that runs outside the screpdb binary (not part of the Go module the enforcement test parses), plus a checksum-dedup guard in the existing storage benchmark test. No iofacade/netfacade allowlist widening, no enforcement-test changes.
2026-07-02 OK. Free first-install UX for macOS/Linux + GUI asset rename (issue #248). New files are the standalone install.sh (curl | sh installer) and scripts/update-homebrew-formula.sh/.github release wiring — these run outside the screpdb binary (they are shell installers/CI, not part of the Go module the enforcement test parses), so they touch no iofacade/netfacade surface. In-binary Go changes are a pure rename: the Windows GUI release asset screpdb-dashboard-windows-amd64.exe → screpdb-gui-windows-amd64.exe and the buildinfo.Variant value dashboard → gui (self-update asset-name string in internal/selfupdate + a one-line dashboard-frontend message), plus the GUI log file screpdb-dashboard.log → screpdb-gui.log. No new os/net calls, no iofacade/netfacade allowlist widening, no enforcement-test changes. Self-update mechanism is unchanged (still minisign-verified, user-initiated); the curl/Homebrew install paths reuse the existing package-manager / writable-dir detection.
2026-07-02 OK (with a deliberate, documented allowlist change + one new sanctioned surface). Windows Low-integrity sandbox (issue #237). Filesystem: writes are consolidated under a single per-OS app-data root via the new internal/appdata package (DB, game-asset cache, logs, crash reports, sample replays) — the iofacade allowlist changes, not widens: the working-directory and OS-user-cache roots are removed and replaced by the one app-data root (the read-only replays root is unchanged). Windows-only: a new internal/winsandbox package performs raw golang.org/x/sys/windows calls (duplicate-token → Low integrity level → CreateProcessAsUser; SetNamedSecurityInfo to Low-label the app-data dir) and a file-drop broker so the Medium launcher performs the one "watch me" write into the read-only replays folder on the Low worker's behalf; it is added to the enforcement-test skip list alongside internal/selfupdate and documented as a chokepoint. golang.org/x/sys is promoted from indirect to direct. Self-update is unchanged in mechanism (still minisign-verified, user-initiated) — on Windows it now runs in the Medium launcher rather than the worker. Net effect is a reduction in attack surface: even a compromised screp/scmapanalyzer parser can no longer write outside the single app-data dir on Windows. Residual risk documented: a compromised Low worker can request one fixed-path (000_screpdb_watch_me/watch_me.rep) brokered write into the replays folder — low impact, no arbitrary paths.
2026-07-02 OK. "N Hatch <tech>" redesign (issue #245): Hydra/Muta/Lurker become composition markers (any N) layered on the supply opener, counted by town-hall builds at the economy→army transition. New internal/unittags.TownHallBuildSeconds reads the already-parsed raw command stream (no new I/O), threaded through the orchestrator into a new worldstate.Engine.TownHallBuildSeconds getter. Pure detection-logic + dashboard-response + testdata changes; no new os/net calls, no iofacade/netfacade allowlist widening, no enforcement-test changes.
2026-07-01 OK. Round-10 follow-up: N Hatch Hydra base count uses a +30s grace at hydra-production start (2jd fix); curate wraiths / muta hit-n-run / 2jd fixtures. Pure detection-logic + testdata changes; no new os/net calls, no iofacade/netfacade allowlist widening, no enforcement-test changes.
2026-07-01 OK. Round-10 curation: beta-exempt deterministic facts (became_, game-phase, viewport, never_); curate 18 BOs/markers (1 Gate no-expa, 7/8 Pool, 3 Starport Valk, Carriers, BCs, Forge Cannon/Forge-Gate-Cannon, 2 Fact Expa Mech, Nukes, Sair/Speedlot, 1 Fact Expa Tankless Mech, Wraith Cloak, 1-Base Mech) with watched fixtures; rename "Mech (no expa)" family → "1-Base"; fix manner_pylon firing vs Zerg opponents. Pure detection-logic + dashboard-response + testdata changes (marker curation registry, worldstate manner-pylon race gate, golden fixtures); no new os/net calls, no iofacade/netfacade allowlist widening, no enforcement-test changes.
2026-06-30 OK. Terran mech taxonomy reformulated (issues #226/#227): mech named by Factories before the first expansion ("N Fact Expa Mech" + Tankless/plain/no-expa variants), a Goliath composition flavor ("Goliath" / "N Fact Expa Goliath", folding the standalone Goliath opener), "2/3 Starport Wraith/Valkyrie" cluster openers, Bunker Rush loosened to 2+ forward bunkers, retired "Factory Expand"/"2 Fact before Expa". New marker-DSL predicates (BuildCountBeforeFirstBuildOf, BuildCountAtLeastBeforeFirstBuildOf, NthBuildWithinGapOfFirst) + a builddedup Tier-A fix for Terran rapid re-placements; definition/allowlist/curation edits, AlgorithmVersion 50→51. Pure detection-logic + dashboard-response changes; no new os/net calls, no iofacade/netfacade allowlist widening, no enforcement-test changes.
2026-06-29 OK. Fuzzy Zerg opener: when a multi-larva Drone morph makes the supply rung indeterminate, emit a "~N Pool/Hatch" label instead of an exact rung (new Custom evaluator + 13 Hatch rung + 3 Hatch Muta → marker). Pure detection-logic + dashboard-response changes; no new os/net calls, no iofacade/netfacade allowlist widening, no enforcement-test changes.
2026-06-29 OK. Zerg pool/hatch supply-count fix: ProduceCountBeforeBuild now counts produces by game-second relative to the building rather than observation order, correcting a dedup-tail miscount (9 Overpool read as 10 Pool). Pure detection-logic change in the marker DSL; no new os/net calls, no iofacade/netfacade allowlist widening, no enforcement-test changes.
2026-06-27 OK. New markers (Maelstrom, Crazy Zerg, Guardians) + timing pills (First Observer, First Mine), proxy-building map overlays, and a "beta" tag on uncurated markers/BOs. Pure detection + dashboard-response changes: marker definitions/evaluators, a new cmdenrich.KindLayMine fact for the PlaceMine / VultureMine orders, a subjectsOfInterest addition, a curated-feature-key registry surfaced via the markers-definitions endpoint, and frontend rendering. No new os/net calls, no iofacade/netfacade allowlist widening, no enforcement-test changes.
2026-06-27 OK. Terran air/specialist openers (issue #228): redefined/renamed build-order markers, a new Wraith Cloak timing pill, a new proxy_starport game-event, and a player-aware proxy spatial gate. Pure detection + dashboard-response changes (marker/worldstate logic, event-type allowlists, frontend rendering); no new os/net calls, no iofacade/netfacade allowlist widening, no enforcement-test changes.
2026-06-25 OK (with a deliberate, documented widening). In-binary self-update (issue #212) introduces the binary's first sanctioned outbound network calls (GitHub Releases API + asset download) and its first writes outside the iofacade roots (atomically swapping the running binary in its own install dir). Both are confined to the new internal/selfupdate package, which is added to the enforcement test's skip list alongside iofacade/netfacade and documented as a chokepoint. Integrity is guaranteed by verifying a minisign signature (embedded public key) over SHA256SUMS and the asset's SHA-256 before any swap; updates are user-initiated only, and package-manager/non-writable installs are excluded. No other package gained os/net access; the rest of the binary stays behind the facades.
2026-06-09 OK. Ingestion crash resilience (issue #165): added a per-replay panic guard in internal/ingest (recover → per-file error) and a guarded type assertion in the parser. Pure control-flow/error-handling change; no new os/net calls, no allowlist or enforcement-test changes. Audited the concurrent parse/detect path and its screp/scmapanalyzer deps for shared mutable state (none found unguarded).
2026-06-09 OK. Debugging/crash-reporting improvements (issue #165): new internal/crashreport writes a crash log via iofacade.WriteFile, and the Windows GUI binary opens a screpdb-dashboard.log via iofacade.Create and registers cwd with iofacade.AllowDir (already an allowed root). No new direct os/net calls, no allowlist widening, no enforcement-test changes; the crash handler's browser-open uses pkg/browser (process exec, not a net/fs primitive).
2026-08-29 OK. Fingerprint-based player identification on the player detail page (issue #322). New read-only SQL query (ListPlayerFingerprintVectors) against the existing player_fingerprint_vectors table; scfingerprint.MatchMany runs in-process on already-stored vectors. No new os/net calls, no allowlist changes.
2026-06-07 OK. Early-game event overlay rework (issue #159): consolidated BO timeline events + map overlays. Pure presentation/dashboard-response changes (Go struct field, frontend rendering); no new os/net calls, no allowlist or enforcement-test changes.
2026-05-31 OK. Introduced the iofacade/netfacade chokepoints, the enforcement test, and removed the AI + fswatch surfaces; this change establishes the I/O rules rather than weakening them.
- This project is licensed under the MIT License. See the LICENSE file for details.
- Due to security safeguards I can no longer accept PRs or other code contributions, but please feel free to file an Issue, and you're more than welcome to contribute non-code improvements.
- Built using the github.com/icza/screp library for StarCraft replay parsing. This project would have been impossible without András Belicza's work.
- Country flags on platforms that lack them (Windows) are drawn with the Twemoji Country Flags font, whose artwork comes from Twemoji and is used under CC BY 4.0.