--- name: "News: Translate Executive Briefs" description: "Translates `analysis/daily/**/executive-brief.md` (English source produced by per-type news workflows) into the 13 non-English target languages as sibling `executive-brief_.md` files. Runs three times daily and picks up at most `max_briefs` (default 1) sources × `max_langs` (default 13) languages per run, **greenfield-first** — sources with one or more missing target files always win batch slots over drift-only sources, and within a source the still-needed languages are translated missing-first then drift up to `max_langs` (remaining languages drain in later runs). Pass 2 is **validator-first**: the bash validator does all structural parity at near-zero model-token cost and the agent only reads back the files it flags. This workflow only uses bash/edit/safe-outputs — its MCP/tool surface is deliberately minimal so the per-turn tool-schema context (the ~3.5× weighting driver that aborted run #26641603577 at 27.0M / 25M weighted) no longer dominates; wall-clock (Timer A/B) and the 200-file safe-outputs cap, not the token budget, are now the governing limits — see `07-commit-and-pr.md §Deadline enforcement`. See `TRANSLATION_GUIDE.md §Executive Brief Markdown Translations` for the authoritative content contract." strict: false imports: - ../prompts/00-base-contract.md - ../prompts/01-bash-and-shell-safety.md - ../prompts/02-mcp-access.md - ../prompts/07-commit-and-pr.md on: schedule: # Three runs every day (7 days/week): morning, midday, evening UTC. # Intentional day-of-week range notation (`0-6`) instead of `*`: avoids the # gh-aw fixed-daily-time schedule warning (same pattern as # news-evening-analysis.md `'0 16 * * 6-6'`). Do not simplify to `*`. - cron: "0 9 * * 0-6" - cron: "0 14 * * 0-6" - cron: "0 19 * * 0-6" workflow_dispatch: inputs: article_date: description: 'Restrict the scan to a single article date (YYYY-MM-DD). Leave empty to scan all dates.' required: false subfolder: description: 'Restrict the scan to a single document type (propositions, motions, committee-reports, interpellations, evening-analysis, week-ahead, month-ahead, quarter-ahead, year-ahead, election-cycle, weekly-review, monthly-review, realtime-pulse, …). Leave empty to scan all.' required: false languages: description: 'Target languages (comma list of `sv,da,no,fi,de,fr,es,nl,ar,he,ja,ko,zh` | `nordic-extra` | `eu-extra` | `cjk` | `rtl` | `all-extra`). Default: all-extra (all 13 non-English languages).' required: false default: all-extra max_briefs: description: 'Maximum number of source executive-brief.md files to translate this run (range 1–7; default 1). Hard cap: 7 (keeps under safe-outputs 200-file ceiling with 13 languages = 91 files). Lowered from 3 to 1 after run #26633644372 exceeded the Copilot effective-token cap before completing source #2.' required: false default: '1' max_langs: description: 'Maximum number of target languages to translate per run (range 1–13; default 13 = a full source in one run). The agent writes one full file per language in a single Copilot session, so the session''s weighted effective-token total scales with the language count **times the per-turn tool-schema surface**. The dominant cost was the latter: run #26641603577 hit 27.0M weighted (> the 25M Copilot per-session cap) at 13 languages while the workflow still loaded the full github `toolsets:[all]` + three data-MCP servers on every turn. That dead-weight schema was removed (the agent only uses bash/edit/safe-outputs), collapsing the per-turn footprint, so a full 13-language source now fits in one run. The worklist builder selects the missing/drifted languages **greenfield-first** and caps them at this value; remaining languages are picked up by the next scheduled run because the source stays classified MISSING/DRIFT until every requested language is present and current. Lower this first if an unusually large brief still approaches the cap.' required: false default: '13' force_retranslate: description: 'When true, ignore the `` drift check and re-translate every matching source. Use sparingly — drains the safe-outputs file budget fast.' required: false default: 'false' analysis_depth: description: 'Analysis depth label echoed into validator output for parity with content workflows (standard, deep, comprehensive). Does not change translation behaviour.' required: false default: standard # Shallow checkout (fetch-depth: 1) for fast safe_outputs Checkout. The prerequisite step # fetches GITHUB_SHA on demand for bundle-apply, making full-history clones unnecessary. # gh-aw v0.76.0+ honours `checkout.fetch-depth` (see compiler_safe_outputs_steps.go). checkout: fetch-depth: 1 permissions: contents: read issues: read pull-requests: read actions: read discussions: read security-events: read copilot-requests: write timeout-minutes: 60 # Cost-focus guardrails. `max-turns` caps agent tool-calling turns (runaway-loop circuit # breaker) and is tunable repo-wide via the NEWS_MAX_TURNS variable. `max-ai-credits` is a # per-run AI Credits ceiling — expressions are unsupported here, so a literal is required. max-turns: ${{ vars.NEWS_MAX_TURNS || 300 }} max-ai-credits: 2500 concurrency: group: gh-aw-news-translate-briefs-${{ inputs.article_date || 'batch' }}-${{ inputs.subfolder || 'all' }} job-discriminator: ${{ inputs.article_date || 'batch' }}-${{ inputs.subfolder || 'all' }} cancel-in-progress: true features: mcp-gateway: true runtimes: node: version: "26" network: allowed: # ── Ecosystem identifiers (gh-aw built-ins) ────────────────────────────── - node # Browser downloads for Playwright CLI mode (kept for cross-workflow network parity) - playwright - github - defaults # ── Container registries (node:26-alpine pulls for SCB + WB MCP) ───────── # `containers` ecosystem covers ghcr.io + *.docker.io + Docker Hub auth/CDN endpoints. - containers # ── Riksdag / Regering MCP server + Swedish parliament + government ────── - riksdag-regering-ai.onrender.com - data.riksdagen.se - www.riksdagen.se - riksdagen.se - www.regeringen.se - regeringen.se - www.government.se - g0v.se # ── SCB (Statistics Sweden) ────────────────────────────────────────────── - api.scb.se - www.scb.se - scb-mcp.onrender.com # ── IMF (canonical economic-data source — see ECONOMIC_DATA_CONTRACT.md) ─ - api.imf.org - data.imf.org - www.imf.org - dataservices.imf.org - datamarketplace.imf.org # ── World Bank (governance / environment / social residue) ─────────────── - api.worldbank.org - data.worldbank.org - datahelpdesk.worldbank.org - governance.worldbank.org - www.worldbank.org # ── European Parliament + EU institutions ──────────────────────────────── - www.europarl.europa.eu - europarl.europa.eu - ec.europa.eu - eur-lex.europa.eu - data.europa.eu - www.consilium.europa.eu - digital-strategy.ec.europa.eu - economy-finance.ec.europa.eu - www.enisa.europa.eu # ── Swedish independent agencies, courts, regulators ───────────────────── - www.statskontoret.se - statskontoret.se - www.lagradet.se - lagradet.se - www.riksbank.se - www.riksrevisionen.se - www.konj.se - www.finanspolitiskaradet.se - www.regelradet.se - www.esv.se - www.ei.se - www.energimyndigheten.se - www.naturvardsverket.se - www.boverket.se - www.do.se - www.domstol.se - www.imy.se - www.konkurrensverket.se - www.kriminalvarden.se - www.migrationsverket.se - www.msb.se - www.av.se - www.svt.se - www.val.se - bra.se # ── Hack23 owned platforms ─────────────────────────────────────────────── - hack23.com - www.hack23.com - hack23.github.io - riksdagsmonitor.com - www.riksdagsmonitor.com - riksdagsmonitor.hack23.com - riksdagsmonitor.pages.dev - euparliamentmonitor.com - www.euparliamentmonitor.com - ciacompliancemanager.com - www.ciacompliancemanager.com - blacktrigram.com - www.blacktrigram.com # ── GitHub raw content is covered by the `github` ecosystem identifier above. # Pinned FQDN remains in `safe-outputs.allowed-domains` below (only FQDNs allowed there). # MCP surface is intentionally minimal: translation reads/writes files via bash + edit # and ships the PR via safe-outputs, so it calls NO data-MCP, github-MCP, web-fetch or # agentic-workflows tool. Every declared tool's JSON schema is re-sent on every model turn, # so the cumulative weighted effective-token total scales with the schema surface × turns — # this was the dominant driver of the 3.5× weighting that aborted run #26641603577 at 27.0M. # Only riksdag-regering is kept (its URL is asserted by tests/network-diagnostics.test.ts) and # narrowed to the single health-gate tool documented in 02-mcp-access.md. The scb / world-bank # data servers are dropped here (their egress domains stay in network.allowed for uniformity). mcp-servers: riksdag-regering: url: https://riksdag-regering-ai.onrender.com/mcp allowed: ["get_sync_status"] tools: startup-timeout: 180 timeout: 120 bash: true edit: cache-memory: key: news-translate-${{ inputs.article_date || 'today' }} retention-days: 14 safe-outputs: threat-detection: continue-on-error: true report-failure-as-issue: false allowed-domains: # Riksdag / Regering MCP server + Swedish parliament + government - riksdag-regering-ai.onrender.com - data.riksdagen.se - www.riksdagen.se - riksdagen.se - www.regeringen.se - regeringen.se - www.government.se - g0v.se # SCB (Statistics Sweden) - api.scb.se - www.scb.se - scb-mcp.onrender.com # IMF (canonical economic-data source) - api.imf.org - data.imf.org - www.imf.org - dataservices.imf.org - datamarketplace.imf.org # World Bank - api.worldbank.org - data.worldbank.org - datahelpdesk.worldbank.org - governance.worldbank.org - www.worldbank.org # European Parliament + EU institutions - www.europarl.europa.eu - europarl.europa.eu - ec.europa.eu - eur-lex.europa.eu - data.europa.eu - www.consilium.europa.eu - digital-strategy.ec.europa.eu - economy-finance.ec.europa.eu - www.enisa.europa.eu # Swedish independent agencies, courts, regulators - www.statskontoret.se - statskontoret.se - www.lagradet.se - lagradet.se - www.riksbank.se - www.riksrevisionen.se - www.konj.se - www.finanspolitiskaradet.se - www.regelradet.se - www.esv.se - www.ei.se - www.energimyndigheten.se - www.naturvardsverket.se - www.boverket.se - www.do.se - www.domstol.se - www.imy.se - www.konkurrensverket.se - www.kriminalvarden.se - www.migrationsverket.se - www.msb.se - www.av.se - www.svt.se - www.val.se - bra.se # Hack23 owned platforms - hack23.com - www.hack23.com - hack23.github.io - riksdagsmonitor.com - www.riksdagsmonitor.com - riksdagsmonitor.hack23.com - riksdagsmonitor.pages.dev - euparliamentmonitor.com - www.euparliamentmonitor.com - ciacompliancemanager.com - www.ciacompliancemanager.com - blacktrigram.com - www.blacktrigram.com # GitHub raw content - raw.githubusercontent.com max-patch-size: 10240 max-patch-files: 100 create-pull-request: labels: [agentic-news, translation, executive-brief] draft: false expires: 14d max: 1 if-no-changes: warn # Don't fail when nothing changed (resilience) fallback-as-issue: true # If org disables Actions PR creation, fall back to an issue with branch link protected-files: policy: allowed add-comment: {} steps: - name: News pre-warm & pre-flight (composite) uses: ./.github/actions/news-prewarm with: imf-sdmx-subscription-key: ${{ secrets.IMF_SDMX_SUBSCRIPTION_KEY }} - name: Resolve workflow inputs uses: ./.github/actions/news-resolve-inputs with: subfolder: news-translate article-date: ${{ inputs.article_date }} analysis-depth: ${{ inputs.analysis_depth }} default-analysis-depth: standard languages: ${{ inputs.languages }} max-briefs: ${{ inputs.max_briefs }} force-retranslate: ${{ inputs.force_retranslate }} translate-subfolder: ${{ inputs.subfolder }} - name: Build executive-brief translation work list id: worklist env: ARTICLE_DATE_INPUT: ${{ github.event.inputs.article_date }} SUBFOLDER_INPUT: ${{ github.event.inputs.subfolder }} LANGUAGES_INPUT: ${{ github.event.inputs.languages }} MAX_BRIEFS_INPUT: ${{ github.event.inputs.max_briefs }} MAX_LANGS_INPUT: ${{ github.event.inputs.max_langs }} FORCE_RETRANSLATE: ${{ github.event.inputs.force_retranslate }} run: | set -euo pipefail # Resolve target languages. LANGUAGES_RAW="${LANGUAGES_INPUT:-all-extra}" case "$LANGUAGES_RAW" in all-extra) LANGS="sv,da,no,fi,de,fr,es,nl,ar,he,ja,ko,zh" ;; nordic-extra) LANGS="sv,da,no,fi" ;; eu-extra) LANGS="de,fr,es,nl" ;; cjk) LANGS="ja,ko,zh" ;; rtl) LANGS="ar,he" ;; *) LANGS="$LANGUAGES_RAW" ;; esac echo "Target languages: $LANGS" # Resolve batch cap (1–7 hard range). MAX_BRIEFS="${MAX_BRIEFS_INPUT:-1}" if ! [[ "$MAX_BRIEFS" =~ ^[1-7]$ ]]; then echo "::warning::max_briefs '$MAX_BRIEFS' out of range; clamping to 1" MAX_BRIEFS=1 fi echo "Max source briefs this run: $MAX_BRIEFS" # Resolve per-run language cap (1–13 hard range; default 13 = a full source # in one run). The agent writes one full file per language in a single Copilot # session, so the weighted effective-token total scales with the language count # times the per-turn tool-schema surface. The latter was the dominant cost: run # #26641603577 hit 27.0M weighted (> the 25M cap) at 13 languages while the heavy # github toolsets:[all] + three data-MCP server schemas were still re-sent every # turn. That dead-weight surface was removed (this workflow only uses bash/edit/ # safe-outputs), so a full 13-language source now fits under the cap. The batching # block below caps the languages handled per run to this value, greenfield-first. MAX_LANGS="${MAX_LANGS_INPUT:-13}" if ! [[ "$MAX_LANGS" =~ ^([1-9]|1[0-3])$ ]]; then echo "::warning::max_langs '$MAX_LANGS' out of range; clamping to 13" MAX_LANGS=13 fi echo "Max target languages this run: $MAX_LANGS" # Discover candidate sources. ROOT="analysis/daily" # Worklist file written to GITHUB_WORKSPACE so it is visible inside the # AWF container (`--add-dir "${GITHUB_WORKSPACE}"`); /tmp on the runner # is NOT mounted into the container and the agent cannot read it. WORKLIST_FILE="${GITHUB_WORKSPACE}/.exec-brief-worklist.txt" # Emit a single canonical bundle to BOTH $GITHUB_OUTPUT (for downstream # GHA steps) AND $GITHUB_ENV (so `awf --env-all` forwards the values # to the agent's bash sandbox — agents have no access to step outputs). emit_bundle() { local worklist="$1" langs="$2" max="$3" missing="$4" drift="$5" { echo "TRANSLATION_WORKLIST=$worklist" echo "TRANSLATION_LANGS=$langs" echo "MAX_BRIEFS=$max" echo "MAX_BRIEFS_RESOLVED=$max" echo "MISSING_COUNT=$missing" echo "DRIFT_COUNT=$drift" echo "EXEC_BRIEF_WORKLIST_FILE=$WORKLIST_FILE" } >> "$GITHUB_OUTPUT" { echo "TRANSLATION_WORKLIST=$worklist" echo "TRANSLATION_LANGS=$langs" echo "MAX_BRIEFS=$max" echo "MAX_BRIEFS_RESOLVED=$max" echo "MISSING_COUNT=$missing" echo "DRIFT_COUNT=$drift" echo "EXEC_BRIEF_WORKLIST_FILE=$WORKLIST_FILE" } >> "$GITHUB_ENV" # Always write the file (empty when worklist is empty) so the agent # can rely on a stable path. : > "$WORKLIST_FILE" if [ -n "$worklist" ]; then printf '%s\n' "$worklist" | tr ',' '\n' > "$WORKLIST_FILE" fi } if [ ! -d "$ROOT" ]; then emit_bundle "" "$LANGS" "$MAX_BRIEFS" "0" "0" echo "No $ROOT directory — nothing to translate." exit 0 fi if [ -n "${ARTICLE_DATE_INPUT:-}" ] && [ -n "${SUBFOLDER_INPUT:-}" ]; then SCAN_PATTERN="$ROOT/$ARTICLE_DATE_INPUT/$SUBFOLDER_INPUT" elif [ -n "${ARTICLE_DATE_INPUT:-}" ]; then SCAN_PATTERN="$ROOT/$ARTICLE_DATE_INPUT" else SCAN_PATTERN="$ROOT" fi # Build list of (date, source-path) tuples, oldest date first. CANDIDATES=$(find "$SCAN_PATTERN" -type f -name 'executive-brief.md' 2>/dev/null \ | sort) if [ -z "$CANDIDATES" ]; then emit_bundle "" "$LANGS" "$MAX_BRIEFS" "0" "0" echo "No executive-brief.md sources found under $SCAN_PATTERN." exit 0 fi # For each source, classify it as MISSING (≥1 requested language file absent # or empty) or DRIFT (every requested language file present but ≥1 has a # stale `` trailer relative to the current source # commit SHA). When `force_retranslate=true`, every source with at least # one existing target file is reclassified as DRIFT (forces a re-pass); # sources missing any target stay in MISSING. # # Selection policy (greenfield-first): # 1. Fill `MAX_BRIEFS` batch slots with MISSING sources, oldest date first # (159-source backlog is drained before any drift-fix work). # 2. Only if MISSING is empty after the scan, fill remaining slots with # DRIFT sources, oldest date first. # This guarantees runs never "redo already-translated briefs" while the # untranslated backlog still has work — see PR description for context. MISSING=() DRIFT=() IFS=',' read -ra LANG_LIST <<< "$LANGS" while IFS= read -r SRC; do [ -n "$SRC" ] || continue DIR=$(dirname "$SRC") SRC_SHA=$(git log -1 --format=%H -- "$SRC" 2>/dev/null || echo "") HAS_MISSING=0 HAS_DRIFT=0 for L in "${LANG_LIST[@]}"; do TGT="$DIR/executive-brief_${L}.md" if [ ! -s "$TGT" ]; then HAS_MISSING=1 continue fi if [ -n "$SRC_SHA" ] && ! grep -q "" "$TGT" 2>/dev/null; then HAS_DRIFT=1 fi done if [ "$HAS_MISSING" -eq 1 ]; then MISSING+=("$SRC") elif [ "${FORCE_RETRANSLATE:-false}" = "true" ] || [ "$HAS_DRIFT" -eq 1 ]; then DRIFT+=("$SRC") fi done <<< "$CANDIDATES" MISSING_COUNT="${#MISSING[@]}" DRIFT_COUNT="${#DRIFT[@]}" echo "Backlog classification: MISSING=$MISSING_COUNT DRIFT=$DRIFT_COUNT" # Fill batch: MISSING first (greenfield), DRIFT only if capacity remains. WORK=() for SRC in "${MISSING[@]}"; do WORK+=("$SRC") if [ "${#WORK[@]}" -ge "$MAX_BRIEFS" ]; then break; fi done if [ "${#WORK[@]}" -lt "$MAX_BRIEFS" ]; then for SRC in "${DRIFT[@]}"; do WORK+=("$SRC") if [ "${#WORK[@]}" -ge "$MAX_BRIEFS" ]; then break; fi done fi if [ "${#WORK[@]}" -eq 0 ]; then emit_bundle "" "$LANGS" "$MAX_BRIEFS" "$MISSING_COUNT" "$DRIFT_COUNT" echo "All discovered sources are up to date for the requested languages." exit 0 fi # Categorise the chosen batch for the log line: how many slots were filled # with greenfield (MISSING) vs drift-fix (DRIFT) work. WORK_MISSING=0 WORK_DRIFT=0 for SRC in "${WORK[@]}"; do for M in "${MISSING[@]:-}"; do if [ "$M" = "$SRC" ]; then WORK_MISSING=$((WORK_MISSING+1)); continue 2; fi done WORK_DRIFT=$((WORK_DRIFT+1)) done WORKLIST_JOINED=$(printf '%s\n' "${WORK[@]}" | paste -sd ',' -) # --- Per-run language batching (effective-token control) --- # Across the selected batch, collect the requested languages that still # NEED work — missing target file first (greenfield), then drift — in the # canonical order of $LANGS, and translate at most $MAX_LANGS of them this # run. This bounds the number of full files the agent writes in one # Copilot session (the dominant driver of the weighted effective-token # total that aborted run #26641603577 with a 429 before the PR shipped). # The remaining languages are picked up by the next scheduled run because # every selected source stays classified MISSING/DRIFT until all requested # languages are present and current. NEED_MISSING=() NEED_DRIFT=() for L in "${LANG_LIST[@]}"; do L_MISSING=0 L_DRIFT=0 for SRC in "${WORK[@]}"; do DIR=$(dirname "$SRC") SRC_SHA=$(git log -1 --format=%H -- "$SRC" 2>/dev/null || echo "") TGT="$DIR/executive-brief_${L}.md" if [ ! -s "$TGT" ]; then L_MISSING=1 continue fi if [ "${FORCE_RETRANSLATE:-false}" = "true" ]; then L_DRIFT=1 elif [ -n "$SRC_SHA" ] && ! grep -q "" "$TGT" 2>/dev/null; then L_DRIFT=1 fi done if [ "$L_MISSING" -eq 1 ]; then NEED_MISSING+=("$L") elif [ "$L_DRIFT" -eq 1 ]; then NEED_DRIFT+=("$L") fi done NEED_LANGS=("${NEED_MISSING[@]}" "${NEED_DRIFT[@]}") # Fallback: if nothing was flagged as needing work (edge case), fall back # to the full requested list so the run never emits an empty language set. if [ "${#NEED_LANGS[@]}" -eq 0 ]; then NEED_LANGS=("${LANG_LIST[@]}") fi RUN_LANG_LIST=() for L in "${NEED_LANGS[@]}"; do RUN_LANG_LIST+=("$L") if [ "${#RUN_LANG_LIST[@]}" -ge "$MAX_LANGS" ]; then break; fi done RUN_LANGS=$(IFS=','; echo "${RUN_LANG_LIST[*]}") echo "Per-run language batch (cap $MAX_LANGS): [$RUN_LANGS] of requested [$LANGS]" emit_bundle "$WORKLIST_JOINED" "$RUN_LANGS" "$MAX_BRIEFS" "$MISSING_COUNT" "$DRIFT_COUNT" echo "✅ Selected ${#WORK[@]} source brief(s) for translation into [$RUN_LANGS]" echo " (greenfield=$WORK_MISSING / drift-fix=$WORK_DRIFT; total backlog: MISSING=$MISSING_COUNT DRIFT=$DRIFT_COUNT)" echo " Worklist file (agent reads this): $WORKLIST_FILE" printf ' %s\n' "${WORK[@]}" model: claude-sonnet-4.6 engine: id: copilot --- # 🌐 News Translate — Executive Brief Markdown This workflow is the **sole writer** of `analysis/daily/$DATE/$SUB/executive-brief_.md` files. It translates the English-master executive brief (`executive-brief.md`, produced by the per-type news workflows) into 13 non-English sibling files — together with the English source these give "all 14 languages". It never generates original analysis and never modifies `executive-brief.md` itself. > 🔗 **Authoritative content contract:** [`TRANSLATION_GUIDE.md §Executive Brief Markdown Translations`](../../TRANSLATION_GUIDE.md#-executive-brief-markdown-translations). Read this **in full** before producing any translation — it defines verbatim-preserve blocks, always-translate blocks, structural parity, tone register per language, RTL rules, and the validator acceptance checklist. ## Pipeline 1. **Wall-clock checkpoint (mandatory)** — as the **very first bash call** in this run, record `JOB_START=$(date +%s)` and export it. After every source completes Pass 2, re-read `NOW=$(date +%s)` and compute `ELAPSED_MIN=$(( (NOW - JOB_START) / 60 ))`; print it. If `ELAPSED_MIN >= 35`, stop selecting new sources and proceed directly to the validate → stage → commit → `safeoutputs___create_pull_request` flow regardless of remaining `$TRANSLATION_WORKLIST` entries. If `ELAPSED_MIN >= 42`, **halt all translation work immediately** and execute the [`07-commit-and-pr.md §Emergency deadline order of operations`](../prompts/07-commit-and-pr.md) bash block to ship a partial PR before Timer A (60-min job `timeout-minutes`) and Timer B (~60-min Copilot API session) fire. A partial PR is always better than zero output — see run [#26633644372](https://github.com/Hack23/riksdagsmonitor/actions/runs/26633644372), where source #2 failed with `CAPIError: 429 Maximum effective tokens exceeded`. 2. The `Build executive-brief translation work list` pre-flight step has already exported four env vars into the agent sandbox: **`$TRANSLATION_WORKLIST`** (comma-separated repo-relative paths — read this verbatim, or parse the same content line-by-line from the file at **`$EXEC_BRIEF_WORKLIST_FILE`** at `${GITHUB_WORKSPACE}/.exec-brief-worklist.txt`), **`$TRANSLATION_LANGS`** (comma-separated language codes — this is the **per-run language batch**, already capped at `inputs.max_langs` (default 13) and pre-selected **greenfield-first** from the languages that still need work; translate **exactly and only** these codes — do not expand to the full 13-language set), **`$MAX_BRIEFS`** / **`$MAX_BRIEFS_RESOLVED`** (1–7, clamp-applied), and the audit counters **`$MISSING_COUNT`** / **`$DRIFT_COUNT`**. **Do not re-scan the filesystem** to rebuild this list — the pre-flight step has already honoured `inputs.article_date`, `inputs.subfolder`, `inputs.languages`, `inputs.max_briefs`, `inputs.max_langs`, and `inputs.force_retranslate`. Languages omitted from `$TRANSLATION_LANGS` this run are intentionally deferred to the next scheduled run (the source stays MISSING/DRIFT until all requested languages are present and current) — this per-run cap is what keeps the session under the Copilot weighted effective-token cap that aborted run [#26641603577](https://github.com/Hack23/riksdagsmonitor/actions/runs/26641603577). The selector is **greenfield-first**: sources with one or more missing target files (`MISSING`) always win the `max_briefs` batch slots over sources that only need drift-fixes (`DRIFT`); `DRIFT` is only consulted when `MISSING` is empty. If `$TRANSLATION_WORKLIST` is empty (so `$MISSING_COUNT=0` **and** `$DRIFT_COUNT=0`), nothing is pending — proceed to improvement-mode (re-validate every existing translation against the current `` trailer and fix any drift the validator flags). If still nothing changes after improvement-mode, follow `07-commit-and-pr.md §No-op policy`. 3. For each source path in `$TRANSLATION_WORKLIST` (or each line of `$EXEC_BRIEF_WORKLIST_FILE`): 1. **Pass 1 — translate**: Read the source `executive-brief.md` in full **once**. For every language in `TRANSLATION_LANGS`, produce `analysis/daily/$DATE/$SUB/executive-brief_.md` following the TRANSLATION_GUIDE rules — **write each file with one `edit` tool call per language** (never via `python3`, `bash` heredocs, or shell redirection — see `01-bash-and-shell-safety.md §File creation & overwrite strategy`). Apply the per-language tone register from TRANSLATION_GUIDE **at write time** (so Pass 2 never needs a full re-read just to adjust register), and for `ar`/`he` start the file with ``. Preserve every verbatim block (YAML, HTML comments except `source-sha`, `dok_id` codes, Mermaid DSL bodies, code fences, URLs, file paths, evidence-anchor canonical column values). Translate every always-translate block (prose, headings, list items, table cell text, image alt-text, BLUF, decisions, link text). **Do not** read a freshly written translation back into model context to "confirm" it — the validator in Pass 2 is the authoritative gate. 2. **Pass 2 — validate-first & targeted refine**: This is the **primary token-efficiency control**. Run `npx tsx scripts/validate-executive-brief-translations.ts --source --lang "$TRANSLATION_LANGS"` in the runtime shell (the `--lang` scope restricts validation to this run's language batch so deferred languages are not reported as missing). That validator already performs every structural-parity check at near-zero model-token cost: heading / table-row / code-fence / Mermaid-block count parity, `dok_id` and URL set equality, banned-English-phrase detection (`Executive Brief`, `Decisions`, `Confidence`, `BLUF`, …), the `` marker for `ar`/`he`, and the `` trailer. **Do not read passing translations back into model context** — reading every target file back in full is exactly what drove run [#26633644372](https://github.com/Hack23/riksdagsmonitor/actions/runs/26633644372) to `7.6M` effective tokens and a `429 Maximum effective tokens exceeded` abort. Only read back (and re-translate / fix) the specific `executive-brief_.md` files the validator reports as failing, then re-run the validator until it is clean for the requested languages. 3. **Append the source-revision marker**: compute `SRC_SHA=$(git log -1 --format=%H -- )` in the runtime shell. Use the `edit` tool to add or replace the final `` line at the end of every translation file **whose language is in `$TRANSLATION_LANGS`** (consistent with the file-write contract — no `>>` or `echo`). Do not touch translation files for languages outside the current batch — their marker must remain stale (or absent) so the next run correctly classifies them as DRIFT/MISSING. This is the drift signal future runs use to decide whether to retranslate. 4. **Title post-processing**: for each language code `$L` in `$TRANSLATION_LANGS`, run `npx tsx scripts/postprocess-translated-brief.ts analysis/daily/$DATE/$SUB/executive-brief_$L.md` to re-apply the renderer's `cleanArticleTitle` pipeline on the translated H1. This strips locale-specific boilerplate prefixes (`Exekutiv sammanfattning — `, `Zusammenfassung — `, `执行摘要：…`) and trailing date suffixes that the Pass 1 translation may have left in place. The helper only rewrites when the cleaned title differs from the source; it never falls back to a BLUF synthesis. Verify the script reports `✓` or `✏️` for every file (no `✗`). Do not pass `executive-brief_*.md` as a glob — that would rewrite deferred-language files outside `$TRANSLATION_LANGS` without subsequent validation, defeating the per-run language cap. 5. **Final validate**: re-run `npx tsx scripts/validate-executive-brief-translations.ts --source --lang "$TRANSLATION_LANGS"` (when available) to confirm the title post-processing in sub-step 4 did not break parity, or fall back to the structural sanity checks listed in TRANSLATION_GUIDE §Acceptance checklist. Fix any failures by re-translating only the offending language (read back only that file). Do not commit a file that fails validation. 6. **Re-evaluate the wall-clock checkpoint** (from step 1) before starting the next source. Never start a new source after `ELAPSED_MIN >= 35`. 4. Stage every newly written / refreshed `executive-brief_.md` file. **Do not** touch `executive-brief.md` itself, any HTML under `news/`, any file under `analysis/daily/*/article.md`, or any non-translation file. 5. Call `safeoutputs___create_pull_request` **exactly once** covering the whole batch, **by agent minute 42 at the latest** (hard deadline 45 per `07-commit-and-pr.md §Deadline enforcement`). The branch prefix `news/translate/briefs/` is enforced by the safe-outputs config (`07-commit-and-pr.md`). ## Inputs - `article_date` — optional. Restrict scanning to a single date folder. - `subfolder` — optional. Restrict scanning to a single document type folder. - `languages` — default `all-extra` (= `sv,da,no,fi,de,fr,es,nl,ar,he,ja,ko,zh`). Aliases: `nordic-extra`, `eu-extra`, `cjk`, `rtl`, `all-extra`. Comma list also accepted. - `max_briefs` — default `1`, range `1–7`. Caps the number of **source** files processed in this run; total file output = `max_briefs × |TRANSLATION_LANGS|` (= `1 × 13 = 13` at the defaults; hard-cap worst case `7 × 13 = 91`, safely under the 200-file safe-outputs cap). The default was lowered from 3 to 1 after run #26633644372 hit the Copilot effective-token limit before finishing source #2. - `max_langs` — default `13` (= a full source in one run), range `1–13`. Caps the number of **target languages** translated this run. The builder pre-selects the still-needed languages **greenfield-first** (missing files first, then drift) and truncates to this many. Run #26641603577 hit 27.0M weighted effective tokens (> the 25M cap) at 13 languages, but that was while the heavy `github toolsets:[all]` + three data-MCP server schemas were re-sent every turn; that dead-weight surface has been removed (this workflow only uses bash/edit/safe-outputs), collapsing the per-turn footprint so a full 13-language source now fits in one run. Lower this first if an unusually large brief approaches the cap. - `force_retranslate` — default `false`. When `true`, every requested language is rewritten even if the `` marker matches. - `analysis_depth` — default `standard`. Echoed into the validator output for parity with content workflows; does not change translation behaviour. ## Batch-size rationale | Quantity | Value | Source | |----------|-------|--------| | Average source size | ~650 words (range 232–2040) | `wc -w` over `analysis/daily/**/executive-brief.md` | | Per-language Pass 1 write + validator gate | ~45–75 s | Sonnet-class translation budget; Pass 2 is bash-validated, not re-read into context | | Per-source wall time (≤`max_langs` languages, sequential) | ~15–25 min at `max_langs=13` | derived | | Default `max_briefs` / `max_langs` | 1 source / 13 languages | a slim per-turn MCP/tool surface keeps the weighted effective-token total under Copilot's per-session cap even at a full 13-language source | | Per-session token driver | per-turn tool-schema surface × turn count | **before the fix** the dominant cost was the fixed per-turn schema (github `toolsets:[all]` + three data-MCP servers) re-sent every turn — the ~3.5× weighting multiplier. The translate agent calls none of those tools, so the surface was stripped to bash/edit/safe-outputs (+ one riksdag health-gate tool); the per-turn footprint now collapses toward the raw token count | | Run #26641603577 (pre-fix) | 13 langs → 27.0M weighted / 25M cap → 429 before PR | gh-aw raw 7.67M; the weighted metric was ~3.5× raw and exceeded the cap — driven by the dead-weight per-turn schema, now removed; **no PR shipped** | | Primary token levers | **strip unused per-turn MCP/tool schema** + validator-first Pass 2 | removing the data-MCP servers + github `toolsets:[all]` cuts the fixed per-turn cost so the 25M cap stops being the limiter; structural parity stays in the bash validator | | Per-run file output | `max_briefs × max_langs` = `1 × 13` = **13 files** (default) | well under safe-outputs `max-patch-files: 100` | | Daily throughput (3 runs) | up to **39 language-files / day** at defaults | wall-clock (Timer A/B) and the 100-file cap, not the token budget, are now the governing limits | | Current backlog drain (≥159 untranslated sources) | greenfield-first, oldest source + missing languages first | linear | If a run is behind schedule at agent minute 30 with translations still pending, the agent MUST trim `max_briefs` downward (drop the last selected source) rather than skip the Pass 2 validator gate — quality over completeness. A partial PR is always better than missing Timer A (job `timeout-minutes: 60`). The default `max_briefs=1` was set after run [#26633644372](https://github.com/Hack23/riksdagsmonitor/actions/runs/26633644372) repeatedly hit `CAPIError: 429 Maximum effective tokens exceeded` while attempting source #2; the validator-first Pass 2 above removes the read-back token sink, and stripping the unused per-turn MCP/tool schema (github `toolsets:[all]` + the scb/world-bank data servers — none of which this workflow ever calls) collapses the fixed per-turn cost that drove the ~3.5× weighting, so the weighted effective-token total stays under the 25M cap and the PR safe-output is reached even at the default 13-language batch. ## Rules specific to this workflow - **Never** generate original analysis or write files outside `analysis/daily/**/executive-brief_.md`. - **Never** modify the English source `executive-brief.md` — it is owned by per-type news workflows. - **Validate every translation** with `scripts/validate-executive-brief-translations.ts` before commit. Re-translate (do not commit) any file that fails. - Keep the PR under the safe-outputs 200-file cap. The default `max_briefs=1` × 13 languages = 13 files leaves ample headroom; the hard cap `max_briefs=7` × 13 = 91 stays well under 200. The previous default of 3 was reduced after run [#26633644372](https://github.com/Hack23/riksdagsmonitor/actions/runs/26633644372) hit the Copilot effective-token cap before the PR call. - **Per-language idempotency, not workflow-level no-op**: - The pre-flight selector is **greenfield-first**: it classifies every candidate source as `MISSING` (≥1 requested language file absent or empty) or `DRIFT` (every requested language file present but ≥1 has a stale `` trailer), then fills the `max_briefs` batch slots with `MISSING` sources oldest-first and only falls back to `DRIFT` once `MISSING` is empty. Drift-fix work **never** displaces an untranslated brief while the backlog still has missing files. - A target language is skipped *for that language only* when **all three** hold: 1. `executive-brief_.md` exists and is non-empty, 2. its `` trailer matches the current `git log -1 --format=%H -- executive-brief.md`, **and** 3. `scripts/validate-executive-brief-translations.ts` passes for that language. - When **all** requested languages for all candidate sources satisfy the three conditions, the run enters **translation-improvement mode**: re-run the validator across every translation, fix any drift / structural-parity regressions / RTL-marker omissions the validator flags, refresh `` trailers if the source has been recommitted since, and commit the resulting changes. Append the improvement-mode rerun marker per `07-commit-and-pr.md §No-op policy`. - `safeoutputs___noop` is only allowed under the conditions in `07-commit-and-pr.md §No-op policy`. "All translations already exist and are valid" is **never** a noop trigger; it is an improvement trigger. ## Time budget > 🟡 **Plan to call `safeoutputs___create_pull_request` by agent minute 42 (hard deadline 45)** to reserve job-level headroom for setup variance and the safe-outputs runner. The operative constraint is Timer A (job `timeout-minutes: 60`) and Timer B (~60-min Copilot API session). See `00-base-contract.md §Session timing` and `07-commit-and-pr.md §Deadline enforcement`. The default `max_briefs=1` is sized to keep runs below the Copilot effective-token ceiling; exceeding it risks repeating run [#26633644372](https://github.com/Hack23/riksdagsmonitor/actions/runs/26633644372), where the session aborted on source #2 with `CAPIError: 429 Maximum effective tokens exceeded`. **Single run** (target ~36–40 agent minutes in a 60-min job, hard deadline 45 agent minutes for the PR call): | Minutes | Phase | |---------|-------| | 0–3 | MCP pre-warm; work-list resolved into `$TRANSLATION_WORKLIST` env var + `$EXEC_BRIEF_WORKLIST_FILE` (`${GITHUB_WORKSPACE}/.exec-brief-worklist.txt`); record `JOB_START=$(date +%s)` | | 3–5 | Read TRANSLATION_GUIDE §Executive Brief Markdown Translations + selected sources | | 5–32 | Pass 1 write + validator-first Pass 2 × `max_briefs` sources × all requested languages; re-evaluate `ELAPSED_MIN` after every source and stop selecting new sources once `ELAPSED_MIN >= 35` | | 32–38 | Final validation with `scripts/validate-executive-brief-translations.ts`; append `` trailers; stage scoped files | | 38–42 | **One** `safeoutputs___create_pull_request` call — **HARD DEADLINE agent minute 45**; if `ELAPSED_MIN >= 42` at any point, switch to the `07-commit-and-pr.md §Emergency deadline order of operations` bash block immediately | If the batch cannot finish under this budget, commit the translations completed so far and call `safeoutputs___create_pull_request` with label `partial`; the next scheduled run picks up the remaining work. A partial PR is always better than losing the whole batch to Timer A. All non-workflow-specific rules are in the imported modules — do not restate them here.