Command Center

○ NOISE (0-1) · ▭ FAINT (2) · △ DETECT (3) · ◇ DIAMOND (4) · ⬠ LOCK (5) · ⬡ CONFIRM (6)

Generate an article on demand. Uses Opus 4.6 Extended Thinking (16K budget). Arbiter runs if ticker provided.

Seed / Prompt Author Grant Bishop Charlotte Cross Mei Lin Chen

SELECT name FROM sqlite_master WHERE type='table';

Runs non-destructive schema additions (creates tables if missing, adds columns if missing). Uses /api/migrate.

Run Migrate

Bypasses cron and KV dedup. Routes: /api/trigger/:bot

Pre-Deploy Validator

Paste index.js. Validator checks critical fixes: Opus 4.6 model string, extended thinking with budget_tokens, Charlotte voice-ID typo (S vs R at position 16), KV dedup, required bot functions.

Flips config.openai_model. Ensemble verification stays pinned to gpt-5.4 by design — generational diversity for Final Call cross-checks.

Confirmation Phrase (must be exactly: RESET THE BOARD) SUPER_ADMIN_OTP (from your Worker secrets)

Every threshold below is read from D1 by the Worker on each cron pass. Edit a value, click Save — no Worker redeploy needed. Changes are logged to pipeline_log with the editor's identifier.

Categories: screen (SG criteria) · cadence (publish rhythm) · longshot (Gambit screen) · gate (sanity gate) · words (article length ceilings) · cron (slot times in ET).

Every overnight Scout evaluation lands here. Captures the candidate ranking, screen results, sanity-gate validation, publication status, and (for picks that publish) T+1 entry marks (VWAP + open/close average).

Defensibility-grade. At any point in the future, you can reconstruct exactly why the system said what it said about a name. No quiet deletions.

Aggregations help identify the credit consumer. Look for:
By voice — voice IDs you don't recognize signal a key compromise (rotate the key immediately).
By day — concentration on dates when nothing was published signals silent retries or admin manual triggers.
Recent items — long character counts on individual calls signal chunked retries that succeeded but failed delivery.

Quota panel shows current tier, chars used / limit, and next reset window. Endpoint pulls up to 5 pages × 100 items = 500 items max with pagination ceiling.

Single endpoint that returns everything Claude needs to evaluate system state in one shot. Replaces ad-hoc console probes and screenshots. Includes: config flags, all 31 tuning rows, articles (recent 5 with full content + arbiter scores), internal_scoreboard funnel, KV brief sample (futures, indices, sentiment), 24h costs by provider, recent pipeline errors, last 10 LLM calls, unread alerts, all API health probes, and synthesized data-quality flags.

Workflow: click Fetch & Copy JSON → paste in next Claude session → diff against tonight to see what changed.

Raw JSON (expand to view)

Path	Method	Purpose
/public/articles	GET	Published articles feed
/public/scoreboard	GET	Scoreboard positions
/public/news	GET	Enriched news (cached)
/public/ticker	GET	Ticker scroll data (60s KV cache)
/public/audio/:id	GET	MP3 from R2
/public/alignment/:id	GET	Word-sync JSON
/public/subscribe	POST	Email capture
/api/queue	GET	Pending review
/api/articles/:id/approve	POST	Publish
/api/articles/:id/reject	POST	Reject
/api/scoreboard	GET/POST	Scoreboard
/api/scoreboard/:id	PUT	Update result
/api/notifications	GET	Alerts
/api/config	GET/PUT	Config values
/api/costs	GET	Cost ledger
/api/trigger/:bot	POST	Fire bot
/api/generate	POST	Seed article
/api/migrate	POST	Schema add-only
/api/reset	POST	Board reset (OTP)
/api/log	GET	Pipeline log
/admin/db-query	POST	Inspector SQL
/admin/kv-inspect	POST	KV value
/admin/kv-list	GET	KV keys
/admin/r2-list	GET	R2 objects
/admin/deploy	POST	Removed v11 — returns 410 (use CF dashboard)
/admin/versions	GET	Deploy history
/admin/rollback	POST	Revert to version

Character	Voice ID	Stab	Sim
Grant Bishop	OmpnmugWtOuude52M0Zb	0.62	0.78
Charlotte Cross	FluZbOk3sMJyf5S5UyPF	0.70	0.80
Sal Russo	2J6TQRRydBqUweLCzmqA	0.48	0.75
Mei Lin Chen	JTSlHBBDbMH7ER5WD0yC	0.72	0.80

⚠ Charlotte voice ID: position 16 is S, not R. A prior typo (FluZbOk3sMJyf5R5UyPF) caused Charlotte to fail silently.

Metric	Alert Threshold	Meaning
no_content rate per bot	> 30%	Writer refusing too often — data feed or prompt regression
data_notice rate	> 10%	Scout or upstream API degraded — feeds need attention
arbiter pending rate	> 5%	Articles flagged for review — prompt discipline slipping
arbiter rejects 24h	low but non-zero is healthy	Zero = Arbiter too permissive; many = writer hallucinating

• invented_price_critical_fail — "SPY closed yesterday at $571.40" vs brief SPY=568.21. Expected: CRITICAL fail, score 0, rejected.
• clean_article_passes — Every price matches brief within 0.5%. Expected: score ≥ 90, approved.
• fabricated_pct_critical_fail — "Revenue growth of 27%" vs brief 19%. Expected: CRITICAL fail, rejected.