GAYA stands for Government Auction Yield Analytics.
An automated analytics system for Indonesia's FR and PBS bond auctions, combining a private per-series machine-learning engine, weekly data orchestration, and a continuously deployed public dashboard.

Dashboard

GAYA turns the next government bond auction into a compact, series-level view. Select an FR or PBS series to compare awarded weighted-average yield (WAY) history against the model path and its published lower and upper bounds.

The dashboard shows:

• the instruments listed for the upcoming DJPPR auction;
• one independently fitted model for each eligible bond series;
• historical actual WAY versus the corresponding backtest prediction;
• the next prediction with lower and upper bounds;
• coupon, remaining tenor, maturity date, and auction window.

Official data source

Auction coverage and realized yields are anchored to announcements published by the Directorate General of Budget Financing and Risk Management (DJPPR), Ministry of Finance of the Republic of Indonesia. The source is kept visible so every series, auction date, and realized WAY shown by GAYA can be traced back to an official release.

_{Direct screenshot of the official DJPPR auction-result announcement, captured in Mozilla Firefox on 9 August 2026. Click the image to open the source page.}

Official publication	Role in GAYA
Rencana Lelang	Establishes the upcoming auction date, settlement date, eligible series, coupon, and maturity.
Hasil Lelang	Supplies realized weighted-average yield, awarded nominal, incoming bids, and bid-to-cover for historical observations.
Public GAYA release	Exposes only the auction context, actual history, model projection, and published bounds required by the dashboard.

Reading the chart

Series	Meaning
	Awarded weighted-average yield published for the historical auction.
	Backtest estimate for historical dates and projection for the next auction.
	Published lower prediction bound.
	Published upper prediction bound.

GAYA Engine

The dashboard is the delivery surface; the GAYA Engine is the private machine-learning computation layer behind it. Every auction cycle is rebuilt from source data, evaluated per eligible bond series, and passed through a release gate before any public artifact is produced.

Three stages. One public release.

Cycle	Schedule	Public-safe outcome
Auction preparation	Monday · 08:15 WIB	Resolve the next auction and persist the eligible FR/PBS universe.
Engine refresh	Tuesday · 09:05 WIB	Refresh validated inputs, rebuild eligible series models, and evaluate the release.
Actual-result refresh	Tuesday · 16:05, 19:05, 22:05 WIB; Wednesday · 09:05 retry	Match official realized WAY by auction date and series, including when the following week's series universe changes.
Continuous delivery	Successful release only	Publish one sanitized payload and trigger the production dashboard build.

Private/public boundary

Engine stage	Responsibility	Exposure
Auction intake	Resolve the next auction, settlement schedule, and eligible FR/PBS universe from official publications.	Private
Data layer	Assemble and validate model-ready histories and contextual inputs for each series.	Private
Per-series ML forecasting	Fit, evaluate, and generate an independent projection for every eligible bond code.	Private
Release gate	Check observation sufficiency, output validity, and diagnostics; retain the last valid release when a run is not publishable.	Private
Prediction contract	Serialize only display-ready actuals, projections, bounds, instrument metadata, and timestamps.	Public

Model boundary: no training dataset, feature recipe, fitted parameter, diagnostic output, serialized model, or callable inference code is shipped to the browser. The public application receives only the sanitized prediction contract. Client-side C++, WebAssembly, and minification are not treated as model protection because browser-delivered code remains inspectable.

If a scheduled auction is unavailable or a series does not have enough observations, the last valid public release remains in place.

Model coverage

	Coverage
	Fixed-rate SUN and Project Based Sukuk series selected from the current auction plan.
	One independently fitted model per eligible bond code, never a pooled model across every series.
	Four historical backtest observations followed by the next auction projection.

SPN, SPNS, and PBSG instruments are intentionally excluded from the public model board.

System stack

Private ingestion, validation, forecasting, and release pipeline.	Historical assembly, schema checks, and model-ready series data.	Monday intake, Tuesday model refresh, validation, and publication.
Typed public contract, model selector, and analytics interface.	Production bundle and dependency-free chart rendering.	Continuous build and global delivery after every public release.

The public application has no login, browser-side model, secret key, or user input form. It reads the current sanitized release from public/data/predictions.json. Dated public-safe snapshots remain under public/data/releases, so realized results are retained even when the next auction contains different series. Model fitting and sensitive inputs remain in the private pipeline.

Operating budget

GAYA is designed to operate at $0 per month on the current schedule and traffic profile. The estimate uses successful workflow durations, configured timeout ceilings, and vendor free-tier limits available on 9 August 2026.

Cost center	Free allowance	GAYA operating estimate	Reset
GitHub Actions	2,000 Linux runner minutes/month	Approximately 17–20 minutes/month under normal runs; below 200 minutes even if both weekly jobs reach their configured timeouts.	Billing cycle
Cloudflare Workers Builds	3,000 build minutes/month	Approximately 4–5 production builds/month; at most 100 minutes if every weekly build reaches the 20-minute timeout.	Monthly
Cloudflare Static Assets	Free and unlimited requests	Dashboard HTML, JavaScript, CSS, images, and the public prediction payload are served as static assets.	Not applicable
Cloudflare Workers runtime	100,000 requests/day	Reserved headroom if a request invokes Worker code instead of matching a static asset.	Daily at 00:00 UTC

The zero-cost policy assumes GitHub Free, Cloudflare Workers Free, a $0 GitHub Actions paid-usage budget, and no paid database, API, VM, container, or external scheduler. If a free limit is exhausted, the preferred failure mode is to pause the run or deployment rather than allow paid overage. Vendor pricing and limits may change; the linked official documentation remains authoritative.

Deployment

Production is served by Cloudflare from the main branch:

gaya.mohammad-raffy.workers.dev

Build command : npm run build
Output         : dist

No frontend environment variable is required.

---

GAYA provides statistical estimates for research and monitoring. It is not an investment recommendation.