openstudy.id

Open data · Indonesia · pipelines, models, dashboards

The analysis is done.
The conclusions are open.

Ten analytical instruments on Indonesian open data: pipelines, models, and evaluation against baselines that already contain the easy knowledge. The measurement is finished on all of them. What none has is a discussion section — because writing one takes someone who knows the subject.

Whythe gap this fills

Journals publish figures. Nobody publishes instruments.

A paper's analysis is usually its most reusable part and its least accessible. The figures are flattened into a PDF, the pipeline lives on one laptop, and the question a reader actually wants to ask — what happens if I change this assumption? — cannot be asked at all.

Repositories archive the code. Notebook hosts run it. Neither builds the thing a non-specialist can use, and no journal will host it for you. That is the gap.

So the work is split by what it takes, not by who owns it. Measurement needs statistics, models and software. Explanation needs the subject. Both get named credit for what they did, and neither has a veto over the other.

Howwhat each side brings

How a case comes together

MeasurementExplanation
Method. Which estimator, which baseline, what would count as working — decided and written down before anything runs The question, and why it matters where it is asked
Models. Built, tuned and scored — statistical or machine-learned, whichever the problem actually needs Domain knowledge. What a result means, what would be surprising, what the literature already settled
Evaluation. Against naive rivals that already contain the easy knowledge, with every gate published including the failures Plausible mechanism. Why the relationship would exist, or why it would not
The instrument. Pipeline, dashboard, figures, all re-runnable The interpretation, the written discussion, and accountability for the finding

Credit follows the work with CRediT roles, so nobody has to pretend: Methodology, Formal analysis, Validation, Software, Data curation and Visualization on one side of that table; Conceptualization, Investigation and Writing on the other. Whoever did a thing is credited for it.

An instrument is published as soon as the left column is finished, without waiting for the right one. Both halves stay open to anyone who can fill them — a domain researcher most obviously, but equally someone who can improve a model or break an evaluation. A conclusion arriving does not close the left column.

Who you arethe terms differ

Two ways in, and they are not the same offer

A domain researcher and someone who builds models are being asked for different things and getting different things back. Rather than write one page that assumes which you are, pick.

Choose how you would take part

You have a finding, a dataset you understand, or a question nobody has measured properly. The measurement gets built for you.

What you get

  • A working instrument for your subject — pipeline, models, evaluation, dashboard — built free
  • First author. The finding and its interpretation are yours
  • You keep copyright and can republish anywhere
  • A DOI on request — archived independently, citable, versioned
  • Thresholds fixed before results, so the evaluation is defensible to a reviewer

What you should know

  • Not peer reviewed. Gates are checking, not refereeing
  • This site has no audience yet. You are betting it gets one
  • Publishing here before a journal makes it a preprint — check your journal's policy first
  • Failed gates are published. If the measurement contradicts your finding, the page says so
  • Implementation is AI-assisted, disclosed on every instrument
  • One person, one server, no grant. It could stop.

The specific gap on each instrument is written on its card below. The longer version.

You do not need a subject or a finding. The most useful thing here takes a day and is the thing this site's credibility actually rests on.

What you doEffortCredit
Break an evaluation. Wrong metric for the problem? Did a baseline see the test set? Does the cross-validation actually prevent leakage? If you succeed, that gets published a dayValidation
Beat a baseline. Build a stronger rival than the one an instrument was scored against. Win or lose, the result is worth publishing a weekFormal analysis
Build an instrument. Pipeline, models, gates and dashboard for someone's finding, to the published standard weeksSoftware · Methodology · Visualization

What you get

  • Named credit on a public artifact, by CRediT role rather than a thank-you line
  • A portfolio piece where the validation is visible — rarer than it should be
  • Your code stays yours, permissively licensed
  • An attack that succeeds is published as a finding, not quietly absorbed

What you should know

  • Thresholds are pre-registered and do not move. If your model loses to a naive baseline, that is what publishes
  • No payment by default. Paid work is separate and offered case by case
  • Existing implementation is AI-assisted — you may be reviewing code a model wrote under direction
  • Capacity is one person's attention. Review is slow

Three openings right now: transit-equity has two failing hard gates — are they fatal or cosmetic? rice-security's R² goes −11.09 to 0.82 through OLS calibration — legitimate, or fitting the answer? fire-haze scores a 3.8% event class with ROC-AUC — is precision-recall the honest metric? The longer version.

Instrumentsnone of the 10 has a conclusion yet

10 instruments, and what each is still open to

Every one is complete as measurement: the pipeline runs, the gates are published including the failures, the dashboard works. What is open differs by instrument — and nothing here ever closes.

StateMeaningOpen to
seeking author No conclusion written yet A domain author, for first authorship — and anyone who can improve a model or break the evaluation
authored A domain author has written the conclusion and holds first authorship Challenge only: a stronger baseline, a flaw in the evaluation, newer data. A successful challenge publishes as its own finding
  • Can satellite lights stand in for regional economic activity in Indonesia?

    What it found. The nowcast is 99.5% intercept and 41% worse than carrying last year forward. The Gibson relationship holds for kota and inverts for kabupaten.

    What is missing. Someone who knows Indonesian regional economic statistics — why it inverts, whether BPS's GRDP methodology explains it, and what lights can honestly proxy in a country with this settlement pattern.

  • How bad will the air be tomorrow, when almost nothing is measuring it?

    What it found. Only 2 of 24 registered public PM2.5 sensors still report. Two US diplomatic stations last published 3,580 days ago.

    What is missing. Someone who knows Indonesian air-quality monitoring — why the network collapsed, what BMKG's own record would add, and what a 24-hour forecast is actually useful for here.

  • Can an ignition model beat the Fire Weather Index over Indonesian peat?

    What it found. It clears both climatology and the FWI at every lead. Two gates fail: transport direction, and the 2015/2019 anchor replay.

    What is missing. Someone who knows Indonesian fire management and peat hydrology — whether those failures matter operationally, what BNPB and KLHK act on, and whether those years are the right blind test.

  • Does a mill catchment tell you anything about who cleared the forest?

    What it found. Three-quarters of alerted hectares fall inside a mill sourcing catchment — but the base-rate test asks what that share would be by chance.

    What is missing. Someone who knows palm supply chains and Indonesian concession law — whether catchment proximity means anything for sourcing responsibility, and what ISPO and RSPO actually require.

  • Reading Java's harvest from radar, and checking it against the official count.

    What it found. Detected harvested area has an R² of −11.09 against BPS KSA before calibration, and 0.82 after.

    What is missing. Someone in agricultural statistics or rice agronomy — whether that calibration is legitimate or is fitting the answer, what KSA's own method assumes, and whether harvest-timing prediction is actionable.

  • How fast, where, and what the ground is doing about it.

    What it found. Vertical land motion across Java, with the fastest kelurahan named and an independent InSAR run as a check.

    What is missing. Someone in geodesy or hydrogeology — how much is groundwater extraction against natural compaction and load, what the coastal defence programme assumes, and which measurements would settle it.

  • Downscaling poverty estimates below the level the survey supports.

    What it found. A gradient-boosted model with spatial cross-validation, taken down to kecamatan.

    What is missing. Someone who knows BPS poverty measurement — whether estimates at that resolution can responsibly be used for targeting, what Susenas sampling supports, and where a sub-district figure would do harm.

  • The Hour seeking author

    What can you reach in an hour, and who cannot reach anything?

    What it found. Access routed on published timetables. Two hard gates fail: timetable sanity and network integrity.

    What is missing. Someone in Jabodetabek transport planning — whether scheduled-not-congested times are usable, what the missing angkot network does to the equity picture, and what planners would actually ask this.

  • Where the economy sits in the product space, and where it is moving.

    What it found. Economic complexity computed from the bilateral trade record, tracked over time.

    What is missing. Someone in trade economics or industrial policy — what complexity does and does not predict for a commodity exporter, and whether the product-space frame fits Indonesia's actual position.

  • What the world's press reported about Indonesia, against its own denominator.

    What it found. Every GDELT record touching Indonesia since 2017, counted against a share-of-everything baseline rather than in isolation.

    What is missing. Someone who studies Indonesian media — what GDELT's Anglophone and online bias does to that picture, whether machine-coded tone survives translation, and what a claim about national narrative can rest on.

If one of these is your subject, the measurement is already done. Writing the explanation earns first author; the measurement earns methods co-author. If instead you can see something wrong with how one was measured, that is a contribution too — see how to contribute.

Read how to contribute first — including the parts that might put you off.