Skip to content

Open Lab / An allocation experiment

What deserves
your support?

A finite budget makes a preference concrete. Try backing a small piece of useful work, then change your mind as you learn.

An optional experiment alongside collaboration—not a replacement for doing or reviewing the work.

100 points available0 allocated / 100 total

100 points is an illustrative design assumption, not a monetary amount.

Reclaim points from one effort to support another. You don’t need to use them all.

Local allocation experiment; no money, global vote, or funding commitment.

Checking local browser storage…

Three possible next steps.

Editorial proposals—not active initiatives, participating researchers, or endorsements by the linked projects.

Support is your preference, not a measure of scientific validity. Useful negative results count as work; popularity does not prove science.

Proposed work / Reproducibility

Reproduce an open tutorial

Proposed work scope
Reproduce the scikit-learn common-pitfalls example on inconsistent preprocessing. Record the documentation version, Python environment, split, commands, and both prediction-error outputs.
Acceptance criterion
A second reader can rerun the example from a clean environment and compare scaled versus unscaled preprocessing. Include exact outputs and explain any difference from the published example; matching a number alone is not enough.
If it fails / negative results
If dependencies fail or the result differs, keep the error log, environment and smallest failing example. Report the discrepancy rather than tuning until the expected answer appears.
Source material—not evidence of completed workscikit-learn: inconsistent preprocessing ↗

Currently 0 points. Maximum for this effort now: 100.

Local allocation experiment; no money, global vote, or funding commitment.

Evidence Optional / local / unreviewed

Add a source and what it shows, including uncertainty or a negative result. No points required. This does not submit a review or a funding request.

One saved note per effort. Only saved notes enter the export. Avoid private information; inspect the file before sharing it yourself.

Back to proposals ↑

A possible next layer / Not implemented here

From an effort to a Hypercert.

Rather than invent an Open Lab claim format, a future integration could use the existing Hypercerts AT Protocol schemas. An activity claim anchors work scope, contributors, time and location. Attachments, measurements and evaluations are separately authored records referencing {uri,cid}; an evaluator keeps their record on their own PDS.

The current data model says activity records are immutable and on-chain tokenization is not implemented. See the AT Protocol quickstart. Claims, evidence and evaluations do not themselves certify truth, allocate funds, or convey equity or IP.

Local design sketch · NOT ISSUED

For “Reproduce an open tutorial”. Not a validated mint payload, signed record or issued Hypercert. Actual identities, work dates and outcomes are deliberately absent.

{
  "status": "NOT ISSUED",
  "format": "Local design sketch; not a validated mint payload or AT Protocol record.",
  "possibleFutureSchema": "org.hypercerts.claim.activity",
  "proposedWork": {
    "title": "Reproduce an open tutorial",
    "scope": "Reproduce the scikit-learn common-pitfalls example on inconsistent preprocessing. Record the documentation version, Python environment, split, commands, and both prediction-error outputs.",
    "acceptanceCriterion": "A second reader can rerun the example from a clean environment and compare scaled versus unscaled preprocessing. Include exact outputs and explain any difference from the published example; matching a number alone is not enough.",
    "negativeResultHandling": "If dependencies fail or the result differs, keep the error log, environment and smallest failing example. Report the discrepancy rather than tuning until the expected answer appears.",
    "sources": [
      {
        "title": "scikit-learn: inconsistent preprocessing",
        "url": "https://scikit-learn.org/stable/common_pitfalls.html#inconsistent-preprocessing"
      }
    ]
  },
  "missingBeforeAnyIssuance": [
    "Actual contributors and their identities",
    "Work dates and relevant location",
    "Actual work and provenance",
    "Schema validation, informed authorization and a separate explicit publish action"
  ],
  "evidenceModel": "Separately authored attachments, measurements and evaluations may reference an issued activity by {uri,cid}; an evaluator retains their own record on their own PDS."
}

Same-browser storage only. Resetting storage or using different devices creates new local state. This is not a one-person-one-budget guarantee. Other tabs are not synchronized; the last saved edit wins. This is not authentication.