Wildcard Scholarship

Selection

How we choose

The bar your application is scored against, word for word, plus what happens after.

Selection is a gate, not a contest. We check that the need is real and that the seat would actually get used this month — and then a lottery decides among everyone who clears the bar. We do not rank applicants against each other, and there is no score to beat beyond the bar itself.

That is a deliberate choice, not modesty about our judgement. This program exists to find out what people outside the usual early-adopter demographic do with a tool like this, and we don't think that can be predicted from an essay. Scoring applications on how impressive the proposed use sounds would just select for people who are good at writing applications, which is not what we're looking for.

What happens to your application

• You submit the form and confirm your email address.
• Your answers are scored against the rubric below. The first pass is done by AI, run three times independently on each application; a human reviews every application that is flagged, that the passes disagree on, or that falls near the bar, plus a random sample of the rest.
• Scoring is per criterion. You qualify only if every criterion is at or above the bar — a strong answer on one does not buy a weak answer on another, and there is no overall average.
• Everyone who qualifies goes into the current round's draw. Rounds close twice a month — on the 15th and on the last day of the month, 23:59 UTC — through December 2026. When more people qualify than there are seats left, seats are drawn by lottery; when fewer do, everyone who qualifies receives one.
• You hear back within 7 days of the round closing, either way.

Answers may be written in any language at all, and second-language or plain prose never loses points. We score what you say, not how polished it sounds. A vague but honest answer beats a fluent empty one.

The lottery

Among qualified applicants, seats are assigned by a random draw rather than by ranking, whenever there are more of them than seats. The draw's seed is the NIST Randomness Beacon pulse published at the moment the round closes — a public source we do not control, at a time fixed before the round opens. The seed is therefore not something we pick after seeing who applied.

The draw has two weightings and no others. Session sharing: applicants who opt out enter at half the weight of those who share. Referrals: a former scholar may give us up to three email addresses in their exit survey, and an application from one of those addresses enters at one and a half times the usual weight — we would rather fund a few people who already know each other than the same number of strangers. It has to be the same address; we match on nothing else, and we never write to an address we are given this way. The two multiply, so someone both referred and opted out draws at 0.75. Mechanically, each qualified applicant gets a number from the seed and their email, that number is divided by their weight, and seats go to the smallest results. Nothing else about the draw is weighted, and neither weighting touches the gate: qualifying is the same for everyone.

The rubric

Below is the rubric itself — the same text given to the reviewers, one section per track. The 0–5 anchors are what a reviewer reads while scoring. The bar is 3 on every criterion.

Claude Code Global Scholarship

Scoring philosophy (read first)

This program's thesis is discovery: we fund seats to find out what people outside the default early-adopter demographic do with them. Review is therefore a gate, not a contest. It checks three things — that the need is real, that the seat would actually get used this month, and that having it would make a real difference rather than a marginal one. It does not rank applicants, and it never scores how impressive, novel, or worthy the proposed use sounds: interestingness judged from an essay rewards its imitation, so taste applies only to actual usage at extension time.

The third check is a counterfactual, which is not the same as interestingness. It asks what changes, not whether the change sounds impressive to us: a mundane task that becomes possible scores higher than an ambitious one that was already underway. That distinction is what keeps the gate from rewarding applicants who write well.

Score each criterion 0–5 using the anchors. Qualified for the lottery: every criterion at or above its bar — 3 unless the criterion heading says otherwise. Criteria do not compensate for each other — there is no overall mean. Score substance, not polish: answers may arrive in any language, and second-language or plain prose must never lose points. A vague-but-honest answer beats a polished-but-empty one.

Criteria

1. Need (0–5)

Would this person plausibly not have access otherwise?

• 5 — Clear, concrete barrier (income context, student with no funding, country purchasing-power gap).
• 3 — Plausible need, thin detail.
• 1 — Need claimed but contradicted by other answers.
• 0 — No need stated, or the applicant could clearly afford access.

2. Live use this month (0–5)

Will the seat actually get used? (Claude Code is a terminal-based tool; it needs a computer, time, and either some programming exposure or a credible plan to start.)

• 5 — A concrete thing they intend to do, a computer, and time set aside.
• 3 — Beginner with a credible plan and a real time commitment, or a clear goal with thin logistics.
• 1 — No computer access mentioned, or a timeline clearly incompatible with a 1-month seat.
• 0 — Cannot use the seat (no device, no time, no discernible intent), or copy-pasted boilerplate.

3. Counterfactual (0–5, bar 2)

Would the seat change what this person can do, or mostly speed up something they would do anyway? Score from their answer to "what would this access change for you". The bar here is deliberately low: it removes answers with no discernible difference, not answers whose difference is modest.

• 5 — Makes something possible that currently is not: work they cannot start, a scale they cannot reach, a barrier they have no way around.
• 3 — A real change in the scope or pace of something they are actually doing.
• 1 — Convenience. They would do much the same thing, a little slower.
• 0 — No discernible difference, or the answer restates the goal without saying what changes.

Deliberately not scored

• Past projects, portfolios, follow-through history — these proxy for privilege (having had a computer, bandwidth, and free hours), not for what we're sampling for.
• Impressiveness or social value of the proposed use — taste is applied to actual usage at extension time, never to promises. Criterion 3 asks what changes, not whether we find the change impressive; score a dull unlock above a glamorous speed-up.

Flags (do not affect score; route to human)

• fraud — duplicate-sounding text, template artifacts, implausible combinations.
• minor — any hint the applicant is under 18.
• unclear — the reviewer could not confidently score a criterion.
• gaming — apparent attempt to game the process. Log it; gaming attempts are research data, not auto-rejections — a human decides.

SEA Scholars

Scoring philosophy (read first)

This program's thesis is discovery: we fund seats to find out what researchers outside the default early-adopter demographic do with them. Review is therefore a gate, not a contest. It checks four things — that the applicant is really a student or faculty member at a university in or near the region, that the need is real, that the seat would actually get used this month, and that having it would make a real difference rather than a marginal one. It does not rank applicants, and it never scores how impressive or worthy the proposed use sounds: interestingness judged from an essay rewards its imitation, so taste applies only to actual usage at extension time.

The fourth check is a counterfactual, which is not the same as interestingness. It asks what changes, not whether the change sounds impressive to us: a mundane task that becomes possible scores higher than an ambitious one that was already underway. That distinction is what keeps the gate from rewarding applicants who write well.

Score each criterion 0–5 using the anchors. Qualified for the lottery: every criterion at or above its bar — 3 unless the criterion heading says otherwise. Criteria do not compensate for each other — there is no overall mean. Score substance, not polish: answers may arrive in any language at all, and second-language prose must never lose points. Affiliation proof is requested only at selection, so score affiliation plausibility — don't demand documents.

Criteria

1. University affiliation plausibility (0–5)

• 5 — University email domain matches a real institution in ID/VN/PH/TH/MY — or a nearby country (e.g. Taiwan, Cambodia; welcome, though not the outreach focus) — and program details are coherent.
• 3 — Personal email but coherent, checkable details (university, degree or position, field, year).
• 1 — Vague or inconsistent institutional details.
• 0 — Clearly not affiliated with a university in or near the region.

2. Need (0–5)

Would this person plausibly not have access otherwise?

• 5 — Clear, concrete barrier (student budget, no funding, purchasing-power context).
• 3 — Plausible need, thin detail.
• 1 — Need claimed but contradicted by other answers.
• 0 — No need stated, or the applicant could clearly afford access.

3. Live use this month (0–5)

Will the seat actually get used for their studies, teaching, or research? (A terminal-based tool; needs a computer, time, and some programming exposure — any language, incl. R/Stata/Python — or a credible plan with advisor/lab context.)

• 5 — A specific task in their research, coursework, or teaching, a computer, and time available.
• 3 — Limited exposure but a credible plan and lab context, or a real task with thin logistics.
• 1 — No device access mentioned, or no time this month.
• 0 — Cannot use the seat, or copy-pasted boilerplate.

4. Counterfactual (0–5, bar 2)

Would the seat change what this person can do in their studies, teaching or research, or mostly speed up something they would do anyway? Score from their answer to "what would that unblock or speed up". The bar here is deliberately low: it removes answers with no discernible difference, not answers whose difference is modest.

• 5 — Makes something possible that currently is not: analysis they cannot run, a scale of data they cannot handle, work blocked for want of a tool or a collaborator.
• 3 — A real change in the scope or pace of something they are actually doing.
• 1 — Convenience. They would do much the same work, a little slower.
• 0 — No discernible difference, or the answer restates the goal without saying what changes.

Deliberately not scored

• Publications, theses, projects, follow-through history — these proxy for privilege and seniority, not for what we're sampling for.
• Impressiveness or social value of the proposed use — taste is applied to actual usage at extension time, never to promises. Criterion 4 asks what changes, not whether we find the change impressive; score a dull unlock above a glamorous speed-up.

Flags (do not affect score; route to human)

• fraud — duplicate-sounding text, template artifacts, implausible combinations.
• minor — any hint the applicant is under 18.
• unclear — the reviewer could not confidently score a criterion.
• gaming — apparent attempt to game the process. Log it; gaming attempts are research data, not auto-rejections — a human decides.

If you think we got it wrong

Email hello@wildcardscholarship.org and a human will read it. We can tell you which criterion fell below the bar. We can't undo a lottery draw — a draw you can re-run on request isn't a draw — but a scoring mistake is a scoring mistake, and we would rather hear about it.