Apply Dossier Evaluator

Further-Education Dossier Evaluator

This skill produces a reproducible, anchored score for a further-education application dossier and a verdict on eligibility and target fit. It does not rewrite any single document; it evaluates the whole package against a fixed scale so two runs on the same dossier land on the same numbers.

Use this when

• The user wants a graded scorecard (per-dimension 1–5 + total), not prose.
• The user asks how competitive they are for a PhD, research master's, scholarship/fellowship, or research internship.
• The user asks whether they are eligible (degree equivalence, cross-border credit expectations, scholarship gating rules).
• The user wants scholarship-mission fit scored (return-home, leadership, development impact) for VEF / DAAD / Fulbright / Erasmus Mundus / MSCA.
• The user wants the same dossier re-scored consistently across iterations.

Do not use this when

• The user wants a qualitative coherence audit and a prioritized fix list with no scores. Use apply-package-auditor.
• The user wants to write/line-edit one artifact: SOP → apply-sop-writer, CV → apply-cv-builder, letters → apply-recommendation-letter-strategist.
• The user wants to choose faculty/programs from scratch → apply-program-fit-mapper.
• No structured profile exists yet → run apply-profile-reader first, then return.

Reproducibility contract

Scoring MUST be anchored, not impressionistic:

• Apply the 1 / 3 / 5 anchor descriptors in references/rubric-descriptors.md for every dimension. Do not score from gut feel.
• Calibrate the total against the worked examples in references/calibration-anchors.md so the scale stays fixed across runs.
• For every score, cite the concrete evidence and the anchor band it matched.

Separate presentation convention from substance. Dimensions 6 (statement quality) and 7 (letters) are the ones most easily corrupted by a convention mismatch: a statement written to a European motivation-letter prompt, a CV on a mandated template, a reference letter in a restrained register, or a grade reported on its native scale can all look weak against expectations set by a different system while the underlying evidence is strong.

The rule: score the evidence, not the packaging. Where a dossier's weakness is a fixable convention mismatch — wrong format, missing section the target expects, unattributed contribution, self-converted grade — record it under Highest-leverage fixes, not as a reduced dimension score. Where the weakness is substantive (no research evidence, no letter-writer who can speak to research ability), score it. State explicitly which of the two you concluded and why, so the call is auditable.

Do not adjust any score up or down on the basis of an applicant's nationality, first language, or home institution's country. Where a credential's recognition genuinely gates admission, that belongs to dimension 1 (eligibility) with references/degree-equivalence.md, not to a quality dimension.

Score sponsor/mission fit (dimension 2) only against published criteria. A funder's mission may legitimately drive scoring where the programme states it — Australia Awards publishes development impact, leadership, and return-home contribution as objectives, and HKPFS publishes four named yardsticks including communication and leadership. Those are scoreable.

What is not scoreable is an inferred national preference. A true geopolitical fact licenses a scoring criterion only when the chain geopolitical objective → programme objective → published selection criterion can be shown. Where the chain breaks, do not score the dimension on it; record the uncertainty instead. Fellowship-specific criteria are also not national criteria — NSF GRFP's Intellectual Merit / Broader Impacts framing applies to GRFP applicants (US citizens, nationals, and permanent residents only), not to US admissions generally.

Score dimension 2 against the scheme's genre, not its nationality. Read the genre off the published objectives: academic-excellence, development/change, leadership/influence, industry/professional, research-cooperation, consortium/mobility, or diplomatic/bilateral. A dossier that is strong for one genre can be genuinely weak for another — a pure research record against a development-change scheme requiring years of professional experience is a real mismatch, not a presentation problem, and should score accordingly. Two schemes from the same national agency frequently belong to different genres with different eligibility, so never infer the genre from the funder's country.

Read the declared theme off the live official page, per run. Like faculty fit (dimension 8), a programme's theme, track, and pathway list are fetched, not remembered: search the web for the programme's or call's own pages — including the specific pathway page where several exist — and take the scope in the programme's own words. Never infer it from the programme's name, and never score against a theme you did not read. Where the page cannot be located or the cycle's call is unpublished, record the theme as unverified and ask for the link instead of scoring it.

A declared theme is published criteria too. Where the programme states a theme, specialisation, track, or topic scope — a consortium master's with named pathways, a mission-scoped doctoral school, a topic-restricted call — relevance to it is scoreable under dimension 2, at the level the applicant actually chooses (fitting the umbrella theme while fitting no individual pathway is a real mismatch). Apply the same substance/packaging split: a dossier whose real work sits inside the theme but never frames itself against it is a highest-leverage fix, while a dossier with no evidence base in the theme and no honest bridge to it is a genuinely low score and may be the wrong target entirely — say so rather than scoring it politely. Do not credit a bridge that would survive substituting a different specialisation; that sentence is filler, not evidence.

Where a programme carries hard post-award constraints — Australia Awards requires returning home and remaining outside Australia for at least two years, with a debt liability for the full scholarship cost otherwise — surface them under eligibility and risk. They change whether the target is viable, not just how it scores.

Volatility rule (read before scoring eligibility or programs)

NEVER treat a cached deadline, GPA cutoff, score minimum, CV/SOP word limit, or funding amount as authoritative — these change every cycle. Where such a number matters, verify it against the live official call and say so in the output. A stale cached cutoff is worse than none, because it will be trusted.

Step 0 — Missing-info gate

Before scoring, confirm you have enough to evaluate. Minimum inputs:

• the target (degree type, country/region, and named program(s) if any)
• CV / publication-project list
• SOP or research statement (or note that it is missing)
• transcript / grade context
• recommender list (who, relationship)
• for scholarships: the named sponsor and its stated mission

If material is missing, list exactly what is needed and score only what is present, marking ungraded dimensions as N/A — insufficient evidence. Do not invent program criteria, professor interests, or eligibility rules.

Step 1 — Target classification

Classify the dossier into exactly one primary target type. This selects the adaptive weighting in Step 3.

If two targets genuinely apply (e.g. a PhD pursued through a scholarship), score both relevant program layers: the academic target with its rubric AND the sponsor layer via Dimension 2.

Step 2 — The 11 evaluation dimensions

Score each 1–5 using references/rubric-descriptors.md anchors. Several dimensions have a different bar per target type; the descriptor file carries the per-type variants.

1. Eligibility & degree equivalence — does the applicant formally qualify? For EU / cross-border targets (does a 4-year Vietnamese BSc satisfy a master's-required EU PhD; US vs EU credit expectations), consult references/degree-equivalence.md. This is a gate: a hard fail here caps the dossier regardless of other strengths — flag it explicitly.
2. Target & program / sponsor fit — alignment to the named program's stated priorities. For scholarships and external programs, consult references/external-programs.md for what each sponsor weights and which documents it requires, then verify all cutoffs/limits against the live official call.
3. Academic record & preparation — transcript rigor, trajectory, relevant coursework, grade context (interpret GPA against institution and grading scale; never against a hardcoded cutoff).
4. Research potential & output — publications, preprints, projects, evidence of independent contribution. Rate publications by venue quality using references/venue-quality.md, not raw counts. A single first-author A*/A paper outweighs several weak-venue entries.
5. Technical & methodological depth — demonstrated command of methods, tooling, and rigor appropriate to the field and target.
6. Statement / research narrative quality — clarity of research identity, credible future direction, specificity (not generic ambition).
7. Letters of recommendation — strength, relevance, and coverage of recommenders relative to the target (research letters for research targets).
8. Faculty / lab fit — concreteness of match to named people or groups. Fetch this live per run (Scholar / DBLP / lab page); do NOT rely on any cached professor file — none exists by design.
9. Leadership, service & impact — especially weighted for scholarships; includes return-home / development-impact / mobility signals where the sponsor requires them.
10. Dossier coherence — does the package tell one consistent story? Work through assets/coherence-checklist.md (SOP lead narrative present in CV; every SOP claim CV-backed; proposal matches target; transcript weakness addressed).
11. Risk & readiness — red flags and submission-readiness. Classify findings Critical / Major / Minor using references/failure-patterns.md.

Step 3 — Adaptive weighting per target type

Combine dimension scores into a weighted total. Weights shift by target type (the four adaptive rubrics). Use these relative emphases; normalize to 100%.

Eligibility is a gate, not a weighted term: if Dimension 1 is a hard fail, report the dossier as ineligible as-is and present the remediation path rather than a competitive score.

Step 4 — Total, tiering, and verdict

Convert the weighted total to a tier, calibrated against references/calibration-anchors.md:

• Top tier (~top 2%) — clear admit/funding signal at competitive targets.
• Strong (~next 8%) — competitive; specific gaps to close.
• Plausible — viable at well-matched targets; material gaps.
• Developing — not yet competitive at the stated target; reframe or build.

State the tier with the per-dimension evidence that drove it, and the single highest-leverage fix.

Output format

## Target classification
[Type A/B/C/D] — [named program(s)] — [reasoning]

## Eligibility verdict
[Eligible / Conditional / Ineligible as-is] — [evidence; live-verify notes]

## Scorecard
| # | Dimension | Score (1–5) | Anchor band matched | Evidence | Weight (this target) |
|---|---|---|---|---|---|
| 1 | Eligibility (gate) | … | … | … | gate |
| … | | | | | |

## Weighted total & tier
[Total] → [Tier], calibrated against calibration-anchors.md

## Strengths
- [Dimension]: [evidence]

## Critical / Major / Minor risks
- [Critical] …
- [Major] …
- [Minor] …

## Coherence findings
- [Checklist item]: [pass/fail + note]

## Declared-theme verdict (per target)
| Target | Declared theme / pathway (programme's own words) | Source page + date checked | Verdict | Target viable? |
|---|---|---|---|---|

<!-- Verdict: core / adjacent, bridged / stretch / outside scope / no declared theme — N/A / unverified — link needed.
     This drops or keeps a TARGET; it is not a second dossier-wide gate (eligibility is the only gate).
     An "outside scope" target is reported as the wrong target, not merely as a low dimension-2 score. -->

## Sponsor / program fit notes
- [Program]: [what it weights] — [fit] — [LIVE-VERIFY: cutoffs/limits]

## Highest-leverage fixes
1. [Fix] — [why it moves the score]

## Live-verification checklist
- [ ] [Each cutoff/limit/deadline that must be confirmed against the official call]

## Ungraded / insufficient evidence
- [Dimension]: [what is missing]

Quality bar

A strong evaluation is anchored (every score traces to a rubric band and concrete evidence), honest about volatility (no cached cutoff presented as fact), and decision-useful (clear tier, clear gate verdict, one highest- leverage fix). If the evaluator cannot defend a score against references/rubric-descriptors.md, the score is not ready.