| name | sg-flw-classifier |
| description | Classify papers, PDFs, abstracts, datasets, intervention descriptions, or study protocols using the SG-FLW measurement framework for serious games and food waste, serious games for food waste reduction, games, gamification, and interactive interventions addressing food loss and waste. Use when Codex needs to assign B/K/D/E/L scores, explain evidence-chain certainty, review whether a sustainability game supports food-waste-reduction claims, audit AI over-crediting errors, or plan evaluation methods, telemetry, and evidence boundaries for a food-waste serious game. |
SG-FLW Classifier
Use this skill as a domain-specific evidence appraisal workflow for serious games, games, gamification, and related interactive interventions addressing food loss and waste. SG-FLW classifies the reported or planned measurement chain behind food-waste-reduction claims; it does not judge game quality, learning design quality, usability, engagement, or causal effectiveness.
Do not classify from the short rubric alone. The framework was created to protect evidence boundaries that AI agents commonly collapse: awareness into behavior, engagement into behavior, in-game outputs into real-world outcomes, and environmental game mechanics into environmental conversion.
Required Reference Loading
For any real classification task, read these files before assigning scores:
references/foundations.md: research rationale, literature anchors, scope boundaries, and the SG-FLW evidence identity.references/coding-protocol.md: extraction workflow, decision rules, full B/K/D/E/L scoring protocol, certainty logic, indices, and error taxonomy.
Load these conditionally:
references/reference-corpus.md: when comparing a case to the accepted 35-study corpus, checking corpus-level patterns, borderline decisions, or classifying a CiFoS-like boundary case.references/corpus-knowledge-base.md: when the source is one of the 35 corpus entries, when using corpus examples for calibration, or when an AI agent needs source-linked study memory and over-crediting traps for the accepted corpus.references/prospective-planning.md: when the input is a study protocol, intervention concept, telemetry plan, municipal deployment context, or serious-game design proposal.references/sg-flw-rubric.md: for a compact scoring reminder only after the full protocol is understood.
Core Workflow
- Define the case scope:
- intervention name, source, year, and citation;
- intervention type: serious game, game, gamification, interactive installation, app, dashboard, or boundary case;
- relation to food loss/waste: direct food-waste reduction, broader food-system sustainability, or indirect/boundary relevance;
- evaluation mode: retrospective published-study classification or prospective study-design planning.
- Build an evidence ledger before scoring. For each dimension, extract the measure, timing, unit, aggregation level, instrument/source, population/site, and a short evidence pointer such as section, page, table, or very short quote.
- Score B/K/D/E/L independently using
references/coding-protocol.md.
- Apply conservative coding: when evidence is absent, implied, ambiguous, or only present as author aspiration, code the lower level and explain what is missing.
- Run the commensurability check between baseline (B) and direct behavior/outcome (D): unit, aggregation level, population/site, time window, and food-waste scope must match for reduction claims.
- Compute measurement-chain certainty separately from mechanism evidence:
- K is mechanism evidence and never upgrades waste-reduction certainty by itself.
- Certainty is about the adequacy of the measurement chain, not causal identification.
- Identify calculable indices only when requirements are met:
- WRR requires B>=2 and D3.
- BII requires B3 and D3 with dispersion or raw data.
- EII requires E>=1 and D3.
- PI requires L3 and D>=2.
- State claim boundaries: what the study can support, what it cannot support, and what evidence would be needed to raise each score.
- Run the self-audit in
references/coding-protocol.md before finalizing.
Non-Negotiable Coding Rules
- Knowledge, awareness, attitudes, intentions, perceived behavioral control, locus of control, responsibility attribution, satisfaction, usability, or acceptance are mechanism evidence (K), not direct food-waste behavior (D).
- In-game choices, simulated outcomes, missions, scores, dashboards, virtual waste, or modeled food-system outputs do not count as real-world behavior unless tied to observed or self-reported participant behavior outside the game.
- Engagement metrics such as logins, session duration, levels completed, points, badges, screen time, or retention do not count as D unless they directly record food-waste-related behavior.
- Environmental indicators inside a game do not count as E unless they convert measured food-waste reduction into environmental impact.
- A low SG-FLW profile means food-waste-reduction evidence is missing or weakly reported. It does not mean the game is ineffective, badly designed, or unimportant.
- Causal claims require separate study-design and risk-of-bias appraisal. SG-FLW alone does not prove effectiveness.
- A planned protocol can receive a target profile only for methods it explicitly commits to implementing. Design intentions, desired outcomes, or uninstrumented mechanics do not receive evidence credit.
Output Format
Use this structure unless the user asks for another format:
## SG-FLW Classification
Source:
[citation or bibliographic details]
Scope:
[one paragraph on whether the intervention directly targets food waste, food systems, sustainability behavior, or a boundary case]
Evidence Ledger:
| Dimension | Reported or planned evidence | Unit / instrument / timing | Boundary issue | Score |
|---|---|---|---|---|
| B | ... | ... | ... | Bx |
| K | ... | ... | ... | Ky |
| D | ... | ... | ... | Dz |
| E | ... | ... | ... | Ew |
| L | ... | ... | ... | Lv |
Profile: Bx / Ky / Dz / Ew / Lv
Measurement-chain certainty for food-waste-reduction evidence:
[Very Low, Low, Moderate, or High]
### Dimension Rationale
- Bx: [baseline evidence and limits]
- Ky: [mechanism evidence and limits]
- Dz: [behavior/waste evidence and limits]
- Ew: [environmental conversion evidence and limits]
- Lv: [follow-up/persistence evidence and limits]
### Commensurability Check
[baseline/outcome unit, aggregation, population/site, time window, and food-waste scope alignment]
### Calculable Indices
[WRR/BII/EII/PI eligibility and any values if the source reports enough data]
### Claims To Avoid
- [unsupported effectiveness, behavior-change, or environmental claims]
### What Would Strengthen The Study
- [specific measurement upgrades, not generic "more research is needed"]
Final Self-Audit
Before responding, check:
- Did I score only evidence that is reported or explicitly planned?
- Did I avoid turning K into D?
- Did I avoid turning engagement, telemetry volume, or in-game actions into real-world behavior?
- Did I avoid turning modeled game environmental indicators into E?
- Did I separate measurement-chain certainty from causal effectiveness?
- Did I state what evidence would change the score?
