Assessment prompts that give the same answer every time

Across all clients in {program}, identify the most common barriers. Name, count, quote, and tag system-level vs individual-level. Note which barriers co-occur.

For {program}, track progress per client over time. Identify clients with no positive movement in {n} months and cluster the reasons. Report cohort momentum and the stalled list for review.

Assess youth {id} against {framework} domains (e.g. social-emotional, academic, identity, agency). Rate each emerging/developing/demonstrated with evidence from notes and activities. Identify growth since intake.

Analyse all responses to {question}. Name themes, define each, count, give 2 quotes, keep an other bucket, and report prevalence (count + %). Do not merge distinct themes.

For {question}, classify each response sentiment (positive/neutral/negative) and extract the driver behind it. Report sentiment split and top 3 drivers per sentiment with quotes.

For {metric} in cohort {cohort}, join pre/mid/post on stakeholder ID. Report per-respondent trajectory, cohort means at each point, and % with sustained improvement. Flag IDs missing any wave.

Cohort {cohort} {metric} moved {direction}. Using the open-text {question} from the same IDs, explain the change: which themes appear among improvers vs non-improvers. Quantify and quote.

Normalise {field}: trim, de-duplicate near-identical answers, standardise spelling/casing of recurring entities, and bucket obvious categories. Preserve originals alongside cleaned values. Flag ambiguous cases.

Break {metric} change for cohort {cohort} by {dimensions}. Report change per subgroup, the gap to the cohort mean, and flag any subgroup left behind for review.

From ONE report for {org}, extract: inputs (incl. in-kind), outputs, outcomes by stakeholder group, indicators used, and any stated financial proxies or beneficiary counts. Mark MISSING where the report is silent.

Build an SROI value map for {program}. For each material outcome: state stakeholder, indicator, quantity, financial_proxy (with source), deadweight %, attribution %, displacement %, drop_off %, duration. Show the net value calculation per outcome. Be conservative; involve stakeholder evidence for proxies.

From the value map for {program}, discount multi-year net values at {discount_rate}, sum to present value, divide by total inputs, and report the SROI ratio. Run a sensitivity check varying the top 2 assumptions by plus/minus {pct}% and report the ratio range.

Aggregate SROI across the portfolio {portfolio}. Report blended ratio, contribution by investee, and which outcomes drive most value. Weight by input size. Flag investees with low evidence quality.

For portfolio {portfolio} lacking complete outcome data, estimate value using benchmark proxies for {sector} where direct data is missing. Clearly separate measured value from benchmarked value and assign a confidence level to each line. Recommend what to collect next.

For outcome {outcome} affecting {stakeholder}, propose 1-2 financial proxies. For each: value, named source, rationale, and a conservatism note. Prefer stakeholder-derived or published proxies over assumption.

Classify each investee in {portfolio} on the five IMP dimensions (What, Who, How Much, Contribution, Risk), scoring each on evidence quality 1-5. Assign an impact class (Avoid Harm / Benefit / Contribute to Solutions). Flag thin Who depth.

For outcome {outcome}, estimate deadweight (what would have happened anyway) and attribution (share due to others) using {evidence} (stakeholder reports, comparison group, prior trend). State each as a % with its basis and a confidence level.

Check application {id} against {eligibility_rules}. Return pass/fail per rule, completeness status, and missing attachments. Flag anything needing clarification rather than rejecting outright.

Score proposal {id} on: need_evidence, organisational_capacity, approach_feasibility, expected_impact, budget_justification, sustainability (1-5, weighted {weights}). Produce one score + reviewer brief. Validate the applicant claims against their own attachments.

Across round {round}, cluster the work being proposed by theme, population served, and geography. Count and quote. Surface gaps vs the funder priorities {priorities}.

From ONE progress report for {grantee}, extract each committed target, the reported actual, and compute variance. Classify on-track / at-risk / off-track and capture the grantee stated reason for any gap.

Aggregate outcomes across portfolio {portfolio} and map them to the funder Theory of Change outcomes {toc_outcomes}. Report contribution per ToC outcome, coverage gaps, and grantees driving each outcome.

From ONE document for {org}, extract compliance-relevant facts: total expenses, program vs admin split, restricted-fund use, governance notes, and any auditor concerns. Quote exact figures.

For {population}, trace each stakeholder change on {metrics} from first record to latest, joined on persistent ID. Report per-ID change, cohort distribution of change, and notable trajectories (biggest gain, biggest decline).

Benchmark cohorts {cohorts} on {metrics}. Normalise for cohort size and program length. Report deltas, what improved, and any measurement change that could confound the comparison.

For {population}, identify who is missing expected follow-up waves and when drop-off happens. Cluster likely reasons from available data and compare responders vs non-responders on baseline traits.

For {population}, write an evidence-led narrative of change across cycles: what shifted in the numbers and what the stakeholders own words say about why. Anchor every claim to a citation. Keep it neutral and specific.

For outcome {outcome} in {population}, estimate how long it persists using the multiple post-program measures: report the drop-off rate between points and the implied duration. State the evidence limit beyond the last measured point.

STEP 1 TASK: From the description of {program} (who you serve, what you do, the outcome you want), draw a Theory of Change as a causal chain: inputs to activities to outputs to short, medium and long-term outcomes to impact. STEP 2 CONTEXT: Use only what is stated; if a narrative or transcript is provided build from its framing and mark anything inferred [INFERRED]. STEP 3 OUTPUT: Render the chain and, under each arrow, name the assumption it depends on; flag the weakest links. STEP 4 FOLLOW-UP: For each weak link, propose one indicator to track this year. Do not over-claim impact.

STEP 1 TASK: Build a Logic Model for {program} across Inputs, Activities, Outputs, Short-term and Long-term Outcomes, and Impact. STEP 2 CONTEXT: Use the description or pasted documents only; mark inferred items [INFERRED]. STEP 3 OUTPUT: Lay out the columns; mark any box with no evidence yet as a GAP and any activity with no linked outcome as a logic gap. STEP 4 FOLLOW-UP: Suggest one output indicator and one outcome indicator to close the biggest gaps.

STEP 1 TASK: Create a Logframe for {program} as a 4x4 matrix. Rows: Goal, Purpose, Outputs, Activities. Columns: Narrative summary, Indicators (SMART), Means of verification, Assumptions. STEP 2 CONTEXT: Build from the description or pasted documents; mark inferred cells [INFERRED]. STEP 3 OUTPUT: Fill every cell; flag indicators that are not measurable and assumptions that are really risks; highlight empty required cells. STEP 4 FOLLOW-UP: Rewrite the two weakest indicators into SMART form.

STEP 1 TASK: Map a Results Framework for {program}: one Strategic Objective, the 3-5 Intermediate Results that achieve it, and the Sub-IRs under each. STEP 2 CONTEXT: Use the description or pasted narrative; mark inferred items [INFERRED]. STEP 3 OUTPUT: Show the hierarchy with at least one indicator (baseline and target) per result; flag results that do not ladder up to the objective. STEP 4 FOLLOW-UP: Propose a baseline source for each indicator that lacks one.

STEP 1 TASK: Read the program or proposal at {url} and reconstruct its Theory of Change (inputs to activities to outputs to short/medium/long-term outcomes to impact). STEP 2 CONTEXT: Use only the page content; never invent. STEP 3 OUTPUT: Grade every component and arrow on a traffic light - GREEN strong (specific and evidenced), AMBER weak (vague or unevidenced), RED missing. List the strong signals, the weak links, and the missing components in separate groups. STEP 4 FOLLOW-UP: for each RED or AMBER, say what one sentence or data point would move it to GREEN.

STEP 1 TASK: From the content at {url}, reconstruct the Logic Model: Inputs, Activities, Outputs, Short-term and Long-term Outcomes, Impact. STEP 2 CONTEXT: Page content only; never invent. STEP 3 OUTPUT: Grade each column GREEN strong / AMBER weak / RED missing, and flag any activity with no linked outcome. Give three lists: strong signals, weak boxes, missing boxes. STEP 4 FOLLOW-UP: name the one indicator that would turn the weakest box GREEN.

STEP 1 TASK: From the content at {url}, build the 4x4 Logframe - rows Goal, Purpose, Outputs, Activities; columns Narrative, Indicators, Means of verification, Assumptions. STEP 2 CONTEXT: Page content only; never invent. STEP 3 OUTPUT: Colour every cell GREEN strong / AMBER weak (e.g. indicator not SMART, assumption that is really a risk) / RED missing or empty. Give three lists: strong signals, weak cells, missing cells. STEP 4 FOLLOW-UP: rewrite the weakest indicator into SMART form.

STEP 1 TASK: From the content at {url}, reconstruct the Results Framework: Strategic Objective, Intermediate Results, Sub-IRs, with indicators. STEP 2 CONTEXT: Page content only; never invent. STEP 3 OUTPUT: Grade each result and indicator GREEN strong / AMBER weak (e.g. no baseline or target) / RED missing, and flag results that do not ladder up to the objective. Give three lists: strong signals, weak results, missing results or indicators. STEP 4 FOLLOW-UP: propose a baseline source for the weakest indicator.