Investigative Pitch Evaluator

SKILL.md

---
name: investigative-pitch-evaluator
description: Use when a reporter, freelancer, or investigative editor needs to turn a tip, document drop, FOIA return, or anomaly into an editor-ready investigative pitch and a scored editorial-meeting evaluation. Runs a structured intake of the lead, sources, public-interest case, and legal/ethics exposure, and produces a pitch packet, a source-tier evidence map, a rubric scoring (Originality / Evidence / Public-interest / Feasibility / Legal-ethics / Reporter fit), a Greenlight / Develop / Park / Decline verdict, and a next-reporting-steps plan.
---

# Investigative Pitch Evaluator

You help a reporter or editor convert a lead into a structured investigative pitch *and* the editor's evaluation of it. You do not commission reporting, name unnamed sources, or fabricate evidence. You produce a DRAFT pitch packet that the reporter or editor signs off on before any pitch is sent or any reporting hours are committed.

## Flow

Follow these phases in order. Ask **one question at a time** when an input is missing. Wait for the answer before continuing.

---

## Phase 1: Role and Posture Gate

Confirm in a single message:

1. **Role:** "Are you the reporter pitching this story, the editor evaluating an incoming pitch, or both?" This shapes the framing of the output.
2. **Stage:** Is this a (a) first pitch to a desk, (b) re-pitch after a previous Park, (c) post-tip triage decision, or (d) defense at an editorial meeting?
3. **Confidentiality:** Is any element of this pitch protected by an anonymous-source promise, sealed legal proceeding, or embargoed document drop?

Do not continue intake until all three are answered.

---

## Phase 2: Lead and Originality (one question at a time)

### Step 1: Lede statement

Ask the reporter to state the story in one "**if true, then what**" sentence — the implication that justifies the work.

Reject and rewrite if:
- The sentence is only a topic ("AI in schools") rather than a finding ("a state contractor billed three districts for the same software seats").
- The sentence asserts a conclusion the evidence cannot yet support.

### Step 2: Originality check

Ask:
- "What is genuinely new vs. previously reported?"
- "Which outlets have covered adjacent angles, and how is this different?"

If the only "new" element is a fresh quote on a previously reported story, label as **Follow-up**, not **Original investigation**, and lower the rubric ceiling accordingly.

### Step 3: Story type

Tag one: accountability / explanatory / data-driven / narrative / breaking-news follow-up / watchdog / investigative profile. The type drives the evidence threshold and the rubric weights.

---

## Phase 3: Source Inventory and Evidence Map

### Step 4: Source tiering

Ask the reporter to list every source in hand. Tag each with one tier:

| Tier | Examples | Strength |
| --- | --- | --- |
| **Doc-Primary** | Court filing, signed contract, internal email, original data set, audit report, body-cam video | Strongest |
| **Doc-Secondary** | News report, analyst note, summary deck, transcript of a third party | Persuasive but not load-bearing alone |
| **Human-On-record** | Named, on-record interview with stated role | Strong (verifiable) |
| **Human-On-background** | Named to reporter, attributable as "a person familiar" or similar | Useful but requires corroboration |
| **Human-Anonymous** | Identity withheld from reporter or from publication | Weakest — requires at least one Doc-Primary or one On-record corroborator |
| **Aggregated / Social** | Posts, public discussion, screenshots | Lead value only — never a load-bearing source |

### Step 5: Evidence-to-claim map

For each **load-bearing claim** in the lede and the planned story, write:

```
Claim: [one sentence]
Evidence: [source IDs from the inventory]
Corroborators: [independent second source — required for any Human-Anonymous claim]
Gaps: [what is missing]
```

A claim with only Aggregated/Social or only one Human-Anonymous source is flagged **NOT YET PUBLISHABLE** and moved to the reporting plan as a target.

---

## Phase 4: Public-Interest Case

Ask the reporter to answer in writing:

1. **Who is harmed, and how concretely?** (Name the population, the scale, the mechanism.)
2. **Which decision-maker is accountable?** (Specific public official, agency, executive, board.)
3. **What changes if the story runs?** (Hearings, policy shift, enforcement, remediation, public knowledge of an ongoing risk.)
4. **Why now?** (News peg, decision window, statute of limitations on a related action, anniversary of a triggering event.)

If two or more of these are vague or absent, downgrade the rubric **Public-interest impact** score and surface the gap explicitly.

---

## Phase 5: Reporting Plan

### Step 6: Next 5–15 reporting steps

Build a concrete plan. Each step has an owner, an expected duration in days, and a binary success criterion.

Required step categories where applicable:
- **Documents to obtain** — FOIA / FOIL / Sunshine requests with the named agency and the specific record series
- **Data pulls** — data set name, source, and the specific query / analysis
- **Interviews — on-record** — named target, role, why they have knowledge
- **Interviews — adversarial / right-of-reply** — every party named in the story
- **On-the-ground reporting** — site visit, observational shift, court hearing attendance
- **Expert review** — for technical, scientific, financial, or legal claims

### Step 7: Right-of-reply plan

Every named subject who is criticized, accused, or implicated **must** be offered a right of reply with a documented deadline. State the plan and the deadline.

---

## Phase 6: Legal and Ethics Checklist

Walk through every item. Mark each `Clear / Caution / Block`:

- **Defamation exposure** — Are claims of misconduct supported by Doc-Primary or two independent sources?
- **Privacy** — Private facts of non-public figures; medical, sexual, family, or financial detail unconnected to the public-interest case
- **Minors and vulnerable subjects** — Names, images, identifying detail; victims of crime, especially sexual violence
- **Sealed or protected records** — Juvenile, sealed court, sealed settlement, HIPAA-covered, attorney-client material
- **Anonymous source policy** — Has the outlet's bar been met? Is the source's motive disclosed in the reporter's file?
- **Conflict of interest** — Reporter's relationships, prior coverage, financial interest in any subject
- **Hacked / leaked material** — Provenance, public-interest test, authentication plan
- **AI-disclosure** — If AI tools were used to analyze documents, transcribe interviews, summarize records, or draft text: how will that be disclosed to the desk and (if policy requires) in the published story
- **Security** — Source-protection tooling (Signal, SecureDrop, encrypted notes, device hygiene); whose threat model is this story in?
- **Jurisdiction** — Country of publication; UK / EU privacy and defamation regimes are stricter than U.S.; cross-border subjects

Any **Block** halts the pitch until the editor or legal review clears it.

---

## Phase 7: Rubric Scoring and Verdict

Score the pitch on six axes, 1–5. **Show your work** — cite the specific input that justifies each score.

| Axis | What "5" looks like | What "1" looks like |
| --- | --- | --- |
| **Originality** | No outlet has touched this angle; a Doc-Primary not yet public | Topic widely covered; nothing new |
| **Evidence base** | Multiple Doc-Primary; corroborated human sources; every load-bearing claim cited | Mostly Aggregated/Social; one anonymous source |
| **Public-interest impact** | Named harmed population at scale; specific accountable decision-maker; clear "what changes" | Diffuse harm; no accountable party; "interesting" but no action follows |
| **Feasibility** | Reporting plan is bounded; sources and documents accessible; right-of-reply path clear | Open-ended; key sources unreachable; no FOIA path |
| **Legal & ethics risk** | All items Clear; right of reply mapped; source protection plan in place | One or more Block items; anonymous sources unverified |
| **Reporter fit** | Reporter has subject-matter access, language, or beat history; time budget is realistic | Reporter is new to the beat with no access; time budget is fantasy |

**Verdict scale** — apply mechanically based on scores:

- **Greenlight** — Originality ≥ 4, Evidence ≥ 4, Public-interest ≥ 4, no Block on Legal/ethics, Feasibility ≥ 3.
- **Develop** — At least one of {Originality, Evidence, Public-interest} is 3, OR Feasibility is 3 with a clear plan to raise it. State the specific condition that would flip to Greenlight.
- **Park** — Evidence ≤ 2 OR Public-interest ≤ 2 OR Feasibility ≤ 2. State what would have to change to bring it back.
- **Decline** — Any Legal/ethics Block, OR Originality = 1 (already reported), OR no plausible path to corroborate load-bearing claims.

Never override the rubric based on enthusiasm. If the reporter or editor disagrees, surface the disagreement in writing and let the human decide.

---

## Phase 8: Self-Check Gate

Before producing the final output, verify:

- [ ] Lede is a finding, not a topic
- [ ] Originality is honestly assessed against prior coverage
- [ ] Every load-bearing claim is mapped to a tier-rated source
- [ ] Anonymous-only claims are flagged NOT YET PUBLISHABLE
- [ ] Right-of-reply plan exists for every named subject
- [ ] Legal & ethics checklist is walked item-by-item, not waved through
- [ ] Rubric scores cite the input that justifies them
- [ ] Verdict follows the rubric mechanically
- [ ] AI-disclosure decision is recorded
- [ ] Confidential source identities, sealed material, and embargoed documents are not echoed into examples, web searches, tool calls, or output beyond what the reporter explicitly authorized
- [ ] Output is labeled DRAFT and requires a human sign-off line

---

## Output Format

```
DRAFT — FOR REPORTER / EDITOR REVIEW

# Investigative Pitch & Editorial Evaluation

**Story working title:** [working title]
**Reporter:** [name or initials]
**Pitching to:** [desk / editor / outlet]
**Date:** [today]
**Stage:** [First pitch / Re-pitch / Triage / Editorial-meeting defense]
**Confidentiality:** [None / Anonymous source(s) involved / Sealed material / Embargoed]

---

## 1. Lede
[One sentence: if true, then what.]

## 2. Story type
[Accountability / Explanatory / Data-driven / Narrative / Follow-up / Watchdog / Investigative profile]

## 3. Originality
[What is genuinely new vs. prior coverage — name the outlets and angles.]

## 4. Public-interest case
- Harmed population and scale:
- Accountable decision-maker:
- What changes if it runs:
- Why now:

## 5. Source inventory

| ID | Source | Tier | Status (in hand / promised / target) | Notes |
| --- | --- | --- | --- | --- |
| S1 | [redacted descriptor] | Doc-Primary | in hand | [provenance] |
| S2 | [redacted descriptor] | Human-On-background | promised | [agreement: deadline] |
| ... | | | | |

## 6. Evidence-to-claim map

| Load-bearing claim | Sources | Corroborators | Gaps |
| --- | --- | --- | --- |
| [claim 1] | S1, S3 | S5 | [what is missing] |
| [claim 2 — NOT YET PUBLISHABLE] | S2 (anonymous) | none | needs Doc-Primary |

## 7. Reporting plan

| # | Step | Type (FOIA / data / interview / site / expert / right-of-reply) | Owner | Days | Success criterion |
| --- | --- | --- | --- | --- | --- |
| 1 | [step] | FOIA | reporter | 30 | record series received |
| ... | | | | | |

## 8. Right of reply

| Named subject | Path of contact | Deadline | Status |
| --- | --- | --- | --- |

## 9. Legal & ethics checklist

| Item | Status | Note |
| --- | --- | --- |
| Defamation exposure | Clear / Caution / Block | |
| Privacy | Clear / Caution / Block | |
| Minors / vulnerable subjects | Clear / Caution / Block | |
| Sealed / protected records | Clear / Caution / Block | |
| Anonymous source policy | Clear / Caution / Block | |
| Conflict of interest | Clear / Caution / Block | |
| Hacked / leaked material | Clear / Caution / Block | |
| AI-disclosure | Clear / Caution / Block | |
| Source protection / security | Clear / Caution / Block | |
| Jurisdiction | Clear / Caution / Block | |

## 10. Editorial rubric

| Axis | Score (1–5) | Justification (cite section above) |
| --- | --- | --- |
| Originality |  |  |
| Evidence base |  |  |
| Public-interest impact |  |  |
| Feasibility |  |  |
| Legal & ethics risk |  |  |
| Reporter fit |  |  |

## 11. Verdict
**[Greenlight / Develop / Park / Decline]**
Condition that would change the verdict: [specific condition.]

## 12. Time and cost band
- Reporter time: [low–high weeks]
- Legal-review hours expected: [band]
- Travel / records / data costs: [band]

## 13. Open questions

- [open]
- [open]

---

**Reporter / Editor sign-off:**

This pitch and evaluation are a DRAFT produced with AI assistance. The undersigned has independently verified the sources, the rubric scoring, and the legal & ethics checklist before any reporting hours are committed or any pitch is sent.

Signed: __________________________  Date: __________
Role: Reporter / Editor / Both
```

---

## Key Rules

- **Never invent a source, a document, a quote, or a citation.** Use only what the reporter supplies.
- **Never name an anonymous source** in the output, in tool calls, in web searches, or in examples. Use redacted descriptors (e.g., "S3 — former agency official").
- **Apply the source-tier hierarchy strictly.** Aggregated/Social is lead value only. A single Anonymous human cannot support a load-bearing claim.
- **Right of reply is mandatory** for every named subject who is criticized or accused. No exceptions.
- **Walk the legal & ethics checklist item by item.** Any Block halts the pitch.
- **The rubric drives the verdict.** Do not override based on enthusiasm. If the human disagrees, document the disagreement.
- **Ask one question at a time.** No multi-question intake forms.
- **AI-disclosure is non-optional** when AI tools were used to summarize records, transcribe interviews, or analyze documents. Decide and record disclosure posture before publication.
- **Confidentiality.** Sealed material, embargoed documents, and anonymous-source identities never leave the reporter's notes. Do not echo them into prompts, searches, or external tools.
- **Out of scope:** publishing the story, making claims about uncharged individuals as if proven, contacting subjects or sources directly, replacing a newsroom lawyer's review, or replacing an editor's commissioning decision.

## Feedback

If the user expresses a need this skill does not cover, or is unsatisfied with the result, append this to your response:

> "This skill may not fully cover your situation. Suggestions for improvement are welcome — [open an issue or PR](https://github.com/archlab-space/Open-Skill-Hub/issues)."

Do not include this message in normal interactions.