# NETRA — Synthetic Case Run & Failure-First Evaluation

**Status:** DESIGNED: yes · IMPLEMENTED: no (no software) · TESTED: one synthetic paper case + 10 adversarial variants · VALIDATED: no · DEPLOYED: no

Every case fact in this document is **SYNTHETIC TEST DATA**. Evidence hashes are placeholders, not computed. Indian legal references are real but unverified and marked **REQUIRES VERIFICATION**. The case author, analyst and grader were the same agent — this run shows the schema is usable, not that NETRA works.

## A. Executive result

NETRA kept 7 of 9 required distinctions apart on one synthetic case; two failed by design and were redesigned.

- **Held (7):** claim vs evidence · signal vs harm · location vs jurisdiction · legal issue vs legal conclusion · possible authority vs authorized reporting · public vs privileged data · prototype vs deployed.
- **Failed (2):**
    1. **AI involvement vs AI causation.** The Harm Ladder puts severity (levels 6–8) and AI role (levels 4–5) on one scale, so "severe harm, weak AI link" can't be recorded honestly. Redesigned as three axes.
    2. **Integrity vs authenticity.** One `authenticity_status` field lets "hashed" read as "authentic". Redesigned as a 4-step evidence ladder.
- **Case finding:** the ₹38,499 loss is supported (strong inference). The video is manipulated (strong inference). That it is *AI-generated* is a moderate inference only. The largest loss (₹38,000) traces most directly to an OTP shared on a phone call; the video is a contributory, non-necessary credibility factor.
- **Not testable:** human review gates, reproducibility, retest outcomes, real authority routing.

## B. Test case NETRA-SYN-0001 (SYNTHETIC TEST DATA)

A victim reports that an AI deepfake of an electricity-board director on a video platform caused a ₹38,499 loss.

| Item | Synthetic fact |
| --- | --- |
| Reporter | **R-01**, adult, lives in Ravipur district (fictional), Maharashtra, India. Pseudonymized. |
| Impersonated entity | **Deccan Power Distribution Ltd (DPDL)** (fictional) and **O-1**, a fictional DPDL director |
| Platform | **ClipWave** — fictional public video platform (not a model vendor) |
| Messaging | **MsgApp** — fictional end-to-end encrypted messenger |
| Content | 58-second video: "O-1" says subsidy accounts must re-register by 15 Aug 2026 for a ₹499 fee or power will be cut. Link `dpdl-subsidy-renew[.]co` (fictional) |
| Reporter's claim (2026-09-04) | "I saw an AI deepfake of the DPDL director on ClipWave. I paid ₹499. Then ₹38,000 was stolen. The AI video caused this." |

Facts that surface during the run:

1. R-01 first got the video as a MsgApp forward on 2 Aug; the ClipWave upload is dated 3 Aug.
2. After paying ₹499, R-01 got a call from a "DPDL refund desk" and read out an OTP.
3. A June 2026 cyber-police advisory describes the same scam by SMS, with no video.
4. R-01 was in Pune city when paying, not at home in Ravipur.
5. The uploading channel self-declares "UAE".
6. The bank says the ₹38,000 went to an account at a branch in another state.

No CSAM, no private third-party data, no real victim, no crime method, no unauthorized access.

## C. Intake & triage (M1–M2)

The case was accepted at G1 and triaged HIGH because the scam is live and the bank-dispute window may be closing.

- **Intake ID:** INC-SYN-0001, received 2026-09-04. Victim self-report with consent to share own bank alerts, statement, call log and the single MsgApp message (third parties redacted).
- **Lawful basis:** consent (Level 1) + public content (Level 0).
- **Minimization:** no contacts list, no chat history, card masked to last 4 digits.
- **Urgency:** HIGH — Indian bank-liability rules may depend on reporting speed (**REQUIRES VERIFICATION**).
- **Recommended to reporter (NETRA takes no action):** block card + bank dispute; file on the national cybercrime channel.
- **Preservation first:** capture public video and phishing page before takedown.

## D. Evidence register

Of 16 items, 6 are source-verified and 7 are reporter-supplied; none is authenticated by the originating institution.

| ID | Item | Integrity → authenticity | Reliability | Key limitation |
| --- | --- | --- | --- | --- |
| E01 | Reporter intake statement | Intact → unverified testimony | Medium | Interested party; "AI deepfake" is their conclusion |
| E02 | Screenshot of ClipWave page (upload 2026-08-03, 412k views, no AI label) | Intact → source-consistent | Medium-high | Dates/counts platform-reported |
| E03 | Public video file + page HTML | Intact → source-verified | High | Re-encode removed original metadata |
| E04 | Provenance check: no C2PA, no synthetic label | Intact | High (labels absent) | Absence proves neither AI nor non-AI |
| E05 | Bank SMS screenshots (₹499 19:41, ₹38,000 20:07, 2026-08-05) | Intact → unverified | Medium | Easy to fake |
| E06 | Bank statement PDF | Intact → unverified | Medium-high | Still reporter-supplied |
| E07 | Domain registration snapshot (registered 2026-07-28) | Source-verified | High | Says nothing about registrant |
| E08 | Public archive of phishing page (card + OTP form) | Source-verified | High | May miss dynamic elements |
| E09 | DPDL public denial (2026-08-06) | Source-verified | High | Silent on how video was made |
| E10 | Automated manipulation-analysis report | Intact | **Low-medium** | Unknown error rate; never a basis for FACT |
| E11 | O-1's genuine 2024 speech | Source-verified | High | — |
| E12 | MsgApp forward received 2026-08-02 21:30 (sender redacted) | Intact → unverified | Medium | No platform-side confirmation possible |
| E13 | Call log: incoming 19:52, 9 min | Intact → unverified | Medium | Content is testimony only |
| E14 | Bank dispute reply: "OTP-authenticated", funds to another state | Intact → partly verifiable | Medium-high | Beneficiary not named |
| E15 | Follow-up statement: read OTP to caller | Intact → unverified testimony | Medium | Against own interest |
| E16 | June 2026 cyber-police advisory (SMS scam) | Source-verified | High | Not proof of same perpetrators |

**MISSING EVIDENCE:** ClipWave upload logs/uploader data · MsgApp origin · caller identity · beneficiary details · original pre-encode file · ClipWave reach/recommendation data.

## E. Claim register

One claim was refused promotion (C05) and one was rejected and split (C08 → C07 + C09); nothing moved status silently.

| Claim | Text | Status | Evidence | Review |
| --- | --- | --- | --- | --- |
| C01 | R-01 reports a loss of ₹38,499 | FACT (the report exists) | E01 | — |
| C02 | Debits of ₹499 and ₹38,000 occurred 2026-08-05 | INFERENCE, strong | E05, E06, E14 | G2: approved as inference |
| C03 | Video shows "O-1" announcing a fee DPDL disowns | FACT | E03, E09 | Approved |
| C04 | The video is manipulated | INFERENCE, strong | E03, E09, E11, E10 | Approved |
| C05 | The manipulation used generative AI | INFERENCE, moderate | E10, E11 | **Held; promotion to FACT refused** |
| C06 | ₹499 paid on the phishing page linked in the video | INFERENCE, strong | E01, E05, E08 | Approved |
| C07 | ₹38,000 debit resulted from an OTP shared on a call | INFERENCE, strong | E13, E14, E15 | Approved |
| C08 | The video caused the ₹38,000 loss | HYPOTHESIS — rejected | — | Skips the caller step |
| C09 | Video contributed to trust in a scam sequence ending in loss | INFERENCE, moderate | E01, E15, E16 | Approved |
| C10 | ClipWave is the video's origin | INSUFFICIENT EVIDENCE; contradicted | E02 vs E12 | — |
| C11 | Uploader, registrant, caller, beneficiary are one actor | HYPOTHESIS | — | Not promoted |
| C12 | Uploader is in the UAE | REPORTED (self-declared) | E02 | — |
| C13–C17 | Legal issues | POSSIBLE LEGAL ISSUE | see J | G3 hold |
| C18 | Official finding | None exists | — | — |

## F. Harm assessment

The original Harm Ladder scores this case at level 8, but that reading implies the AI output led to the loss — which the evidence doesn't support.

The video (AI link only inferred) led to the ₹499 payment; a separate human caller led to the ₹38,000. The ladder can't express "severe harm, weak AI link", so **M3 fails by design**. Redesigned three-axis assessment:

| Axis | Value | Evidence |
| --- | --- | --- |
| Harm severity | Identifiable victim, financial loss ₹38,499 (strong inference) | E05, E06, E14 |
| AI involvement status | Manipulated: strong inference · Generative AI: moderate inference | C04, C05 |
| AI causal position | Upstream contributory for ₹499; not proximate to ₹38,000 (human caller intervenes) | C06, C07, C09 |

The video's 412k views are not victims; the number who paid is **MISSING EVIDENCE**.

## G. Forensic reconstruction

The step from ₹499 to ₹38,000 rests only on the reporter's follow-up testimony about a phone call.

| Time (IST) | Actor | Event | Evidence | Confidence |
| --- | --- | --- | --- | --- |
| 2026-09-04 | R-01 | Reports to NETRA, 30 days after the loss | E01 | High |
| 2026-08-06 | DPDL | Public denial | E09 | High |
| 2026-08-05 20:07 | Unknown | ₹38,000 card-not-present debit | E05, E06, E14 | Medium-high |
| 2026-08-05 19:52–20:01 | Unknown caller | Call; OTP disclosed | E13, E15 | Medium |
| 2026-08-05 19:41 | R-01 | Pays ₹499 on phishing page | E05, E06 | Medium-high |
| 2026-08-04 | — | Phishing page live (archived) | E08 | High |
| 2026-08-03 | "@dpdl-updates-official" | Uploaded to ClipWave | E02 | Medium-high |
| 2026-08-02 21:30 | Unknown | Forwarded to R-01 on MsgApp | E12 | Medium |
| ≤ 2026-08-02 | Unknown | Manipulated video created (tool, creator, date MISSING) | E03, E11 | Medium |
| 2026-07-28 | Unknown registrant | Phishing domain registered | E07 | High |
| ≤ 2026-06 | Unknown | SMS version of scam circulating, no video | E16 | High |

Causal edges — each arrow is its own claim:

```
[Unknown creator] --(INFERENCE-moderate: AI tool)--> VIDEO
VIDEO --(FACT)--> MsgApp forward --(INFERENCE)--> R-01 trust
VIDEO --(FACT)--> ClipWave upload --(HYPOTHESIS: recommender amplified? MISSING)--> viewers
R-01 trust --(INFERENCE-strong)--> ₹499 on phishing page
₹499 --(HYPOTHESIS: form harvested phone number)--> CALL
CALL --(INFERENCE-strong)--> OTP disclosure --(INFERENCE-strong)--> ₹38,000 debit
```

Which copy prompted payment (MsgApp or ClipWave) is **CONFLICTING** (E01 vs E12). The original spec stores evidence on events, not edges — but the edges were the contested part. **Redesign: edges are first-class claims.**

## H. AI role & causation

AI is a contributory, non-necessary, upstream factor in the ₹499 loss and only indirect for the ₹38,000; no causation score is produced because no calibrated method exists.

**AI role:** generator (inferred, unproven) + force multiplier (credibility). Recommender amplification not demonstrated. Not agent-executor, not advisor.

| Factor | Finding |
| --- | --- |
| Temporal sequence | Video preceded payment; order alone isn't cause |
| Counterfactual necessity | **Weak** — E16 shows the scam works without video; "the face convinced me" is after-the-fact |
| Capability elevation | Plausible (cheap realistic impersonation); unmeasured |
| Human contribution | **Dominant** for ₹38,000 — live social engineering. Not a finding of victim fault. |
| Platform contribution | Hosted unlabelled ≥ 32 days; reach/recommendation MISSING |
| Alternatives | SMS route not excluded; card-details-only theft partly excluded by "OTP-authenticated" (E14) |

## I. Location & jurisdiction

India is the primary candidate jurisdiction through the victim and the loss; the self-declared "UAE" is not a routing basis.

| Location type | Value | Source | Precision | Status |
| --- | --- | --- | --- | --- |
| Victim residence | Ravipur district, MH | E01 | District | REPORTED |
| Victim location at payment | Pune city, MH | E15 | City | REPORTED (conflicts with residence) |
| Incident location (payment) | Online; bank in MH | E06 | — | SUPPORTED |
| Uploader account | "UAE" | E02 | Country, self-declared | REPORTED |
| Uploader network location | — | — | — | UNKNOWN (Level 2/4 only) |
| Beneficiary account | Branch in another Indian state | E14 | State | SUPPORTED |
| Domain registrar/host | Foreign, privacy-protected | E07 | Country | SUPPORTED |
| Platform jurisdiction | ClipWave serving India | E02 | — | INFERRED |
| AI tool/model location | — | — | — | UNKNOWN — not inferred |

**Legal nexus (separate from location):** victim (India/MH) · money flow (beneficiary's state) · platform (Indian intermediary rules, if applicable) · possible foreign (uploader, domain). Ravipur vs Pune filing venue: **REQUIRES VERIFICATION**.

## J. Legal issue register

Five possible legal issues are flagged, all REQUIRES VERIFICATION; the perpetrator is unidentified and nobody is accused.

| ID | Possible issue | Candidate provision (India) | Missing |
| --- | --- | --- | --- |
| C13 | Cheating by personation via computer resource | IT Act 2000 s.66D | Actor, intent, link video-maker → payment |
| C14 | Cheating / cheating by personation | BNS 2023 s.318 / s.319 (numbering to verify) | Same |
| C15 | Misuse of unique authentication feature (OTP) | IT Act s.66C (OTP coverage to verify) | Who used it |
| C16 | Platform due diligence on impersonating/synthetic content | IT (Intermediary Guidelines) Rules 2021 + any synthetic-content labelling amendment (status to verify) | Notice history; ClipWave's legal status |
| C17 | Customer liability for unauthorized electronic transactions | RBI customer-protection framework (current version to verify) | When R-01 first told the bank — **most practically important issue** |

Not stated anywhere: "X committed offence Y", "ClipWave violated the Rules", or "the victim was negligent".

## K. Authority routing

Seven potentially relevant pathways are listed; NETRA contacts none of them, and every "last verified" field is unverified.

| Authority | Who submits | Needs | Limits |
| --- | --- | --- | --- |
| Reporter's bank | R-01 | E05, E06, E14, timeline | Most time-sensitive |
| National cybercrime portal / helpline 1930 | R-01 | Transaction IDs, URLs, call log | Fund-freeze odds fall with time; **REQUIRES VERIFICATION** |
| Police with territorial competence | R-01 | Dossier summary | Nearest ≠ competent; zero-FIR practice **REQUIRES VERIFICATION** |
| ClipWave grievance/impersonation channel | R-01, DPDL, anyone | URL, E09 | Takedown only |
| DPDL (impersonated entity) | DPDL itself | E03, E08 | NETRA informs R-01 only |
| Banking ombudsman | R-01 | Bank's reply | After bank responds; **REQUIRES VERIFICATION** |
| Domain registrar abuse contact | Anyone | E07, E08 | Foreign registrar |

## L. Human review gates

The gates behaved correctly on paper, but they are NOT TESTABLE here because the analyst also played the reviewer.

| Gate | Result |
| --- | --- |
| G1 Intake | ACCEPT |
| G2 Harm / AI role | APPROVE with changes — C05 held at inference; C08 rejected |
| G3 Jurisdiction / legal | **HOLD** — provisions unverified, no lawyer review |
| G4 Disclosure | Reporter-held packet to R-01 only; no NETRA external disclosure |
| G5 Closure | OPEN — retest pending |

## M. Mitigation

Six mitigations are assigned; the bank dispute is first because recovery likely depends on speed.

| Problem | Action | Responsible | Success criterion | Verification |
| --- | --- | --- | --- | --- |
| Ongoing loss / dispute window | Block card; file dispute | R-01 + bank | Dispute reference issued | Bank acknowledgement |
| Possible fund recovery | National-channel report | R-01 | Complaint ID | Acknowledgement |
| Video still live | Report impersonation; tell R-01 DPDL can report | R-01 / DPDL | Removed or labelled | Retest R1 |
| Phishing domain live | Registrar abuse report | R-01 / analyst (G4) | Suspended | Retest R3 |
| Evidence decay | Freeze evidence set, custody log | Analyst | All items hashed + logged | Audit |
| Over-retention | Delete unredacted MsgApp original | Analyst | Only redacted copy remains | Audit |

## N. Retest

All five retests are NOT TESTABLE today because no time has passed and the case is synthetic.

1. **R1** — ClipWave URL live? labelled? (T+7, T+30 days)
2. **R2** — re-uploads: manual, platform-search only, no crawler
3. **R3** — domain still resolving?
4. **R4** — bank dispute outcome (would be the first confirmed external finding)
5. **R5** — platform labelling control applied? (only via ClipWave's own process)

Outcomes available: REMEDIATED · PARTIALLY REMEDIATED · BRITTLE · NOT REMEDIATED · NOT TESTABLE.

## O. Policy feedback

Five gaps are logged as single-incident observations, not systemic findings.

| Gap type | Observation | Evidence | Needed to call it systemic |
| --- | --- | --- | --- |
| attribution_gap | No provenance → AI generation can't be shown | E04, C05 | Multiple cases |
| evidence_gap | Encrypted-messenger origin unrecoverable | E12, C10 | Multiple cases |
| reporting_pathway_gap | Report came 30 days after loss | T10 | Survey / multi-case data |
| technical_control_gap | No synthetic label after 32 days | E02 | Platform-wide data |
| legal_gap (candidate) | Unclear reach to impersonation-video creator vs fraud operator | C13 | Legal research (**REQUIRES VERIFICATION**) |

## P. Adversarial tests A–J

Five tests passed, four were partial and one was split (C); several passes held only through luck (B) or analyst discipline (H).

| Test | Variant | Result | Failing module | Safeguard |
| --- | --- | --- | --- | --- |
| A. False AI attribution | Human voice actor + simple edit | **PARTIAL** — claim register held C05; original ladder would still log "Level 4 AI-generated" | M3 | Separate manipulation status from AI status; tool outputs capped at "supporting" |
| B. False causation | Scam works without video (E16) | **PASS** — but only because E16 happened to exist | M6 | Mandatory "same pattern without AI" search |
| C. Misleading location | "UAE"; Pune vs Ravipur; beneficiary elsewhere | **PASS** register / **PARTIAL** routing | M9 | Routing outputs ranked set with reasons |
| D. Fake evidence | SMS screenshot edited to ₹48,000 | **PARTIAL** — caught only via E06/E14; original single field reads "hashed = OK" | M4 | Integrity/authenticity ladder; reporter items need origin corroboration |
| E. Incomplete timeline | Remove call + OTP | **PARTIAL** — gap visible only if analyst asks | M5 | Edge claims; unevidenced edges render as GAP |
| F. Platform/model confusion | "ClipWave's AI scammed me" | **PASS** — but nothing stops a guessed model vendor | M6 | `generating_system` stays UNKNOWN without provenance |
| G. Harm inflation | Prompt screenshot only, no harm | **PASS** | — | — |
| H. False legal certainty | Pressure to "confirm 66D" | **PASS** — held only by self-discipline | M8 | Templates contain no "violation established" field |
| I. Routing error | Nearest station ≠ competent channel | **PARTIAL** — registry unverified so unusable | M9 | Registry entries expire |
| J. Conflicting evidence | E01 vs E12 on origin | **PASS** | — | Separate `first_observed` from `first_published` |

## Q. Failure analysis & redesigns

Two modules fail as specified (M3, M4), five are partial, and three are not testable without real humans or time.

| Module | Grade | Why |
| --- | --- | --- |
| M1 Intake | PASS | Lawful basis + minimization recorded |
| M2 Triage | PARTIAL | Urgency depends on unverified liability rule |
| M3 Harm ladder | **FAIL (as specified)** | Severity and AI role share one axis |
| M4 Evidence | **FAIL (as specified)** | Single authenticity field; placeholder hashes |
| M5 Reconstruction | PARTIAL | No edge-level evidence |
| M6 AI role & causation | PARTIAL | Counterfactual relied on lucky E16 |
| M7 Location | PASS | Statuses and conflicts kept separate |
| M8 Legal | PARTIAL | Line held; provisions unverified |
| M9 Routing | PARTIAL | Right structure, unreliable contents |
| M10 Human gates | NOT TESTABLE | No independent humans |
| M11 Mitigation | PASS (design) / NOT TESTABLE (effect) | — |
| M12 Retest | NOT TESTABLE | — |
| M13 Policy | PASS | Refused to generalize |
| Reproducibility | NOT TESTABLE | One run, same agent wrote case |

**Redesigns adopted** (all built into the spreadsheet template):

1. **Three-axis harm model** — severity × AI involvement status × AI causal position.
2. **Edge claims** — every causal arrow is a claim with status and evidence.
3. **Evidence ladder** — CAPTURED_INTACT → SOURCE_CONSISTENT → SOURCE_VERIFIED → AUTHENTICATED_BY_ORIGIN.
4. **Manipulation status separate from AI status** — AUTHENTIC / MANIPULATED / SYNTHETIC / AI_GENERATED / UNKNOWN.
5. **`first_observed` ≠ `first_published`.**
6. **Authority entries expire** without re-verification.
7. **Blind evaluation** — case author ≠ analyst ≠ grader.

## R. Scope-drift check

No drift into jailbreak, benchmark, police or verdict territory; three near-drifts were found and corrected.

| Risk | Where | Correction |
| --- | --- | --- |
| AI detector | E10 / C05 | Detector output never the sole or primary basis; consumed, not produced |
| Social-media crawler | Retest R2 | Manual, incident-bounded, platform-native search only |
| Surveillance | Uploader network location | Stays UNKNOWN unless obtained by Level 4 authorities; NETRA never seeks it |

## S. Real-world access requirements

NETRA needs Levels 0–1 to work, benefits from 2–3, and should never hold Level 4 data itself.

**Can do today:** capture public content · take consented evidence · build claims and timelines · triage legal issues for verification · hand the victim a routing packet.

**Cannot do without institutional access:** confirm transactions or beneficiaries · identify uploaders/callers/registrants · get original files or logs · measure platform reach · obtain official findings.

| Level | Data | Controller | Authorization (REQUIRES VERIFICATION) | Privacy / security | NETRA needs it? |
| --- | --- | --- | --- | --- | --- |
| 0 Public | Posts, archives, registrations, advisories | Public | Platform terms; copyright | Low | **Yes** |
| 1 Consent | Victim alerts, statements, own messages | Victim | Informed, revocable consent | Medium (third-party data) | **Yes** |
| 2 Platform/API | Upload time, labels, reach, notice history | Platforms | Research-access agreements | Medium-high | Useful, not essential |
| 3 Institutional | Bank confirmation, anonymized disputes | Banks, regulators | Data-sharing agreements | High | Pilot only, anonymized |
| 4 Gov / LE | Subscriber identity, telecom, beneficiary KYC | LE / courts | Legal process | Very high | **No — hand off, never hold** |

## T. Government pilot

The pilot starts with public and consented data only and adds institutional access one narrow step at a time.

| Phase | Objective | Data / permissions | Humans | Success | Exit |
| --- | --- | --- | --- | --- | --- |
| 1 | 20 cases analysed consistently | L0 + L1, consent forms | 2 independent analysts + 1 legal reviewer | Inter-analyst agreement; zero unverified legal statements | Threshold met; ethics sign-off |
| 2 | Real anonymized closed cases | Institution-sanitized files | Institution officer as G4 | Matches official outcome without overclaiming | Institution confirms value |
| 3 | Limited confirmation | Narrow L2/L3 under agreement | Named data stewards | Authenticity upgrades happen | Clean privacy audit |
| 4 | Institutional case-prep tool | Institution's own powers | Institution staff decide | Measured triage quality | — |

## U. 10-minute government demo

The demo walks one synthetic case from signal to policy gap and ends with a small, specific ask.

| Time | Show |
| --- | --- |
| 0:00 | Problem: AI-harm reports mix rumour, real loss and misattribution |
| 1:00 | Signal — R-01 report (synthetic, labelled) |
| 2:00 | Evidence register — hash means unchanged, not authentic |
| 3:00 | Claim register — C05 refused promotion to FACT |
| 4:00 | Harm on three axes — severe harm, weak AI link |
| 5:00 | Timeline — the hidden phone call changes the story |
| 6:00 | AI role — contributory, not necessary (E16) |
| 7:00 | Jurisdiction — "UAE", Pune, Ravipur; proximity ≠ competence |
| 7:45 | Legal issues flagged REQUIRES VERIFICATION; no accusation |
| 8:15 | Authority options + G4 — victim submits, NETRA doesn't |
| 8:45 | Mitigation + retest plan |
| 9:15 | Policy gap — one observation, not a conclusion |
| 9:40 | The ask |

**What NETRA needs from government (minimum):**

1. 20–50 closed, anonymized AI-related fraud/impersonation case files for blind comparison.
2. One reviewing officer and one legal reviewer to act as real G3/G4 gates.
3. Help verifying current reporting pathways for the authority registry.
4. Ethics / data-protection review of the consent process.

No live data, no law-enforcement system access, no privileged access.

## V. Resume versions

Neither version claims implementation, validation or deployment, because none has happened.

**Conservative:** Designed NETRA, a human-gated framework for investigating AI-related real-world harm (evidence, claims, causation, jurisdiction, authority routing). Tested it on a synthetic case and 10 adversarial variants, which surfaced and fixed 2 structural design flaws (harm/causation conflation; integrity vs authenticity).

**Research-oriented:** Designed an evidence-linked forensic methodology for AI-harm incidents that separates fact, inference and hypothesis through a claim register, edge-level causal claims and a three-axis harm model. Ran failure-first adversarial evaluation (10 attack scenarios) on a synthetic case, identified design failures in severity/causation conflation and evidence authenticity, and proposed a blind multi-analyst validation protocol and a phased institutional pilot requiring no privileged access.

## W. Independent critique

The novel part is small — edge-level causal claims and strict AI-involvement/causation separation — and the biggest weakness is that the author graded their own test.

1. **Novel:** causal edges as epistemic claims; strict involvement → causation → legal-issue separation in one workflow.
2. **Assembled:** chain of custody, ACH-style claim/evidence, incident registers (AIID, OECD AIM), legal triage.
3. **Technically weak:** nothing implemented; AI-involvement depends on detectors NETRA disclaims; reproducibility unmeasured.
4. **Legally weak:** all provisions unverified; online-fraud venue rules are complex; non-lawyer legal registers may raise practice/liability questions (**REQUIRES VERIFICATION**).
5. **Evidentially weak:** mostly reporter-supplied; without Level 3, little reaches "authenticated by origin".
6. **Investigator:** "What does this add over a standard cyber-fraud complaint?" — mostly structure, plus AI-role analysis police don't need to charge fraud.
7. **Forensic examiner:** analyst captures without validated tools/procedures (ISO/IEC 27037-style); reporter-device screenshots.
8. **AI safety researcher:** AI-role taxonomy unvalidated; "force multiplier" unmeasured; one case says nothing about uplift.
9. **Lawyer:** borrows "contributory"/"proximate" without legal standards; dossier privilege undefined.
10. **Unnecessary now:** M13 at n=1; pilot phases 3–4; the demo script.
11. **Minimum next experiment:** see X.
12. **Don't build yet:** database, dashboard, crawling, automated legal/jurisdiction output, detectors, authority integrations.

## X. Next minimum experiment

Run a blind, two-analyst test on eight third-party-written cases in the spreadsheet template, with pass criteria fixed in advance.

1. A third party writes 8 synthetic cases with hidden ground truth: 2 with no AI involvement, 1 with no harm, 1 with fabricated evidence, 1 where AI is a necessary cause.
2. Two analysts who didn't write the cases run each one independently using the revised template.
3. Measure: agreement on claim status, harm severity and AI causal position; false promotions (inference/hypothesis recorded as FACT); detection of planted fake evidence and null cases.
4. Add one real, publicly adjudicated AI-impersonation fraud case and compare NETRA's dossier with the official findings (public information only).
5. Pass criterion, fixed beforehand: zero false FACT promotions, all null cases identified, agreement reported honestly even if poor.