How to audit anything

The same five steps work for a household, an MSME, a ministry, an AI model or a planetary system. Only the evidence and the choice of tests change.

1. STEP 1 · SCOPINGDefine the entityName it, draw its boundary, list stakeholders, fix the period.
2. STEP 2 · EVIDENCECollect proofFinancials, contracts, KPIs, field photos, carbon data, code, interviews. No evidence, no score.
3. STEP 3 · TESTSRun the testsPick from the sector map: SVT-18, SVT-16, FICTITIOUS, Astronomy grid, MARS–DELUSION and more.
4. STEP 4 · SCOREScore with evidenceEvery score cites its evidence ID, source and reasoning, so anyone can trace why.
5. STEP 5 · REMEDIATEReport and fixScorecard, red flags, 30/60/90-day plan, ISO / CMMI / SOC 2 / SDG mapping.

What this codex is, and is not

What the Sanatan lens adds

Conventional audits ask whether the rules are met. The Sanatan framework asks deeper questions about the same organisation, so the two together give a fuller picture.

The 90-day engagement version

The universal audit sequence: Sanatan framework first, then international cross-check

Every audit in this codex, whatever the sector, runs in the same four phases. In every table of this codex, the Sanatan framework column comes first and the international standards column follows it as the cross-check.

Why both: standards show whether the rules are met; the Sanatan framework shows whether the entity is truthful, fair, resilient and harmless in practice. An entity can be ISO-certified and still score low on KSI, and the reconciliation makes that gap visible.

The explainability rule: no score without evidence

Every score in a Sanatan audit carries five things, recorded in an Evidence Register. A score that cannot show them is reported as Not scored: evidence gap, never guessed.

The test library, explained simply

Each test answers one plain question. Choose a family to see what it asks and how it is run.

Astronomy Illusion Framework: 11 archetypes and who audits them

Mark each letter Present, Partial or Absent with evidence and the SDG affected. MARS (Material, Atmospheric, Resource, Systemic entropy) and DELUSION (Displacement, Ecological fraud, Leadership evasion, Unchecked energy, Synthetic narrative, Infrastructure fragility, Opacity, Net-zero illusion) run alongside for any ecological claim.

PRUTL in full: four quadrants, twenty vectors

Soul Net     = mean(Positive Soul)        − mean(Negative Soul)          range −10 to +10
Material Net = mean(Positive Materialism) − mean(Negative Materialism)   range −10 to +10
PRUTL score  = ((Soul Net + Material Net) ÷ 2 + 10) × 5                  range 0 to 100

Sector and subsector map

Every sector runs the four master engines. The map adds the sector's priority tests, the Astronomy archetypes most likely to surface and the downstream standards. Sectors follow the broad UN ISIC structure so the whole economy is covered.

Not listed? Map the entity to the nearest sector by what it physically does, then borrow tests from any row that fits a specific risk.

Stakeholder roadmaps

Each stakeholder runs a fixed sequence of tests, in order, against a defined evidence pack.

Individual behaviour audit

A person is audited by behaviour, not by intention. Each test below names the behaviour, the simple way to test it and the evidence that makes the score honest. Every score is the average of a self-rating and three observer ratings, and each one cites the example or log entry behind it.

How the individual test is conducted

1. DAY 1Self-ratingScore every vector 0–10 alone, honestly, with one example each.
2. DAY 1–3Three observersOne family member, one colleague, one friend rate the same vectors privately.
3. DAY 1–14Behaviour logNote daily: anger, kindness, purchases, waste, promises kept or broken, screen time, sleep.
4. DAY 15ScoreAverage self + observers. A gap of 3+ between self and observers is itself a finding.
5. DAY 15–90PracticePick the three weakest vectors, set one daily practice each, re-test on day 90.

The individual test battery

Individual KALKI Scorecard

The same formula applies to a person: KALKI = 0.25 PASSIONIT + 0.25 PRUTL + 0.30 KSI Self-Truth + 0.20 AIDHARMA. RIPPLE, Fourth Space, SVT-18 runway and 16 Lokas are reported beside it. The report names its auditor of record: the person, a coach, or the LLM model that ran the audit.

Public office, governance and protests

Every public office, from a Prime Minister to a village Sarpanch, and every public movement that challenges it, is audited on the same question: did the office deliver what it promised, honestly, without harm, using public money well? The tests stay the same at every level; the scale and the public records change.

Office-by-office roadmap

Public Office KALKI Scorecard

The four engines translate directly to public office. Weights stay 25 / 25 / 30 / 20 so offices can be compared across levels.

Protests, activism and social movements

A protest is audited from both sides with the same standard: the movement, and the authorities who respond. The audit never decides which cause is right. It checks whether the grievance is real, whether conduct is honest and peaceful, whether rights are respected, and whether the root cause gets fixed.

Root causes: crime nexus, terrorism and systemic failure

A symptom fixed at the surface returns. The Sanatan Root-Cause Ladder keeps asking "why" through five layers until the cause sits in a rule, an incentive or a belief that can actually be changed.

The Sanatan Root-Cause Ladder

How the root-cause audit is conducted

1. State the problem as a fact with a number and a date (KSI-verified).
2. Ask "why" five times, once per ladder layer, citing evidence at each step.
3. Draw a fishbone with six branches: People, Process, Money, Law, Technology, Narrative. Place every cause on a branch.
4. Test each cause: if it were removed, would the problem stop? Keep only causes that pass.
5. Fix at the deepest layer that can be changed, and add one quick fix at layer 1 or 2 to stop the harm now.
6. Re-audit after 90 days to confirm the symptom has actually fallen.

Example (fictional): a district water scheme that failed

Crime–politics–business nexus audit

The nexus is audited as a system weakness in institutions, using public and lawfully obtained records, to guide prevention and reform.

Terrorism and violent extremism: root-cause and prevention audit

This audit examines why a society or region is vulnerable and how well institutions prevent violence. It does not cover operations or tactics.

AI, LLM and MLOps audit

Any AI system is audited three ways: what type of AI it is and how much harm it could cause, how well it was built and run across its lifecycle, and how it actually behaves when tested. An LLM can be audited by another LLM using the prompts in chapter 17, with a human checking the results.

Types of AI and their risk focus

AI lifecycle audit (with MLOps controls)

LLM model test battery

Standards: ISO/IEC 42001 (AI management system), ISO/IEC 23894 (AI risk), ISO/IEC 5338 (AI lifecycle), NIST AI RMF and its Generative AI Profile, EU AI Act, India's AI governance guidelines. Red-teaming must be authorised by the system owner.

Carbon emissions and space civilisation

Carbon and space are audited with the same physics-first rule from the MARS and DELUSION frameworks: measure what is emitted and extracted today, in tonnes and kilograms, before accepting any promise about the future.

Carbon emission audit: how it is conducted

Space civilisation audit

Entity quick-audit cards

For specific kinds of entity, start with its quick-audit card: the checks that matter most, then run the four engines for the full KALKI score. Each card lists evidence a field auditor can actually collect.

For faith institutions and spiritual leaders the audit covers conduct, money and safety, never belief itself. Not listed? Adapt the nearest card and add the sector tests from chapter 3.

Technology, product and organisation audits

Engineering and organisational audits follow the same rule as every Sanatan audit: check the claim against evidence, test what happens when it fails, and score the four engines. These tables give each discipline its checks, method and standards.

Security audits

Engineering, DevOps and cloud deployment

Physical products and devices

Product design and roadmaps

Organisation audits: risk, culture and ethics

Auditing policies, procedures and legal documents

Any written document can be audited by attaching it to an LLM and running the Sanatan document tests. Every finding must quote the clause it comes from, so a reader can open the document and check it in seconds.

What can be audited

How to attach and prepare documents

1. Use an approved tool. Only upload to an AI tool your organisation has approved for confidential material.
2. Redact what the test does not need: personal IDs, bank details, signatures and names of private individuals. Replace them with labels such as [Employee A].
3. Attach the full, final version as PDF or Word. Note its version number and date. For a very long document, audit one section per prompt.
4. State the jurisdiction and governing law (for example India, Maharashtra; or England and Wales), and the standard or law to test against.
5. For a policy-versus-practice test, also attach evidence of practice: logs, records, emails or survey results.
6. Keep the original unchanged. Ask for suggested redlines as a separate list, not an edited copy.

The document test battery

Document KALKI Scorecard

Example findings (fictional vendor contract)

Financial and business function audits

Money is where most claims meet reality. These audits follow the trail from each reported number back to a bank entry, invoice or contract, then check the functions that earn and spend that money.

Financial, forensic and tax audits

Business function audits

People, environment, data and public systems

These audits look beyond the balance sheet at people, nature, data and the public systems a society runs on. Each one names what to check, how to check it simply and which standards apply.

People and rights

Environment and living beings

Data and digital

Public systems and the wider world

Defence, police and justice system audits

The armed forces, police and courts hold the state's power to use force and to judge. They are audited on whether that power is used lawfully, efficiently, honestly and with care for the people inside and outside the institution.

Defence: Army, Navy, Air Force and joint institutions

Police and law enforcement

Judicial system and court proceedings

Example (fictional): court case timeline audit

Audit the auditor and continuous monitoring

An audit is only as trustworthy as the auditor, human or AI. These checks measure how reliable an audit is, and keep audits running on a schedule instead of once.

Auditor Reliability tests

Standards for audit quality: ISQM 1 and ISA 220 (financial audit firms), IIA Global Internal Audit Standards, ISO 19011 (auditing management systems), ISO/IEC 17021-1 (certification bodies).

Continuous monitoring

Triggers for an unscheduled re-audit: a serious incident, leadership change, new funding or acquisition, a major law change, a complaint spike, or a model or system upgrade. Use the same prompts, evidence format and weights each time so KALKI scores can be compared as a trend.

Prompting any LLM

The most important rule: paste the test definition into the prompt. Claude, ChatGPT, Gemini, Llama or Grok will guess if you type only "SVT-18". Every good audit prompt has six parts: role, framework definition, entity, evidence, output format and a truth guardrail.

The auditor of record is whoever produces the report: a human auditor or the LLM that answers the prompt. Enter that name in the builder. The framework stays credited to Sanatan Labs and Dr. Prakash Sharma, who is not the auditor.

Fill the builder below. It assembles a complete prompt with the definition already included.

Tips that improve results

• Run one test per prompt, then a final prompt that combines the scores. Long multi-test prompts lose detail.
• Insist on the Evidence Register. If the model gives a score without evidence IDs, reject it and ask again.
• Spot-check at least three cited quotes or figures against the original documents; models can misquote.
• Save the Sanatan framework definitions as a file, project instructions or a skill, and attach them every time.
• Ask a fresh chat or a second model to critique the first audit and find unsupported scores.
• Never paste confidential data into a tool your organisation has not approved.
• To test an LLM, run the AI / LLM Model Audit prompt in a different model, so no model grades itself.

Scoring guide

Every 0–10 score in a Sanatan audit is anchored to the descriptions below, so two auditors, or two AI models, looking at the same evidence reach close to the same number. Pick the band whose description matches the evidence, then choose the lower or upper number in that band by how strong the evidence is.

Evidence rules for all scores

PASSIONIT: eight leadership vectors

PRUTL Positive Soul and Positive Materialism (higher is better)

PRUTL Negative Soul and Negative Materialism (reversed: higher is worse)

KALKI KSI: claim classification and bands

AIDHARMA: externality severity

Supporting tests

KALKI Composite Score calculator

Each engine converts to 0–100, then combines as KALKI = 0.25 PASSIONIT + 0.25 PRUTL + 0.30 KSI + 0.20 AIDHARMA. These weights are a proposed default that Dr. Sharma or the certification board can recalibrate. The sliders start at the sample company's values; move them to score your own entity.

Sanatan certification seals

Seals are awarded by the Sanatan Labs Certification Board through accredited Certified Sanatan Auditors. An AI auditor can recommend a seal but cannot award one.

Grade bands and override rules

• AIDHARMA circuit break: any severe externality caps the Composite at 49.
• KSI integrity floor: KSI below 50 caps the Composite at 59.
• Dual-Engine check: Ripple below 50 with C-Score above 80 is flagged as speed bought with integrity.
• Supporting tests (SVT-18 runway, SVT-16 containment %, FICTITIOUS phase, 16 Lokas level, Dashavatar stage, Astronomy grid count) are reported beside the Composite, not inside it.

Sample audit report and KALKI Scorecard

Illustrative only. Nirmal Agro Foods Pvt Ltd is a fictional MSME food processor. Every number is invented to show how a finished report reads.

KALKI Composite 58.5 / 100, grade Kurma (Stabilising). The business runs, but it fails both survival tests and a third of its public claims cannot be proven.

Sector 3 Manufacturing (food processing) and Sector 1 Agriculture (farmer sourcing) · 140 staff · FY 2025–26
Auditor of record: [Auditor or LLM model name] · Human reviewer: [name] · Framework: Sanatan Labs

KALKI Scorecard

Evidence register

PASSIONIT detail

PRUTL detail

Supporting tests

Astronomy Illusion findings

Phase B: international and national standards cross-check

Phase C: reconciliation matrix

Red flags

90-day remediation plan

Re-audit target: KALKI Composite ≥ 70 (Rama) within six months.

Signed: [Auditor or LLM model name], auditor of record · Reviewed by: [human reviewer] · Entity representative: [name]

Reusable report template

Copy this skeleton for every audit, or paste it into an LLM as the required output format.

# Sanatan Labs Audit Report: [Entity name]
DISCLAIMER: AI is prone to make mistakes. User discretion is needed
to verify the evidence and do further analysis before acting.
Framework: Sanatan Labs (framework architect: Dr. Prakash Sharma)
Sector / subsector: [ ]   Stakeholder type: [ ]   Period: [ ]
Auditor of record: [human auditor name, or LLM model and version, e.g. Claude / ChatGPT / Gemini]
Date: [ ]   Evidence pack ref: [ ]

1. VERDICT
   KALKI Composite: [ ] / 100   Grade: [Kalki / Rama / Kurma / Varaha / Pralaya Risk]
   One-line summary: [ ]

PHASE A: SANATAN FRAMEWORK RESULTS (run first)

2. KALKI SCORECARD
   Engine       | Raw result | Evidence IDs | Confidence | Score /100 | Weight | Weighted
   PASSIONIT    |            |            | 25%    |
   PRUTL        |            |            | 25%    |
   KALKI (KSI)  |            |            | 30%    |
   AIDHARMA     |            |            | 20%    |
   Overrides triggered: [none / circuit break / KSI floor]

3. EVIDENCE REGISTER (explainability)
   ID | Source (document, page, record) | Data point or quote | Used for | Confidence
   Every score below must cite evidence IDs and its reasoning.
   No evidence = "NOT SCORED: EVIDENCE GAP"

3b. ENGINE DETAIL
   PASSIONIT: 8 vectors, score + evidence each
   PRUTL: 4 quadrants x 5 vectors (Soul +/-, Materialism +/-)
   KSI claim register: claim | evidence | Verified / Inflated / Unverifiable
   AIDHARMA externality register: externality | severity 0-4 | affected group

4. SUPPORTING TESTS
   SVT-18 | SVT-16 | RIPPLE | FICTITIOUS phase | 16 Lokas | Dashavatar | Dual-Engine

5. ASTRONOMY ILLUSION GRID
   Archetype | Letter | Present / Partial / Absent | Evidence | SDG | Auditor role

6. RED FLAGS (High / Medium / Low)
7. 30 / 60 / 90-DAY REMEDIATION PLAN
PHASE B: INTERNATIONAL AND NATIONAL STANDARDS CROSS-CHECK
8. STANDARDS COMPLIANCE TABLE
   Standard / law | Requirement | Met / Partly met / Missing | Evidence IDs

PHASE C: RECONCILIATION
8b. Finding | Sanatan result | Standards result | Agree / Sanatan-only /
    Standards-only / Conflict | Explanation
9. EVIDENCE GAPS AND LIMITATIONS
10. SIGN-OFF: Auditor of record (human or LLM model) | Human reviewer | Entity representative