Open Indian Accountability Stack.
openIAS is a docs-first foundation for building an AI-native public accountability system for India. The project models how public money moves, which officials and systems touch it, what evidence should exist at each stage, and how AI agents can help officials, auditors, journalists, civil society, and citizens detect breaks in the chain.
The working product codename is PaperTrail: an evidence graph plus Slack-style agent workspace where each government role can have an accountable AI agent that explains, reconciles, escalates, and records its work.
The reference map above shows the full system PaperTrail is modeling: decision bodies, fund tiers, digital rails, influence actors, datasets, and oversight paths. The bridge story below is the simpler front door into that larger graph.
Imagine a state announces a Rs 50 crore rural bridge.
Most people assume leakage happens when money is transferred. PaperTrail starts from a sharper question: what if the overpayment was already baked into the paperwork before the first rupee moved?
The bridge story is simple:
- Parliament and the state legislature approve money for road or bridge schemes. At this point, it is permission to spend, not cash in a contractor's account.
- A department prepares a Detailed Project Report. If a bridge that should cost Rs 38 crore is estimated at Rs 50 crore, the extra Rs 12 crore has already become legal-looking paperwork.
- The tender goes live. If there is one bidder, repeated winners, bids clustered around the estimate, or a winning bid suspiciously close to the inflated estimate, PaperTrail marks the tender for review.
- Money sits in the scheme account and moves to the contractor in running account bills: mobilization advance, foundation, pillars, deck, finishing.
- A Junior Engineer measures work in the Measurement Book, the Executive Engineer clears the bill, and the payment moves through PFMS or treasury rails into the contractor's bank account.
- The bridge on the ground is the final truth. If the record says premium steel and full concrete volume, but photos, satellite imagery, inspection reports, or first-monsoon failure say otherwise, the leakage has become physical.
So the public-money path is:
Consolidated Fund / state share
-> scheme account
-> payment instruction
-> contractor bank account
-> bridge on the ground
But the accountability path is:
DPR estimate
-> tender competition
-> running account bills
-> measurement book
-> payment record
-> physical bridge
PaperTrail's job is to reconcile those two paths. Paper says X, money moved Y, ground reality shows Z. The mismatch is the product.
PaperTrail should feel like a civil-services command room built inside a Slack-style workspace. A bridge case becomes a shared thread, agents take roles, tickets track unresolved work, evidence chips show the source trail, and the workshop panel shows what every agent is doing.
In the bridge example, the Collector Agent creates PTL-384, the Works Agent checks DPR, tender, running account bills, Measurement Book, and quality evidence, the Finance Agent reconciles sanction and payment records, the Data Steward Agent flags missing IDs, and the Audit Agent prepares a source-linked review note. Humans approve any external action.
The repo now includes a deterministic PTL-384 bridge demo that turns synthetic
records into source-linked Sense Signal tickets.
Run:
python3 scripts/generate_bridge_signals.pyThe demo reads DPR, tender, payment, Measurement Book, physical-evidence, and
grievance records from examples/ptl-384-bridge/records/, then emits review
tickets such as DPR_COST_OUTLIER, SINGLE_BID_OR_BID_CLUSTERING,
MEASUREMENT_BOOK_MISMATCH, and QUALITY_DELIVERY_GAP.
This is the smallest product proof: records go in, explainable accountability tickets come out, and every ticket cites the exact fields behind the claim.
The same evidence-spine pattern applies to police complaints. A citizen should not have to guess whether a complaint became a station diary entry, FIR, zero FIR, non-cognizable report, preliminary inquiry, refusal, transfer, or oversight matter.
The Police Complaint Agent does not decide guilt or publish allegations. It structures the complaint, finds the missing receipt/status/forum, checks safety risk, and drafts human-approved next questions for the police station, SP office, PCA, NHRC/SHRC, cybercrime portal, RTI, or legal review.
Every public rupee should have an inspectable trail:
- Where it was allocated, released, spent, and physically delivered.
- How much moved at every custody boundary.
- When each event happened.
- Why a mismatch, delay, duplicate, or exception exists.
- Who had formal responsibility, custody, discretion, review power, or escalation duty.
openIAS does not begin with a corruption accusation. It begins with a structured question: "What evidence should exist if this public-money claim is true?"
India has already digitized large parts of the fiscal stack: PFMS, DBT rails, Aadhaar/APB, scheme MIS systems, GeM, eProcurement, state IFMS systems, and public reporting portals. The hard product opportunity is not just another dashboard. The opportunity is a reconciliation and accountability layer that connects the dots across fragmented systems and turns mismatches into explainable work items.
The 2026 product thesis:
- DBT reduced many last-mile cash leakages in direct-benefit flows.
- The harder leakage and accountability gaps often sit upstream: eligibility gatekeeping, fund release delays, works measurement, procurement, vendor billing, data mismatch, and weak escalation loops.
- AI agents are useful only if they are constrained by evidence, permissions, audit logs, and human approval gates.
- The workspace should feel like a government-grade agent-coordination pattern: channels, tickets, PRDs, approvals, budgets, agent heartbeats, and immutable receipts, but adapted to public administration rather than software teams.
.
|-- README.md
|-- CONTRIBUTING.md
|-- docs/
| |-- 00-bridge-story.md
| |-- 00-vision.md
| |-- 01-india-public-money-flow.md
| |-- 02-accountability-model.md
| |-- 03-agent-workspace-architecture.md
| |-- 04-sense-signals.md
| |-- 05-product-roadmap.md
| |-- 06-graph-visualization-spec.md
| |-- 07-civil-services-agent-workspace.md
| |-- 08-police-complaint-agent.md
| |-- 09-bridge-signal-engine.md
| |-- adr/
| | `-- 0001-docs-first-accountability-stack.md
| |-- assets/
| | |-- india-public-money-flow.png
| | `-- papertrail-workspace-mockup.png
| `-- references/
| `-- official-anchors.md
|-- data/
| `-- taxonomy/
| |-- edges.json
| |-- nodes.json
| `-- signals.json
|-- examples/
| `-- ptl-384-bridge/
| `-- records/
|-- papertrail_demo/
| `-- bridge_signal_engine.py
|-- prompts/
| |-- civil-services-workspace-super-prompt.md
| `-- money-flow-visualization-super-prompt.md
|-- scripts/
| `-- generate_bridge_signals.py
`-- tests/
`-- test_bridge_signal_engine.py
- Read docs/00-bridge-story.md for the simplest version of the product story.
- Read docs/00-vision.md for the mission and product boundary.
- Read docs/01-india-public-money-flow.md for the fiscal plumbing model.
- Read docs/02-accountability-model.md for responsibility, custody, discretion, and oversight.
- Read docs/03-agent-workspace-architecture.md for the agent system design.
- Read docs/04-sense-signals.md for the evidence and mismatch taxonomy.
- Read docs/07-civil-services-agent-workspace.md for the Slack-style interface spec.
- Read docs/08-police-complaint-agent.md for the police complaint workflow and source-backed signal design.
- Read docs/09-bridge-signal-engine.md and run the PTL-384 demo for the first executable proof.
- A system design and taxonomy repo for Indian public-money accountability.
- A seed schema for graph visualizations and reconciliation agents.
- A product strategy base for a PaperTrail-style evidence workspace.
- A place to make assumptions explicit before code starts enforcing them.
- Not an allegation engine.
- Not a political scoring product.
- Not a replacement for CAG, PAC, RTI, vigilance, courts, Lokayukta, departmental proceedings, social audit, or citizen grievance systems.
- Not a personal-data warehouse.
- Not a system that takes coercive action without authorized human decision-makers.
The product borrows interface and operating-system ideas from modern agent-coordination tools, but the domain model is grounded in Indian public administration, public finance, and auditability. openIAS is not affiliated with any such product.
This repository is initialized as a docs, taxonomy, and first deterministic demo base. The next useful build is either:
- an interactive D3 graph that renders
data/taxonomy/*.json, or - a workspace prototype that renders the
PTL-384generated Sense Signals as agent-owned tickets with evidence chips and human approval states.