pft-task-audit

An independent consistency check for a single Post Fiat Task Node task, built only from that task's public forensic events.

Task Node records every task's lifecycle as a chain of forensic events (offered, accepted, submitted, verification, reward outcome), each viewable under a task's Forensics tab. But the status shown on a task's Overview is derived separately from that event history — and the two can disagree. Publicly documented cases include tasks displaying "Refused" with no refusal event in their forensics at all.

pft-task-audit checks that a task's story is coherent: that its events form a valid chain, that its displayed status is actually backed by an event, and that a rewarded task's payout lines up. It turns a class of manual QA checks into a reusable, reproducible tool.

What it proves — and what it doesn't

It proves a task's lifecycle is internally coherent:

• the events form a valid, contiguous, correctly-ordered chain,
• the displayed status is backed by a real terminal event (catches "phantom" statuses),
• a terminal status is classified as user- or system-initiated,
• a rewarded task's reward outcome matches any supplied payment evidence.

It does not assert fraud, missing funds, or anything requiring backend access. Every finding is derived only from the pasted public forensic payloads, so anyone can re-run it and get the same receipt.

The checks

Check	What it verifies
C1 events present	at least one valid forensic event payload was supplied
C2 single task	all events belong to the same task_id
C3 transition chain	each event's previous_status matches the prior event's status_after, and every transition is legal
C4 status backed	core check — the displayed status is backed by an event; a status with no matching event is flagged as PHANTOM
C5 terminal origin	a terminal status is recorded by an actual event, classified user- vs system-initiated
C6 reward linkage	for a rewarded task, the reward event is present and consistent with any supplied tx/amount

Verdict is CONSISTENT, CONSISTENT_WITH_WARNINGS, or INCONSISTENT. The tool exits non-zero on an inconsistent task.

Usage

1. Open a task's Forensics tab. For each event, expand Raw payload and copy the JSON.
2. Paste the events into one file as a JSON array ([ event1, event2, ... ]).
3. Run:

python3 pft_task_audit.py events.json --status refused
# rewarded task with payment cross-check:
python3 pft_task_audit.py events.json --status rewarded --reward-tx <hash> --reward-amount 3500

It writes task-audit-<task_id>.md and .json — a human- and machine-readable receipt with an SHA-256 fingerprint of the events audited.

Examples

Real receipts in examples/, generated from authentic task forensic data:

• rewarded-CONSISTENT.md — a full 6-event rewarded lifecycle; all checks pass.
• real-refusal-CONSISTENT.md — a genuine user refusal (offer + refuse); consistent.
• phantom-refusal-INCONSISTENT.md — a task showing "Refused" with only an offer event; flagged INCONSISTENT (phantom status).

The fixtures/ directory holds the authentic event data behind these receipts so they can be reproduced offline.

Tests — proving the checks bite

A verifier is only trustworthy if it fails when it should. tests/ contains fixtures that each break one rule (chain gap, illegal transition, multi-task, phantom status) and a runner that confirms the auditor catches each one:

python3 tests/run_tests.py

It runs the auditor against the clean fixtures (expected CONSISTENT) and every broken one (expected INCONSISTENT, with the specific check that must fail), and exits non-zero if any case misbehaves.

Requirements

Python 3.8+. Standard library only — no dependencies.

Scope

This tool audits whatever events you paste into it; it does not fetch from the network. It verifies internal consistency of public forensic data, not the correctness of the network's backend decisions. Related public QA reports: pft-qa-reports.

License

MIT — see LICENSE.