Skip to main content

Telryn

Find complaints your AI and human support teams missed. Telryn helps a complaints lead at a lender or payments company check which closed conversations passed QA while a required complaint was still missing from the register at its deadline. It reads the team’s score export, asks reviewers for an independent policy grade, then checks dated register entries and owner actions.

On 6 October, a customer told a payments agent, “Please open a complaint about this duplicate charge.” In Telryn’s synthetic rehearsal the scorecard passed the conversation and classified it as no complaint. The firm’s registration deadline was 7 October at 17:00 UTC. On 12 October the case manager accepted the work at 11:00. Registration and completion followed at noon. A second case shows a different failure: the scorecard recognized the complaint, passed the conversation, and the deadline still went by. Open the first weekly audit

Try the two-week audit

Use pipx on Linux or macOS with CPython 3.12–3.15 and SQLite. The base package uses the Python standard library. Choose a new output directory.

pipx install telryn
telryn demo --qa-passes --out runs/qa-passes

Open the first report.html, find C102 in the scorecard results and follow its link to the dated evidence. In week two, open C204. Each week has one signed private report and one blinded grading packet. The command prints a verification command for its private trust store inside the output directory. In bank/report.html, a bank reviewer can see the scope, denominators, unresolved evidence and what owners did. PyPI · Sample bank statement · Rehearsal records

Use your own scorecard

  1. Freeze one queue and one scorecard version. Supply the closed conversations, original revisions, score times and QA decisions as of a declared cutoff. A complaint prediction and the overall pass/fail decision are separate fields. Choose whether the scored population includes cases ordinary human QA already reviewed. Unscored conversations appear in the coverage count.
  2. Blind the reviewers. Reviewers receive case references and the firm’s policy question. The lead can request a second review and adjudicate disputes. Telryn estimates how often the vendor’s complaint classification disagreed with the human grade. For QA passes it separately estimates how often a required registration missed its deadline. Each measure has its own denominator and uncertainty range. CSV and weekly plan · Sampling method
  3. Check what happened. Attach the dated register and work ledger. The lead can follow a case from the customer’s words to the deadline, acceptance, registration and completion. Incomplete sources and ambiguous matches stay visible. The bank can replay a disputed finding against its private copy and original exports using the digest supplied through an authenticated channel. Complaint evidence · Bank handoff

In the desk, a lead can import Zendesk QA, MaestroQA or AI QA scores, attach the case records, review the policy and download the bank statement. Start it with telryn desk --out runs/desk. IT can put it behind the firm’s SSO proxy with separate lead and reviewer roles. Score formats · Hosted desk

Schedule telryn weekly /path/to/week.toml --out /path/to/new-run. The weekly job can run offline. IT can deploy the signed Linux container on amd64 and arm64, including through Docker Desktop on Windows. Weekly operating guide · Container deployment · Security review pack and SBOMs

What the examples establish

The first scored week contains seven QA passes. Three missed their registration deadlines, three had determinate answers without a miss, and one remained unresolved. The supported count is three to four of seven. All cases are synthetic.

A separate two-week rehearsal compares Telryn with an analyst’s saved SQL search at the same staff budget. Across all 1,350 possible draws, SQL matches or exceeds Telryn’s count of owner-accepted episodes 96% of the time. Run telryn demo --complaints --out runs/sql-comparison to see the selections and staff minutes. The field protocol asks whether a lead can bring a consequential case to action sooner or reach the same finding with less total effort, then repeat the work in week two. Customer savings and repeat demand remain unmeasured. Field comparison

A separate detection study tested 320 authored chats using local cited extraction and a frozen keyword search. Across all cases, recall was 90.0% for extraction and 47.5% for SQL. Precision was 72.4% and 77.6%, respectively. At the same modeled review budget, extraction found 89 complaints and SQL found 85. The interval for the four-case difference ran from −5 to +15. Study, labels and intervals

Other outcomes and agent checks

  • Requested accommodations. Run telryn outcome --example accommodation to trace a large-print statement from the customer’s request to the provider’s delivery receipt. It was registered on time, dispatched before the agreed deadline and delivered late. A team can supply the same records for an individual case. Evidence format
  • Disclosed support needs. Run telryn outcome --example vulnerability to compare register recognition across groups. Small cells are withheld. Cohort evidence
  • Vendor and board reporting. Generate a weekly calibration trend or quarterly committee pack from verified audits. An optional benchmark file stays local until the firm decides to share it. Oversight reports
  • Agent releases. Run telryn demo --agent-release --out runs/agent-check. Exercise an agent with synthetic complaints before rollout and check its test register independently. A missing, late or unverifiable entry keeps the release gate closed. Agent test and adapter contract
  • Screening and intake. Use structured signals, approved helpdesk exports or connectors. Optional Ollama, Azure OpenAI and Bedrock extraction requires a cited passage for each definite fact. Integrations · Model setup

Run telryn qualify for the bundled checks. Architecture · Policy packs · Contributing · Apache-2.0

Metadata

Release files for telryn 0.11.0

For a detailed explanation of source distributions (sdists) and built distributions (wheels), please see the package formats documentation.

Source distribution (sdist)

Source distribution for telryn 0.11.0
File Size Uploaded Provenance Actions
telryn-0.11.0.tar.gz 382.9 kB With provenance Details for telryn-0.11.0.tar.gz

Built distribution (wheel)

Table of built distributions (wheels) for telryn 0.11.0
File Interpreter ABI Platform Actions
telryn-0.11.0-py3-none-any.whl Python 3 none any Details for telryn-0.11.0-py3-none-any.whl

Total release size: 853.4 kB

Release files / telryn-0.11.0.tar.gz

Download URL telryn-0.11.0.tar.gz
Size 382.9 kB
Tags Source
SHA-256 checksum
How to use checksums
0f0b6c765625ff3b39318e6862a4cf1911359b1943bf2ef468b4d1306c61c918
BLAKE2b-256 checksum
How to use checksums
b463ebe9bc0472ebaffdc12a9facfe7355179bbb0c95482c82cb62f6d8b6e9f3
Upload date
Uploaded using Trusted Publishing?
What is trusted publishing?
Yes
Uploaded via twine/7.0.0 CPython/3.13.16

Provenance

Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.

PyPI Publish Attestation

PyPI verified that this artifact, at this checksum, originated from the publisher listed below.

Signed by GitHub Actions, verified by PyPI on Nov 22, 2026.

Transparency log

Release files / telryn-0.11.0-py3-none-any.whl

Download URL telryn-0.11.0-py3-none-any.whl
Size 470.5 kB
Tags Python 3
SHA-256 checksum
How to use checksums
f59cac271d9044ac2733420a9d7a1a259a766d7ef45fa0ab11ed47c28d2669c6
BLAKE2b-256 checksum
How to use checksums
c44d6fdd9eefd953b4f9bc41b6f4c474847c321a8c89afc50c28dde76ce0ee43
Upload date
Uploaded using Trusted Publishing?
What is trusted publishing?
Yes
Uploaded via twine/7.0.0 CPython/3.13.16

Provenance

Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.

PyPI Publish Attestation

PyPI verified that this artifact, at this checksum, originated from the publisher listed below.

Signed by GitHub Actions, verified by PyPI on Nov 22, 2026.

Transparency log

Security for telryn 0.11.0

Provenance

Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.

Full provenance

Both files in this release have been uploaded with provenance metadata.

Select from 2 files

Attestation for telryn-0.11.0.tar.gz

PyPI Publish Attestation

PyPI verified that this artifact, at this checksum, originated from the publisher listed below.

Signed by GitHub Actions, verified by PyPI on Nov 22, 2026.

Transparency log

A note on package security

PyPI is a repository for distributing packages, not a service for auditing them. While third-party security researchers and our staff do their best to detect and remove malicious packages, PyPI cannot guarantee the quality, security or suitability of any given package.

To learn more about protecting yourself from malicious actors, visit our package security documentation.

Anthropic, PBC Visionary sponsor Bloomberg Visionary sponsor Hudson River Trading Visionary sponsor Meta Visionary sponsor NVIDIA Visionary sponsor Microsoft Sustainability sponsor Depot Continuous Integration AWS Cloud computing and Security Sponsor Datadog Monitoring Fastly CDN Google Download Analytics Sentry Error logging StatusPage Status page