Code Effects VS QA Batch Scorer

QA Batch Scorer's answer:

Three things, and the third is the one nobody else does.

It scores 100% of your transcripts rather than the 2 to 5% sample a supervisor can physically get through. Every criterion comes back with a verbatim quote from the transcript as evidence, so a score can be checked rather than trusted, and no agent has to accept a number they cannot see the basis for. And it runs entirely on your own machine using your own AI provider key, so transcripts never reach a server we control.

That last point is the structural difference. There is no account, no upload, and no vendor in the data path. It is a single HTML file you open in a browser.

QA Batch Scorer's answer:

Mostly you should not, and it is worth being clear about that.

If you need real-time agent assist, a coaching workflow, dialler integration, or audio transcription, buy a platform. Observe.AI, Scorebuddy, MaestroQA, Convin and Enthu.ai all do things this does not. This is one job done properly, not a suite.

Choose this if you are a floor under about fifty seats and the platform economics do not work. Entry pricing in this category starts around 35 to 59 dollars per agent per month, usually billed six or twelve months up front, which puts a 25-seat team somewhere between ten and seventeen thousand dollars a year on a contract. This is 99 dollars once, unlimited agents, no renewal.

Also choose it if your transcripts contain things you would rather not upload. Nothing here touches a server we run.

And choose it if you want to know what is actually on your calls before committing budget to a platform. Score a month of calls for a few dollars in API credit and you will know whether the problem justifies the spend.

QA Batch Scorer's answer:

Owner-operators and ops leads at contact centers roughly 5 to 60 seats, in businesses where nobody has budget for a QA platform but the calls still carry risk.

In practice that means small BPOs, collections and receivables desks, healthcare intake teams, inbound sales floors, and in-house support teams inside small and mid-sized businesses.

The common shape is a supervisor who reviews a handful of calls a week, knows that is not coverage, and has no realistic path to changing it. The buyer is usually the owner or the person accountable for compliance exposure rather than a dedicated QA analyst, because at this size that role often does not exist.

It is not built for enterprise contact centers. Above roughly a hundred seats you want workflow, calibration tooling and integrations that this deliberately does not have.

QA Batch Scorer's answer:

It started from a number that most people in this industry have stopped questioning: the share of calls that actually get reviewed. On a floor under fifty seats it is usually somewhere between two and five percent, because that is what one supervisor can physically listen to in a week. Everyone treats that as a capacity fact rather than a quality problem.

Every tool built to fix it is priced for enterprise. Entry pricing in this category starts around 35 to 59 dollars per agent per month and is usually billed six or twelve months up front, which means the smallest operators, the ones with the least QA slack, are the ones priced out entirely.

The other thing that kept coming up was data. Call recordings and transcripts contain account details, payment information and disclosed hardship. Asking a small operator to upload all of that to a vendor they have never met, to solve a problem they are not sure they have yet, is a real barrier.

So this was built the other way round. One file, no account, no upload, and the buyer's own AI key so the transcripts stay in their control. One price, paid once, so trying it does not require a contract. It does one job and deliberately stops there.

It is new, and honest about that. Version 1 shipped in August 2026.

QA Batch Scorer's answer:

Deliberately almost none, and that is the point.

The whole product is a single HTML file: vanilla JavaScript, no framework, no build step, no dependencies, no package manager. It opens directly in Chrome, Edge, Firefox or Safari from your file system. There is no backend, no database and no hosting, because there is nothing to host.

Scoring is done by whichever large language model you point it at. It supports OpenRouter, Anthropic and OpenAI, calling their APIs directly from your browser with your own key. The model list is fetched live from the provider at runtime rather than hardcoded, so it does not break when a model is retired.

CSV, TSV and JSON parsing is written from scratch inside the file, so nothing has to be installed and no data passes through a third-party library.

The constraint is a feature rather than minimalism for its own sake. No server means no place for your transcripts to sit. No install means it runs on a locked-down machine without an IT ticket. No dependencies means it will still open in five years.