I built a grader that refuses to grade

Most accountability sites have the same problem. They will always give you a number.

Thin evidence, missing votes, a record with holes in it, does not matter. Something gets published, because a page with a grade on it gets shared and a page that says "we do not know yet" does not.

I built RepWatchr to do the opposite, and I want to walk through the part that most people would call a bug.

It refuses to answer.

The gate

The published grading method sets three tiers of output, and which one you get is decided by the evidence, not by the official.

To get a full number and a letter grade, at least 80% of the planned weight has to be scoreable, with 70% weighted coverage and 65% confidence, four categories present, and ethics scoreable.

Clear 60% of planned weight and 45% confidence and you get a number labeled provisional, with no letter attached.

Below that, you get NR. Not rated. No score, no letter, and a visible checklist of exactly what evidence is still missing.

The page says it plainly. NR is not a favor and it is not a punishment. It means the record has not cleared the same gate everybody else has to clear.

That is the whole point. The gate is identical for a congressman you like and a congressman you cannot stand.

What cannot move a score

The other half is the evidence ladder, and this is where the arithmetic gets honest.

Every source that enters the system carries a weight. Official roll calls, journals, filings, audits, final ethics orders and court records carry 1.0. Transcripts, meeting video, agency datasets and direct official statements carry 0.9. Named independent reporting tied to primary records carries 0.7. Campaign, advocacy and social material carries 0.4, and the page is explicit about what that 0.4 actually proves: that the speaker said it, not that the accusation is true.

Unverified material carries zero. It is a research lead. It cannot move a score at all.

Evidence weight by source tier

Source: RepWatchr performance grade method v1.0 ↗

I want to be very clear about why that rule is in there.

I have been on the other side of a record built out of repetition. Allegations become deductions when enough people say them enough times, and by the time the actual disposition arrives nobody is reading anymore.

So the system will not do it. An allegation can be reported. It cannot be scored. Only a final authoritative finding touches the number.

There is due process built into the software too. Notify the official of a score-moving event. Ten business days to respond. A second reviewer on high-impact records. Publish the source, the calculation, and the response. Route the appeal to someone who did not make the first call.

And corrections create a new version instead of quietly overwriting the old calculation, so you can see what changed.

The empty table

Here is what all that discipline actually looks like in production, and it is not glamorous.

The water rights scorecard has no officials in it.

The table is empty. And instead of hiding that, the page explains it: a vote-record scorecard only appears after every vote on it is matched to a reviewed bill file carrying a public source link, and an official missing from that table has not cleared review yet.

Then it adds the line I care about most.

Do not read that as a clean record.

An empty table with an honest explanation is worth more than a full table of numbers nobody can defend.

The lesson, and it is not about politics

This is the difference between a website and software, and it is the most expensive thing to explain to a business owner.

A website displays what you type into it. If you type a grade, it shows a grade. It has no opinion about whether you should have typed it.

Software has rules. It can compute an answer, check whether the answer is defensible, and then decline to display it. A website cannot decline. It has nothing to decline with.

That is not a small distinction and it is not a technical one.

If your business makes any claim that a customer might check, ask yourself which one you have. Do you have a page where somebody types a number, or do you have a system that will not publish a number it cannot back?

An estimate. A turnaround time. A capacity claim. An availability calendar. A price. A result.

Every one of those is a place where a website will happily let you publish something you cannot defend, and a system will stop you.

Most small businesses do not need software. I say that regularly and I mean it. Four honest pages beat a bad app every day.

But the moment you are publishing claims that people will make decisions on, you have crossed a line, and a page you type into is no longer the right tool.

Do not threaten anyone. Do not harass anyone. Do not contact anyone in my name.

Read it. Check it. If a record on RepWatchr is missing a source, send the source and it gets weighed by the same rules as everything else.

That is the deal. Same gate for everybody.