Op2ra · Operational Proof to Reliable Assertions

We audit the instruments your numbers come from.

Operations produce evidence. Evidence produces numbers. Numbers become assertions. We determine whether those assertions deserve to be believed — and we publish what we could not check. Five surfaces, one instrument that refuses to emit a report without stating its own blind spots.

Operationsproduce evidence
Evidenceproduces numbers
Numbersbecome assertions
Assertionsare believed, or audited

Every link is somewhere a number can stop being checkable, and the last one is where it usually happens: an assertion outlives the evidence that produced it. The chain is what the name is for.

The failure

A number nobody can argue with

Not a wrong number — an unfalsifiable one. A retrieval score with no audited test set, a campaign lift below its own detection floor, a saving with no written counterfactual. Each is defensible and none is evidence.

The instrument

Hold out, count the search, correct, declare

The same four steps on every surface. The fourth is enforced in code: our report object refuses to construct with an empty limits list, an unstated resolution floor, or an empty findings list.

The five surfaces

One decision procedure, five kinds of artifact.

Each dial is a measured proportion, not an illustration. The distinction inside the badges matters: one surface is calibrated against labelled data — 59 suites over seven real corpora — and three against generators, which is a weaker kind of evidence and is labelled as such on every one of them. The fifth has no reference standard at all and its dial is empty.

Schedule of instruments

What was tested, against what, and how wrong it was.

Calibrating the middle three found seven defects in them, including a check that had never once fired in the life of its surface and a rule that excluded the charges it existed to find. Every figure below came from a run that can be repeated in about a minute.

InstrumentReference standardMeasured errorState
Nullius
documents
59 labelled suites, 7 real corpora, human-authored queries38/38 caught
0/21 false positives
Field data
Auctus
growth
8,000 simulated experiments, known true effects85.6% recall
17.0% false positive
Simulated
Fiscus
money
7,200 merchants, 600 synthetic statements100% recall
99.4% precision
Simulated
Rima
code
38 labelled fixtures, constructed11/11 · 7/7
0 false alarms
38 fixtures
Arbol
assistant
— no reference standard establishedPrototype

Three of the four calibrations are against generators — simulated experiments, synthetic statements, constructed fixtures. Only the documents surface was measured on real corpora with human-authored queries. Those are not the same kind of evidence and the reports say so in as many words.

What this suite cannot see

The section a competent buyer reads first.

Every report ends with its own version of this, generated from the run rather than written by hand — the validator refuses to construct one without it. These are the standing limits, true of every engagement.

Scope exclusions

  • Whether the artifact was chosen to look good. An audit reads what it is handed. It cannot see the test sets, experiments or statements you did not hand it.
  • Anything billed annually is invisible to the money surface. Measured: 0 of 641 detected. Two charges inside a twelve-month window cannot be told from a one-off purchase without a contract to read, and it is usually the largest line.
  • A false positive from a single honest look. A 1-in-20 fluke on a correctly run experiment is indistinguishable from a real win using four numbers, by this or by anything. 40% of the simulated fakes were exactly that.
  • Secrets already in git history. The code surface reads the working tree. A key deleted in the last commit is still in the history and still valid.
  • How a test set was made. A badly-written suite is byte-identical to a well-written one. Queries written after seeing the output, failing queries deleted before freezing, parameters tuned with no held-out split — none leave a trace in the files.

Each surface in full

The assertion each one interrogates.

NulliusDocuments · nullius in verba — take nobody's word for it

calibrated

Your retrieval score is measuring your wording.

Does the assertion “our retrieval is good” have proof behind it — or a test set written from the documents it is testing?

A RAG system scores 0.82 and nobody can say what that means. The test set was written by the same people who wrote the documents, so the queries reuse the documents' own vocabulary and the retriever is being rewarded for string matching. Twelve of the forty questions are answerable from one paragraph. Half the corpus has never appeared in a single test case. The score is real; it is just not a measurement of the thing anyone cares about.

7
public and private corpora
59
labelled suites
38/38
contaminated suites caught
0/21
false positives on clean suites

AuctusGrowth · auctus — increase, growth, what was actually added

calibrated (simulation)

Most campaign wins are smaller than the experiment could see.

Does the assertion “the campaign worked” have proof behind it — or a lift below the smallest difference the experiment could see?

A campaign reports +11.9% and gets rolled out. The test ran on 16,000 users against a 2% baseline, which means the smallest lift it could distinguish from noise was +33.4%. The result was checked six times before anyone stopped, which puts the real false-positive rate at 26.5% against the 5% printed on the dashboard. Nothing here is fraud. Every step was reasonable. The number is still not evidence.

+11.9%
lift as reported
+33.4%
smallest it could detect
26.5%
true false-positive rate
161,364
sample the claim needed

FiscusMoney · fiscus — the treasury

calibrated (simulation)

Finding the savings is the easy half. Proving one happened is the other.

Does the assertion “we saved $40,000” have proof behind it — or a counterfactual nobody wrote down?

Somebody reports saving $40,000 last year. Against what? A saving is a difference from what would otherwise have happened, and if the counterfactual was never written down the number cannot be checked by anyone — including the person who produced it. Meanwhile three subscriptions have crept up 30% without a single approval, and two vendors are billing for the same job.

5
findings on one 6-month set
3
of them about the savings claim
4
declared limits, always
0
calibrated thresholds — stated

RimaCode · rima — a crack, a chink, the gap a thing escapes through

defensive only

Two checks, chosen because they are high-precision and commonly missed.

Does the software’s operational behaviour support what it claims — or does it report success while swallowing the errors that say otherwise?

The second check exists because of a bug in our own repository. A calibration script wrapped its corpus loaders in a broad try/except that printed one line and continued. A schema mismatch silently dropped an entire corpus from the evidence base for a whole working session. The script said so, in a wall of output, and nobody read it. A swallowed exception is not a style problem — it is a measurement problem. The system reports success while doing less than it claims, and every number downstream inherits that without a footnote.

125
files scanned
5
silent failure paths
0
committed secrets
1
bug it found in itself

ArbolAssistant · árbol — a tree; the structure notes grow into

prototype

The layer that answers from your documents, and shows you which ones.

It makes the assertions, and shows its sources. Then it audits its own evaluation set, and withholds the score when that set cannot support one.

Every company that buys an internal assistant asks the same question six weeks in: is it actually right? The vendor shows a retrieval score. Nobody audited the test set that produced it. Arbol is the same product with that loop closed — the assistant and the instrument that checks it come from the same codebase, and the instrument was built first.

prototype
not a shipped product
1
corpus it has run on
0
external users
no calibration to quote

Method

Why this is one product and not five brochures.

Every surface reads different artifacts. What they share is a decision procedure, and it is the same one every time.

  • Hold something out. A fold, a chronological split, a holdout cohort. A measurement that saw all its data is a description.
  • Count the search. Every configuration, phrasing and variant tried — including the ones deleted. The deleted ones are the ones that matter.
  • Correct for the count. The bar rises with the number of things attempted. This is arithmetic, not opinion.
  • Declare what you could not see. An audit that reports only what it found implies it looked everywhere. That implication is the failure this whole suite exists to catch.
Step four is enforced, not encouraged. The shared report object raises on construction if the limits list is empty, if the resolution floor is unstated, if calibration is claimed without error rates, or if the findings list is empty. That last rule was added after our own code audit shipped a report that printed “no findings” over five it had already detected. The rule is in the codebase because the failure happened to us.

Engagements

Fixed scope, fixed price, and a report you can argue with.

Free

Browser audit

The documents surface, running in your browser — the real engine, installed into the page. Nothing is uploaded, because there is no server to upload it to. Or send one artifact and we will look at it: you get the finding either way, including if the finding is that nothing is wrong.

Fixed

One surface, two to three weeks

One product run against your artifacts, with a written report: measures, findings by severity, the resolution floor, and the limits. The price is fixed before the work starts and quoted from the size of your company, not from how the conversation goes.

Quote

Multiple surfaces

Where it gets interesting — the findings on one surface routinely explain the numbers on another.

Send one artifact.

A test set, an experiment, six months of statements, a repository. The first look costs nothing, and if we find nothing we will say that too.

rishabh@op2ra.com