Sample output

One run, as Corelata delivered it.

Everything below is the product's own output from one run, with nothing rewritten: the exported report, what corelata trace prints about the run and about one claim, and the auditor's findings as the API returns them (filtered to the failed verdicts and indented).

The run used Corelata 0.1.8 with its synthetic test model, a deterministic stand-in that ships with Corelata for tests and demonstrations. No language model is involved, so the prose is placeholder text: a real model writes the assessment itself. The pipeline, the checks, the revision loop, the exports and the trace are the real ones.

The second source ends with UNSOURCED, a marker that tells the synthetic writer to add a claim with no evidence, so that the auditor has something to find.

Input

Two sources and a value statement

`varna-port.txt`, the sample source in Corelata's bundle

The port of Varna expanded its container capacity in 2025.

Capacity expansion was financed by a state bank.

Container traffic through Varna rose 40 percent.

`trade-note.txt`

A trade newsletter repeats, without naming a source, that the terminal will reach full capacity early. UNSOURCED

Value statement

The port of Varna's 2025 container expansion, financed by a state bank, is already raising traffic; shippers and the port authority should plan for higher volumes.

The audit

The auditor failed the first draft

The writer's first draft included claim c99 with no evidence. The auditor failed three rules over it. These are its findings for iteration 0 from GET /api/runs/{id}/findings, keeping only the failed verdicts (the other fourteen rules were passed or attested):

[
  {
    "rule_id": "icd203.as.5",
    "verdict": "fail",
    "evidence": "A tradecraft standard fails.",
    "cited_units": [],
    "claim_ids": [
      "c99"
    ],
    "recommendation": "Resolve the failing tradecraft findings.",
    "severity": "major",
    "route_to": "writer",
    "iteration": 0
  },
  {
    "rule_id": "icd203.tc.8",
    "verdict": "fail",
    "evidence": "Claims without evidence: c99.",
    "cited_units": [],
    "claim_ids": [
      "c99"
    ],
    "recommendation": "Cite a source unit for each claim or remove it.",
    "severity": "blocking",
    "route_to": "writer",
    "iteration": 0
  },
  {
    "rule_id": "icd206.src.2",
    "verdict": "fail",
    "evidence": "Claims without evidence: c99.",
    "cited_units": [],
    "claim_ids": [
      "c99"
    ],
    "recommendation": "Cite a source unit for each claim or remove it.",
    "severity": "blocking",
    "route_to": "writer",
    "iteration": 0
  }
]

All three are routed to the writer and two are blocking, so the engine sent them back. The writer's revision removed c99, and the second audit passed every verifiable rule. The auditor changed nothing in the report itself: it only returned findings.

The trace

The whole run, from `corelata trace`

run 6756ba82-e4a7-4613-a204-a51852bb33c1: completed clean after 1 revision(s) (max 2)
pack: icd203-base@cf6fc9be4160e0e0
statement: The port of Varna's 2025 container expansion, financed by a state bank, is already raising traffic; shippers and the port authority should plan for higher volumes.
sources:
  src1 varna-port.txt (txt, 3 units)
  src2 trade-note.txt (txt, 1 units)
gate: supported, sufficient; 7 of 16 key terms appear in the sources.
stages:
  ingesting   it0  ok        0 ms
  gating      it0  ok       22 ms
  drafting    it0  ok       10 ms
  conveying   it0  ok        7 ms
  auditing    it0  ok        8 ms
  drafting    it1  ok        8 ms
  conveying   it1  ok        7 ms
  auditing    it1  ok        8 ms
  rendering   it1  ok        0 ms
findings:
  iteration 0: failed icd203.as.5, icd203.tc.8, icd206.src.2; open 3; passed 9; attested 5; n/a 0
  iteration 1: failed none; open 0; passed 12; attested 5; n/a 0
usage: 7 calls, 70 in / 70 out tokens; auditor 2 call(s), 20 in / 20 out; conveyor 2 call(s), 20 in / 20 out; gate 1 call(s), 10 in / 10 out; writer 2 call(s), 20 in / 20 out

The same run on the review screen's Process tab: a stage timeline with drafting, editing for the reader and checking against the standards in iterations 0 and 1, and a Drafts list ending with writer, iteration 1: title changed; removed c99; sections changed: Analysis. — The same run on the review screen's Process tab: each stage per iteration, and the drafts, including the writer's revision that removed `c99`.

The source trail

Claim `c0` and its evidence

corelata trace --claim c0 prints a claim, its likelihood term, the source units it cites with their locators and text, and where the report cites it:

c0 (judgment): It is likely that The port of Varna's 2025 container expansion, financed by a state bank, is already raising traffic; shippers and the port authority should plan for higher volumes.
likelihood: likely
  src1/u2  varna-port.txt, para 2
    Capacity expansion was financed by a state bank.
cited by: key_judgments[0], analysis[0].paragraphs[0]

The report

The delivered report

The Markdown export, as downloaded:

UNCLASSIFIED

# Conveyed: Synthetic assessment: The port of Varna's 2025 container expan

## Key judgments

- **Moderate confidence.** The port of Varna's 2025 container expansion, financed by a state bank, is already raising traffic; shippers and the port authority should plan for higher volumes.

## Scope note

Synthetic scope note covering all sources.

## Sources

- varna-port.txt — open-source report; Synthetic: current and complete. (sha256 6d559cdc)
- trade-note.txt — open-source report; Synthetic: current and complete. (sha256 b1fe2895)

## Analysis

It is likely that The port of Varna's 2025 container expansion, financed by a state bank, is already raising traffic; shippers and the port authority should plan for higher volumes.

Source states: Capacity expansion was financed by a state bank.

Source states: The port of Varna expanded its container capacity in 2025.

Source states: Container traffic through Varna rose 40 percent.

## Alternatives considered

- **The opposite holds.** Less likely.

## Gaps and indicators

- Synthetic gap.

UNCLASSIFIED

Download the exports

Run 6756ba82-e4a7-4613-a204-a51852bb33c1. corelata report re-renders every export from trace.json alone.