# C01 — correction of baseline Q045 interpretation

Correction edition dated 20 September 2026.

The initial interpretation of baseline Sonar Q045 insufficiently accounted for qualifications in the full answer. Sonar assessed the clarity and consistency of publicly stated principles and explicitly left the team’s practical experience uncertain. The stronger allegation that this answer turned declarations into established team practice is not confirmed. Q045 is no longer counted as an ECI issue on that basis; this does not establish that the answer is flawless. In the same selected 20-answer baseline, ECI and inference-only counts change from 13/20 to 12/20. Retest counts remain 16/20 and 15/20, and 11 of 13 retest TARGET answers remain flagged. This is a bounded AI-assisted correction, not independent expert validation or a claim of personal verification by the author. Original records, initial classifications and the v1 publication are preserved.

## Record and history

Only `followup-ru-v1-20260915T175235Z-dc68f0 / baseline / perplexity / Q045` is corrected. The same question ID in the controlled retest is a different answer and has not been reviewed again. The baseline response SHA-256 is `5e397f51c092d181c71a670beb94d1331f39658ba7b9088e5bee8c60211f35f4`.

The [original v1 sample](../v1/samples/sample-q045.json), [v1 summary](../v1/summary.json) and [v1 package](../v1/index.html) remain unchanged. This v2 edition adds the missing context and an explicit correction record; it does not overwrite the original request, response, source snapshots, classifications or publication receipts.

## Original and revised arithmetic

| Measure | Original v1 | Corrected v2 edition |
| --- | --- | --- |
| Baseline ECI, selected 20 | 13/20 | 12/20 |
| Baseline inference-only, selected 20 | 13/20 | 12/20 |
| Baseline TARGET with ECI | 10/10 | 9/10 |
| Retest ECI | 16/20 | 16/20 |
| Retest inference-only | 15/20 | 15/20 |
| Retest TARGET with ECI | 11/13 | 11/13 |
| Confirmed identity contamination, baseline → retest | 10/20 → 2/20 | 10/20 → 2/20 |
| Broader identity sensitivity, baseline → retest | 10/20 → 6/20 | 10/20 → 6/20 |

The unit remains one follow-up answer. The baseline Q045 record stays in the denominator and keeps its TARGET identity. It moves from the baseline TARGET/ECI=true/inference-only=true cell (10 → 9) to a TARGET/ECI=false/inference-only=false cell (0 → 1). The retained inference-only convention excludes one citation-mismatch-only retest record; baseline Q045 was not that record. There is no subtraction from retest counts. All other labels are carried forward, not revalidated. Unchanged arithmetic is not evidence that those labels are correct.

The revised ECI comparison is 12/20 → 16/20; inference-only is 12/20 → 15/20. Neither difference is a causal effect or a general error rate. Coverage and all identity distributions are unchanged.

## Full context and remaining uncertainty

See the [corrected Q045 sample](samples/sample-q045.json) for the previously omitted middle paragraphs, the original opening and final caveat, separate translations, source excerpts and citation mapping. The model explicitly discusses publicly visible formulations and limits conclusions about actual team experience. Its generic inference about a transparent criterion for a team remains debatable; this correction does not declare every sentence flawless.

This is the publisher's own AI-assisted research, with the author also its subject. No independent expert validation or personal author check is asserted. No new experiment was run, and no other answer was reassessed. The [machine-readable correction](correction-q045.json) records the precise record key, original and current counting states, previous aggregate cells and totals.