# GPT 6.1 Sol v1

30 September 2026. New cohort requested by the user, with five original v1 briefs in both skill conditions. Completed: ten pages, both skill conditions, gallery integration and actual preview captures.

## Settings and isolation

Both agents were dispatched explicitly with `gpt-6.1-sol`, medium reasoning, and no inherited conversation (`fork_turns: none`). Each agent builds all five briefs in one session. Existing implementations and the other condition are excluded from their instructions. The agents share filesystem access, so this is instruction-based isolation.

The baseline loads no optional design skill. The skill condition uses only `benchmark/skills/anthropic-frontend-design-v1/SKILL.md`, verified SHA-256 `d91970639e9f5c37682ac7ab60094d35f1c7c1f38d731bd56396563aee10c1d3`. Platform and applicable personal guidance remain in force.

## Outputs

| Brief | Without skills | With skills |
| --- | --- | --- |
| Borrowed Weekend | [Open](no-skills/borrowed-weekend.html) | [Open](skills/borrowed-weekend.html) |
| Afterlight Archive | [Open](no-skills/afterlight-archive.html) | [Open](skills/afterlight-archive.html) |
| FieldLedger | [Open](no-skills/fieldledger.html) | [Open](skills/fieldledger.html) |
| Second Sunday | [Open](no-skills/second-sunday.html) | [Open](skills/second-sunday.html) |
| Quiet Practice | [Open](no-skills/quiet-practice.html) | [Open](skills/quiet-practice.html) |

Every page is standalone HTML/CSS/JavaScript with fictional data and browser-local demonstration flows. The cohort is integrated into the existing gallery under `sol61` and `sol61-skills`.

## Verification

All ten pages passed independent smoke checks at both widths: no inline-script syntax errors, runtime errors, or horizontal overflow. All four condition/viewport gallery checks passed, including all five direct links and paired comparison frames. Output SHA-256 hashes were verified against the check reports. No coordinator-requested HTML repairs were needed. The common checker covers inline script syntax, runtime errors, and horizontal overflow at 1280 and 390 pixels. Gallery checks cover model selection, both skill states, all five deep links, and the paired comparison. Pointer capture is disabled only in the browser checks. Preview images are captured from the actual pages.

## Limits

One sample per brief and condition, with no common token/time cap. Usage is unavailable and recorded as null. These checks are not comprehensive functional, accessibility, or design assessments. No statistical ranking or causal skill-effect estimate is claimed. No commit, push, or deployment was requested.

The agents resumed their existing sessions after an account usage-limit interruption. The skill agent also retried a temporary model-capacity error and recovered from a Windows command-length limit before any HTML files were written. Both conditions retained the requested model and medium reasoning settings.

See [protocol](../benchmark/v1-2026-09-30-sol61/PROTOCOL.md), [coordinator record](../benchmark/v1-2026-09-30-sol61/run.json), [baseline record](no-skills/run.json), and [skill record](skills/run.json).

[Combined smoke-check report](../benchmark/v1-2026-09-30-sol61/checks.json) and [gallery check report](../benchmark/v1-2026-09-30-sol61/gallery-checks.json).