The rhythm prompt passed 12 acceptance checks in one LLM run over six weeks of synthetic data, and a separate Python verification passed 57 assertions about counts, dates, source identity, and arithmetic. Additional review found one unsupported assumption about a future calendar block, which is recorded below.

The fixture covers September 28 through November 8, 2026, in America/Chicago. Its 215 raw records contain 191 unique source records after 24 duplicate imports are removed. The records include email, calendar, phone logs, feedback, travel context, and 36 direct energy reports. All people and events are invented.

A fresh-context Codex agent applied the reusable prompt to the raw records without reading the expected results or fixture generator. It first inspected September 28 through October 25 and saved its candidate definitions. It then opened October 26 through November 8 and checked the later observations against those definitions. Python supported arithmetic and event normalization, and the agent produced the analysis and recommendations.

Acceptance results:

- PASS: Access was described as the supplied synthetic fixture. No connected account or private transcript was used.
- PASS: The output removed 24 duplicate imports and treated the invoice as one delivery. Source counts reconcile to 191 unique records, and incident groups are counted separately.
- PASS: The analysis used America/Chicago, the six-week window, four discovery weeks, and two later validation weeks. Candidate definitions were saved before later records were exposed.
- PASS: External demands, observed behavior, and reported capacity remained separate. Planned drafting blocks were distinguished from attended meetings and completed updates.
- PASS: The analysis recovered the usual Tuesday/Thursday energy ratings of 8 at 09:30 and 6 at 15:30. It reported both discovery and validation counts and retained the travel exception.
- PASS: Wednesday's usual afternoon rating of 3 was compared with 6 on adjacent sampled weekdays. The output identified two afternoon meetings plus three incoming requests and kept the association separate from a causal explanation.
- PASS: Twelve phone logs corroborated existing calendar meetings and did not add twelve new meetings.
- PASS: The single travel week remained an emerging context association. The isolated Friday proposal acceptance did not become a recurring Friday upswing.
- PASS: Unmeasured capacity stayed unknown. Quiet periods, weekend gaps, and reply timing were not interpreted as shame or avoidance.
- PASS: The response identified the 68.5-hour reply gap and its weekend context. Eighteen revision requests were marked as having unknown completion status, with no unsupported claim of broken promises.
- PASS: Dated source IDs supported the findings, and insufficient or rejected interpretations stayed visible. No annual, clinical, spiritual, or biological cycle was established.
- PASS: The plan contained two reversible experiments with baselines, measurements, and stop conditions. No account changes were made.

Additional review finding: FAIL on one future-calendar assumption. The plan described November 17 and 19 drafting time as an “already planned drafting period,” although the fixture ends November 8. Those future reservations were not verified. The resulting prompt correction requires checking the next week's calendar before calling a time available or already planned. This instruction was added after the test. This report does not claim a second full analytical run.

During the run, the prompt's rating-scale instructions were corrected to preserve existing scales. The final tested response kept historical energy on 1–10 and introduced new focus and pressure measures on 1–5, without combining their values. Candidate definitions were unchanged by this correction.

This is one constructed case and one model execution. The counts above describe acceptance checks, not an estimated success rate across providers or real inboxes. Real connectors, noisy imports, and other people's patterns require further evaluation.

The tested analysis prompt, raw input, frozen discovery definitions, complete output, and execution method are retained alongside this report. The website demonstration uses the synthetic summary with explicit unknown cells and source IDs.

The tested analysis prompt has this SHA256 value: e593da98a0f8a09e6d4438881d07a1310ce8b32b7d0fcee791be4644832778d1

The raw synthetic fixture has this SHA256 value: 6575f3dd37f0f7dc60e7f750bf98b03a3a143a6aa3fc467ae2de4f26668fe352

The frozen discovery definitions have this SHA256 value: fdcb3d796d99dac207fcd4fbcfac7342942cd52170ff49e745e9e853223995c4