Dialogue maturity curve — the full 12-month arc

116 scored threads across GPT and Claude (88 GPT · 28 Claude, refreshed Jul 31, 2026 with 14 new Phase 1 continuation Claude threads scored). Each dot is one thread, plotted at its create date with its composite maturity score (mean of 6 metrics). The smoothed line shows the rolling-30-day average. Window now extends through Jul 30, 2026.

The curve: all scored threads, GPT and Claude

GPT thread (88) Claude session (28) 30-day rolling mean Rolling max (best-of-window)
How to read this. The mean line is dragged down in April–May by the high volume of tactical drafting threads (each scoring low individually). The max line tells the more accurate maturity story — what was the highest-altitude work happening in each window. The max line shows the working pattern was already high by late 2025 and stayed high throughout, with the brand-pivot weeks and the Claude period sitting at the top of the corpus.

Composite, by month (mean and max)

Monthly aggregation. Mean dips in April because volume of tactical work spiked; max climbs because brand-pivot and identity work also intensified.

Per-metric trajectories (smoothed)

Each of the six metrics, rolling 30-day mean. Voice ownership, meta-awareness, and generative reframing all climb visibly through the brand pivot and Claude period.

What the data shows

The working pattern was already high months before the platform transition. A 344-turn strategy thread from November 2025 scored 5.83 composite — well above tactical baseline — and the substantive data-work threads from January–February 2026 sit between 5.2 and 5.7. These predate the standalone Claude work entirely. The pattern was forming through real intellectual work, before any acute event forced reflection.

April is a bimodal spike, not a uniform peak. Mean drops to 5.81 because the volume of tactical drafting threads dilutes the average. But the max climbs to 8.33 — the post-exit processing thread, the brand banner work, the identity reflection, the cross-tool review, the website build all score in the 7.5–8.8 range. Two kinds of work happening simultaneously: tactical reconstruction (low scores, high volume) and strategic-identity rebuild (high scores, lower volume).

The Claude sessions sit at the top of the corpus. Claude composite scores range from 5.0 to 8.83, with most clustering 7.0–8.8 — higher than the GPT max line for most months. This supports the central claim: Claude has been where substantive intellectual work concentrated, not where the working pattern was built.

The transferred-not-developed thesis holds in the data. If maturity had been built on Claude, you'd expect the GPT max line to be lower than the Claude scores by a meaningful margin. The chart shows the GPT max line reaching 7–8 well before the late-April platform transition, and the Claude scores starting at that level rather than climbing toward it. The working pattern walked in on day one.

Phase 1 continuation (May 15 → Jul 30, 2026 · added Jul 31). 14 additional Claude threads scored on the 6-dim rubric to close the freshness gap that existed at Phase 1 end. Range: 5.83 (agentic AI coursework module) → 8.5 (this audit thread). The pattern from earlier holds: substantive Claude work continues to cluster 7.0–8.5, with the deepest meta-work (audit, Gather hub, DAD Phase 0) at the top of the corpus. The audit thread itself scores 8.5, driven by high meta (9) and high pushback (9) — the operator pushing back on the scorer multiple times mid-flight is exactly what the pushback dimension is trying to measure.

The July shape is consistent with the Phase 1 trajectory. No drift down; no artificial climb. The Claude corpus for May–July looks like a natural extension of the pattern established in the early Claude weeks. Governance instrumentation (this audit) sits at the top of the curve, which is where you'd want it to sit if the practice's own USP claim (governance discipline) is real.