Home

The Receipts

Data extracted from 366K messages. Patterns that emerged.

432 phrases·1,778 corrections·151 weeks tracked

Weekly Velocity

354
msgs/week (2023)
12,991
msgs/week (peak)
36× growth in 3 years

"i want to build it in 1 week"

2024-06 | MVP Planning | chatgpt

"Great and I want to build this product in a week"

2025-03 | Project kickoff | claude-code

Peak: July 2025 — 12,991 messages in a single week. 151 weeks tracked total. The velocity correlates with shipping: highest output during WOTC and Brain MCP builds.

Signature Phrases

Recurring expressions that define communication style. Not what was said — how it was said.

PhraseCountInsight
"give me the"789xDirect request, expects output not process
"and give me"429xChaining requests, transactional style
"no no no"333xStrong disagreement, immediate redirect
"please give me"262xPoliteness + directness combined
"what is the"251xDirect questions expecting direct answers
"step by step"244xSequential, monotropic processing
"now give me"215xUrgency, immediate delivery expected
"make sure to"206xPrecision focus, need for certainty
"all of the"180xComprehensive, exhaustive coverage
"want to be"180xAspirations, future-oriented thinking

432 total phrases tracked across 8 categories: command, correction, process, emphasis, intention, question, evaluation.

The Core Rhythm

"give me the" → "no no no" → "step by step" → ship

This pattern is consistent across 2023, 2024, 2025, 2026. The THINKING didn't change. The TOOLS adapted to match it.

Teacher + Conductor

"Mixture between kindergarten teacher and orchestra conductor. Early ChatGPT was like a garden hose. As context windows grew, it became a fire hose. Being a firm conductor has become critical."

2026-01 | Language discussion | claude-code
Kindergarten Teacher
Shaping, guiding, patient. Used when exploring, teaching context.
Orchestra Conductor
Firm, precise, controlling. Used when directing output.

The Evolution

Early ChatGPT (2023)Garden hose
Easy to shape, small output. More teacher.
Growing context windowsFire hose
Minutes of autonomous output. More conductor.

As AI output grew from trickle to flood, firm conductor became critical. This is what SHELET is — the framework for conducting.

Correction Patterns

When and how AI gets redirected. The "no no no" pattern quantified.

1,778
Total Corrections
1.9%
Average Rate
9.1%
Peak Rate
Jan 2026
238
Peak Count
Sep 2025 (SHELET)

Monthly Breakdown

MonthCorrectionsRate
2026-0149.1%
2025-12810.6%
2025-11341.5%
2025-10721.9%
2025-092382.5%
2025-08441.0%

September 2025 peak: 238 corrections during SHELET development.

Deep Dive Stats

Metrics that show depth, not just volume.

581
Deep Conversations
(100+ messages each)
116
AI Models Tested
Platform agnostic
187
High Output Days
(productivity outliers)
35
Domains Touched
Sequential mastery
581 conversations with 100+ messages — not casual chats. Deep problem-solving sessions averaging 4-6 hours each. 1,117 messages in the longest session (WOTC pipeline).

What This Shows

1.
Active not passive — 1.9% correction rate shows engagement
2.
Command-driven — "Give me" dominates (direct requests for output)
3.
Sequential — "Step by step" reveals monotropic processing
4.
Precision-focused — "Make sure" emphasis shows need for accuracy

Intellectual Evolution

How language changed over time. As fluency increased, verbosity decreased.

Q1 2023Exploratory

"what if...", "can you help..."

Q4 2023Direct

"give me...", "show me..."

Q2 2024Efficient

Minimal words, assumed context

Q4 2025Integrated

Shorthand, shared vocabulary