Franck PachotDatabase field notes Corpus analysis Article archive

AI corpus analysis · 1,369 publications · 2010–2026

Did my employer change how I write about databases?

Rather than judge my own writing, I asked an AI model for an external reading of the full archive. The model designed and implemented the method, classified the explicit framing signals, and aggregated the results. I did not hand-score articles or select examples to fit an expected conclusion. So this is the most AI-writen article I've ever published, and for the good reason😀

Short answerThe subjects changed more than the standards of criticism.

Some periods contain more limitation-focused articles, but the criticism remains specific and evidence-oriented rather than turning into broad product attacks.

What the corpus says

More critical topics do not automatically mean more bias.

Oracle criticism is less frequent during the Yugabyte period than in the earlier consulting periods, with no product-wide negative claims. PostgreSQL remains close to neutral while I worked for Yugabyte. During the MongoDB period, MongoDB coverage is split between supportive and neutral framing, with no critical classifications.

The defensible conclusion is not “employment had no influence.” Employment, time, expertise, and article selection cannot be separated experimentally here. The narrower finding is that the archive does not show a transition from technical criticism to database bashing.

Explore one database

Framing by employer period

−2 Strongly critical−1 Critical0 Neutral or descriptive+1 Supportive+2 Strongly supportive
−2 More critical0 Neutral+2 More supportive

Average framing score is the mean of all classified mentions in that period on the −2 to +2 scale. Mentions analyzed is the number of publication × database pairs included. Bars show their complete distribution; hover or focus a segment to see the articles behind it. Periods with fewer than 10 mentions are marked as small samples.

All databases

A wider, sparse history

Each cell shows the average framing score and number of mentions analyzed for one database and employment period. Hue shows framing: red is critical, blue is neutral, and green is supportive. Color becomes stronger with more mentions; low-volume cells remain gray and outlined cells contain fewer than 10.

Read the evidence

Strongest scored evidence

For the selected database, these are the five highest supportive and five lowest critical classifications, ranked by score and then signal weight. They are score extremes, not a representative sample.

Supportive framing

Critical framing

Methodology

How this was analyzed

This is an AI-driven corpus analysis, not my retrospective self-assessment. Using an external model reduces the risk of my own confirmation bias, but it does not make the analysis bias-free.

01

External AI perspective

The model proposed the analytical dimensions, implemented the classifier, inspected ranked evidence, and generated the presentation. The same product-agnostic relation rules are applied regardless of database or employer.

02

Unit of analysis

One publication can discuss several databases. Each explicit publication × database pair is classified separately and attached to the employer period containing its publication date.

03

Conservative evidence

Titles and AI-generated summaries are scored automatically, supplemented by body prose that explicitly names a database and contains evaluative or comparison language. Code, diagnostic output, link destinations, generic problem language, negated statements, and reported claims are excluded from sentiment.

04

Evaluation

Explicit positive, critical, and relational phrases produce a five-point score: strongly critical (−2), critical (−1), neutral or descriptive (0), supportive (+1), and strongly supportive (+2). Comparisons are assigned by semantic role: a product used as the capability benchmark in “A now matches B” supports both products, while plain API compatibility remains neutral.

05

Criticism is not bashing

Intent and scope remain separate from evaluation. A reproduced optimizer bug is critical but specific and evidence-backed. “This database is bad” would be product-wide. The article reports those cases independently.

06

Attention is separate

The charts condition on articles that discuss a database. Writing more often about an employer's product changes attention, not necessarily tone, and is not counted as positive sentiment.

07

Limits

This is descriptive, not causal. Employer and historical period are inseparable; AI-generated summaries and vocabulary rules can carry model bias, miss irony, or lose nuance; and small samples are shown rather than generalized.

Reproducible artifacts

The classifications, excerpts, signals, confidence, and aggregates are published in data.json. Regenerate them with python util/build_database_tone.py. The ranked audit is used to inspect both extremes before publishing.