Archive

Browse the complete record or search for a topic. Entries follow the event or original note date. Reconstructed accounts are labeled “event” and show their actual publication date on the article.

October 2026

17 Oct
Deep Research Needs an Evidence Budgetapplied research note
16 Oct
Merging LoRA Removes a Branch, Not Parametersresearch note
14 Oct
A Test Set Cannot Choose Its Own Thresholdresearch note
09 Oct
A Distillation Weight Needs Its Temperatureresearch note
07 Oct
Perplexity Has a Tokenizer in Its Denominatorresearch note
05 Oct
NFC Cannot Fix the Wrong Diacriticengineering note
02 Oct
Three Judge Calls Are Still One Itemresearch note

September 2026

30 Sep
Count the Raster After Decodingengineering note
28 Sep
A Ranking Metric Cannot Choose a Thresholdresearch note
25 Sep
The Perceptron Convergence Bound Has Unitsresearch note
24 Sep
One Field Is Not a List Operationengineering note
21 Sep
Test the State Space, Not the Noiseengineering note
20 Sep
A Holdout Is a Set of Equivalence Classesresearch note

August 2026

27 Aug
Defining a Reproducible Arithmetic Experiment in CalcGPTengineering note
27 Aug
When Noisy Text Changes the Winning Modelresearch note
27 Aug
Change One Slot, Watch What Else Movesresearch note
27 Aug
Ten Thousand Adaptation Steps Could Not Remove One Scraper Artifactresearch note
event 27 Aug
The Panel Was Weak on Items and Useful for Ranking Systemsresearch note
27 Aug
Three Emulator Bugs, Three Different Testsengineering note

May 2026

event 18 May
A Defensive Ollama Client for Local Judge Inferenceengineering note
event 02 May
The Interval Belongs to the Comparisonresearch note

April 2026

14 Apr
The Artifact Chain Behind TinyFabulistresearch note

March 2026

17 Mar
What Must Be Fixed Before the Diacritics Comparisonmethod note

February 2026

24 Feb
Decision Gates for an Open-Weight Judge Panelmethod note

January 2026

event 15 Jan
TF3-RO: Training and Compressing a Romanian Model from Scratchrelease note

December 2025

08 Dec
Designing a 51M-Parameter Romanian Model from Scratchengineering note

October 2025

27 Oct
One Year In: The Decisions Behind the Outputretrospective
event 27 Oct
What the Diacritics Study Measured at InnoComp 2025research note

September 2025

29 Sep
Romanian Diacritics: The Errors Hidden by Overall Accuracyresearch note
event 09 Sep
TF2: Open Models for English–Romanian Literary Translationrelease note

August 2025

04 Aug
What LoRA Changes in the TF2 Translation Systemengineering note

July 2025

event 26 Jul
A Logistic-Regression Baseline Worth Keepingretrospective
07 Jul
When BLEU and a Literary Rubric Answer Different Questionsresearch note

June 2025

09 Jun
Why TF2 Chose Translation Before Romanian-Native Generationresearch note

May 2025

12 May
Treating LLM Judges as Measurement Instrumentsresearch note

April 2025

event 29 Apr
Three Million Stories Are Still a Sparse Sampleresearch note
event 29 Apr
TF1-EN-3M: From Six Slots to Three Million Fablesrelease note

March 2025

event 18 Mar
A Map of LLM-Generated Synthetic Datarelease note
10 Mar
What Breaks When Fable Generation Reaches Millionsengineering note

February 2025

03 Feb
Why TinyFabulist Starts with a Specificationresearch note

January 2025

07 Jan
The 546× Speedup That Wasn't One Numberretrospective
04 Jan
A Perceptron You Can Auditretrospective

December 2024

09 Dec
Papers That Changed the TinyFabulist Designretrospective

October 2024

28 Oct
Three Decisions at the Start of the PhDretrospective