chore(classifier): reconstruct wall-clock for v0.2.0 collection run#2
Merged
Merged
Conversation
…lection run
The author's machine still had /tmp/mnemoscope-full.log from the original
collection — its created/modified timestamps bound the wall-clock end-to-end:
started_at: 2026-04-29T04:11:20-04:00
ended_at: 2026-04-30T03:13:50-04:00
wall_clock_s: 82950 (23h 02m 30s, ~207 s/cell over 400 cells)
Both fields are flagged as *reconstructed* from the log mtime in
measurements-meta.json#_note and model.json#dataset_collection.wall_clock_human;
they bound the orchestrator's elapsed time but include any idle gaps between
cells, so they are not equivalent to a sum of per-cell durations.
Token totals stay null. The OpenAI-compat usage block was not persisted
during this run, and inferring tokens from the haystack target sizes alone
would be a guess, not an audit. b33b5dc already wired
research/replication/run.py to capture tokens_in / tokens_out for every
future cell.
train.py re-ran on the unchanged measurements.csv with the same seed (42),
re-selected rf at R²=0.5827 / MAE=0.1386, and re-emitted an ONNX file
identical (byte-for-byte) to the published one. Only model.json changed,
to embed the updated measurements-meta.json under dataset_collection.
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
The author still had
/tmp/mnemoscope-full.logfrom the original v0.2.0 collection — its created/modified timestamps give a tight bound on the wall-clock end-to-end. Both fields are flagged as reconstructed from the log mtime, not measured by the collector, inmeasurements-meta.json#_noteandmodel.json#dataset_collection.wall_clock_human.started_at2026-04-29T04:11:20-04:00ended_at2026-04-30T03:13:50-04:00wall_clock_s82950(23h02m30s)wall_clock_s_grading_onlyand the token totals staynull— they would be guesses, not audit. b33b5dc already wired the collector to record both for every future run.Test plan
train.pyre-ran withseed=42on the unchangedmeasurements.csv, re-selectedrfat R²=0.5827 / MAE=0.1386model.onnxis byte-identical to the published one (sklearn is deterministic with the same seed)model.json#dataset_collectionnow reads the updated meta sidecar (embedded dataset_collection metadata from classifier/measurements-meta.jsonprinted)🤖 Generated with Claude Code