feat: drop the focus doc's tracking tables and the impact log's older rows - #69
Draft
Bhekani Khumalo (bhekanik) wants to merge 4 commits into
Draft
feat: drop the focus doc's tracking tables and the impact log's older rows#69Bhekani Khumalo (bhekanik) wants to merge 4 commits into
Bhekani Khumalo (bhekanik) wants to merge 4 commits into
Conversation
… rows Two product decisions, both prompt-only. Neither file on disk is touched, and the writers and the cleanup still read them whole. `My Focus.md` is the engineer's own document and the coach reads all of it, so it stays whole apart from one thing: a markdown table with more than twenty data rows is replaced by a line saying how many rows it had. On the real vault that document is 55k and one 128-row tracking table is 44k of it. The priorities are in the prose and the headings; the table is a spreadsheet that happens to live in a markdown file. Headings, prose and small tables survive verbatim. The impact log now reaches back a year from the week being generated, not from today, so regenerating an old week sees what that week could have seen. Older rows become one line counting them. The coach's job with this file is the gap analysis — how long since the last significant impact, is the pace healthy — and a year of rows answers that; the older ones have already been promoted into a brag book. The last-impact and current-gap lines the schema depends on are not dated rows and pass through untouched. Both walk tables the way the memory trim does: fence-aware through `markdown-scan.ts`, CRLF preserved, kept rows passed through byte for byte rather than re-rendered. The table finder they share treats a run of pipe lines with no separator as prose, so it cannot eat one. `--verbose` gains a line counting what each left out, so a thin section is never a mystery. On fixtures shaped like the real files: focusDoc 24,172 to 925, impact 16,041 to 5,554.
Both survived mutation. The undated-row test used a value that sorts above any ISO date, so dropping undated rows happened to keep it; it now uses one that sorts below. And nothing checked that the window is measured from the week being generated rather than from today, which matters for regenerating an old week; the dropped count is now compared against the fixture's own rows.
Only the lower bound was applied, so regenerating an old week showed it achievements from two years after the fact, and the two status lines walked through untouched telling a 2024 week what its author's last impact would eventually be. That is the history rule broken in the one place the coach is asked to reason about time. The window is now `[week start - 52 weeks, week end]`. Rows past the upper bound are removed without comment, because to that week they had not happened; there is no count for them, since a count would itself leak that something came later. Rows below the lower bound stay a count, as before. Both status lines are recomputed from every row dated on or before the week's end, including rows too old to be listed: an engineer whose last impact is fourteen months back has a fourteen-month gap, not "none recorded". The gap is measured from the week's end or from today, whichever is sooner, so the current week reads exactly as it does now. `gapText` and the status vocabulary are imported from `vault-updates.ts` rather than reimplemented. Three more from the same review: `findTables` recorded only the first separator in a run of pipe lines, so a small table written directly under a large one was swallowed and its rows went with it. A table now starts at every header-and-separator pair. Dates are validated with the round-tripping `isIsoDate` from `vault-updates.ts`. The local copy checked the shape only, so `2024-13-01` counted as a date, was dropped as too old, and could have won the latest-impact comparison. Tables written without their outer pipes, and tables indented up to three spaces, are both found now. A line is normalized once, then handed to the writers' own `isTableSeparator` and `scanRow` so this file does not keep a second copy of their rules. Four spaces still means an indented code block.
Three mutations survived. The four-space guard was covered only through the impact trim, where stripping it shifts the date column and nothing is dropped anyway; the focus trim counts rows instead, so it notices. And nothing built a prompt for a historical week, so neither the upper bound nor the gap basis was pinned at the builder level.
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.
What
Two prompt-only trims. Neither file on disk is touched, and the writers and the cleanup still
read both whole.
Measured on fixtures shaped like the real files. The fifteen sections still sum to exactly the
prompt length.
1. Focus doc tables
My Focus.mdis the engineer's own document and the coach reads all of it, so it stays wholeapart from one thing: a markdown table with more than twenty data rows is replaced by
_(table of N rows omitted from the coaching prompt)_. On the real vault that document is 55k anda single 128-row tracking table is 44k of it — the priorities are in the prose and the headings,
and the table is a spreadsheet that happens to live in a markdown file.
Headings, prose, and tables of twenty rows or fewer survive verbatim.
2. Impact log window
The prompt gets impact rows dated within 52 weeks of the week being generated, not of today,
so regenerating an old week sees what that week could have seen. Older rows become one
_(N older entries not shown)_line after the table.The coach's job with this file is the gap analysis — how long since the last significant impact,
is the pace healthy — and a year of rows answers it; the older ones have already been promoted
into a brag book.
**Last significant impact:**and**Current gap:**are not dated rows and passthrough untouched, along with the heading and the table header.
How both work
Both walk tables the way the memory trim does: fence-aware via
markdown-scan.ts, CRLF preserved,kept rows passed through byte for byte rather than re-rendered. They share one table finder, which
treats a run of pipe lines with no separator as prose so it cannot eat one.
--verbosegains a third line:Verification
bunx tsc --noEmitclean,bun run lintclean,anti-slop-checkclean on changed lines.bun run test— 839 passing across 25 files (818 before).Note on the fixture
The focus doc fixture is 24k against the real 55k. The shape is what matters here — one table
holding most of the bytes — and the trim removes it either way; the real saving will be larger.