docs: re-measure the example CRM size on the four pages that state it (31 files, 1,929 lines, ~18.5k tokens) by objectstack-fleet[bot] · Pull Request #22267 · objectstack-ai/objectstack

objectstack-fleet · 2026-10-08T10:46:32Z

Fixes #22248
Clause-②: no

What this changes

Four content pages state the size of the bundled example CRM (examples/app-crm) as 31 files, 1,792 lines, about 16k tokens, about 8% of a 200k-token window. The concepts page prints the command that reproduces the line count and invites the reader to run it ("Count it yourself"). Run against this PR's base it answers 1,929, not 1,792.

This PR re-measures the figure with the page's own command and re-derives the token and percentage estimates by the method the original figure was made with. Only the figures change; nothing else on any page does.

Page	Location	Before	After
`content/docs/concepts/metadata-driven.mdx`	table, section 4	Lines 1,792 · ~16,000 tokens · "about 8%"	Lines 1,929 · ~18,500 tokens · "about 9%"
`content/blog/context-window-is-the-constraint.mdx`	table + next sentence	Lines 1,792 · ~16,000 tokens · "about 8%"	Lines 1,929 · ~18,500 tokens · "about 9%"
`content/docs/getting-started/how-ai-development-works.mdx`	prose, "Why it's fast"	"1,792 lines across 31 files, about 16k tokens"	"1,929 lines across 31 files, about 18.5k tokens"
`content/blog/the-ontology-is-the-software.mdx`	prose, one sentence	"31 files and roughly sixteen thousand tokens"	"31 files and roughly 18,500 tokens"

The file count (31) is unchanged and still correct on all four.

The fourth page

The card names three pages. A grep of the whole tracked tree (release notes and CHANGELOGs, which record history, excluded; case-insensitive; every spelling of the line, token and percentage figures) found one more restatement of the same measurement inside content/**: the ontology post, which says "roughly sixteen thousand tokens" for the same 31 files. Left alone it would contradict the three corrected pages by 13%, so it is part of this card (nothing outside content/** carries the figures; the root README already carries the command only, with no number). No open PR touches any of the four files (10 open PRs' file lists read at base).

Measured at the PR's base

Base: origin/main 3513ac7, in this PR's worktree. origin/main has since moved to 9f0de32; the diff between the two over examples/app-crm/src and the four pages is empty, so the figures hold at the tip as well.

$ find examples/app-crm/src -name '*.ts' -not -name '*.test.ts' | xargs cat | wc -l
1929
$ find examples/app-crm/src -name '*.ts' -not -name '*.test.ts' | wc -l
31
$ find examples/app-crm/src -name '*.ts' -not -name '*.test.ts' | xargs cat | wc -c
73889

How the token and percentage figures were re-derived

Neither page states how tokens were estimated, so the method was recovered from the figure's origin rather than assumed. The page figure was first written in the commit that added the context-window post (parent tree a6058728de00816279204adb6579ba6797a156e0). Counting that tree with the same command, filters and glob gives:

a6058728: 31 files, 1,792 lines, 63,993 bytes   63,993 / 4 = 15,998  ->  "~16,000";  16,000 / 200,000 = 8%

So the method is bytes divided by 4 (the repo's own rule of thumb, as in scripts/check-skills-token-ratchet.mjs), not a fixed tokens-per-line ratio. The same method reproduces the post's other two figures at that tree: the Todo example (25 files, 58,255 bytes, 14,564, "~14.5k") and the showcase (83 files, 405,447 bytes, 101,362, "~101k").

Applied at this base:

73,889 bytes / 4 = 18,472   ->  "~18,500" (nearest 500, the precision the post already uses for "~14.5k")
18,500 / 200,000 = 9.25%    ->  "about 9%"

A fixed per-line ratio (1,929 x 16,000/1,792 = 17,223) would have understated it: the average line grew from 35.7 to 38.3 bytes.

Verification

Docs-only, no executable code. The gate list was derived from the actual diff at the head commit with node scripts/pm/dispatch-gates.mjs --commands --repo objectstack-ai/objectstack (no path arguments; 4 paths against merge base 3513ac7), every command run with its exit code recorded to a file before any pipe, and the record reconciled with --ran:

✓ dispatch-gates --ran: 40 derived famil(ies) accounted for — 40 run, 0 NOT-MEASURED (a DERIVED zero — all 40 recorded an exit code and none of them is 3).

All 40 exit 0 at head 35744bd (including check:docs-single-h1, check:doc-authoring, check:doc-anchors, check:docs-transcript-drift, check:nul-bytes, check:corpus-claim-drift, the doc-frontmatter and doc-route-spelling gates and their self-tests). Five of them (check:doc-formula-expressions, check:doc-security-posture, check:docs, check:skill-examples, check:docs-transcript-drift) first refused on a fresh worktree because their built inputs did not exist (exit 3, or exit 1 for the missing packages/spec/json-schema); the first-pass record was kept, the dependency closures were built (@objectstack/lint... and @objectstack/client-react..., under the shared verify lock) and those five were re-run and exit 0. A refusal to measure is not a finding and is not counted as a pass.

No changeset: no published package ships content/** (69 non-private package.json files scanned; none references the root content/ tree in files or scripts; the same pattern matches a known-true control string). The PR carries skip-changeset.

Acceptance notes

Sibling figures on the context-window post are stale by the same method and are not touched. The post's next paragraph says "The Todo example is ~14.5k tokens. The kitchen-sink showcase app ... 83 files of it ... is ~101k. Even that one ... leaves half a modern context window free". At this base, bytes / 4 gives Todo 27 files, 95,787 bytes, ~23.9k tokens; showcase 91 files, 695,310 bytes, ~173.8k tokens, which leaves about 13% of a 200k window, not half. The card says whether that post gets a fuller rewrite is a separate maintainer decision, and "nothing else on the pages changes", so this is reported, not edited.
The figure will drift again with every change to examples/app-crm/src. The card names no gate and none is added.
The pages do not say how tokens were estimated. Left as is under the same "nothing else changes" clause; the method is recorded above for the next person to re-derive.

Generated by Claude Code

The pages stated the bundled example CRM as 31 files, 1,792 lines, about 16k tokens, about 8% of a 200k window, and invite the reader to re-run `find examples/app-crm/src -name '*.ts' -not -name '*.test.ts' | xargs cat | wc -l`. At origin/main 3513ac7 that command answers 31 files, 1,929 lines. Tokens are re-derived the way the original figure was made: bytes / 4. The original tree (a605872) is 63,993 bytes -> 15,998 -> "~16,000"; today's is 73,889 bytes -> 18,472 -> "~18,500", and 18,500 / 200,000 is about 9%. Only the figures change: Lines, Approximate tokens, the percentage, and the same sentence on the fourth page that restated the token figure. Claude-Session: https://claude.ai/code/session_0115N1oNnQS5WqofZ2DzaT3q Co-authored-by: Claude <noreply@anthropic.com>

…dow post and correct the "half a window free" sentence (objectstack-ai#22293) Fixes objectstack-ai#22273 Clause-②: no ## What this changes One paragraph of `content/blog/context-window-is-the-constraint.mdx` (lines 49-52): the Todo and kitchen-sink showcase sizes, and the clause built on them. Nothing else on the page changes; the CRM table above it (corrected by PR objectstack-ai#22267) is untouched. ```diff -The Todo example is ~14.5k tokens. The kitchen-sink showcase app — which -deliberately exercises every metadata type in the protocol, 83 files of it — is -~101k. Even that one, the maximalist case, leaves half a modern context window -free for the actual work. +The Todo example is ~24.5k tokens. The kitchen-sink showcase app — which +deliberately exercises every metadata type in the protocol, 91 files of it — is +~174k. Even that one, the maximalist case, still fits in a 200k-token window — +about 87% of it, leaving about 13% free for the actual work. ``` **The argument is kept; one clause is not.** The paragraph argues that even the maximalist example fits in one window. At ~174k of the 200k-token window the post already uses for the CRM paragraph two lines above, it still does, so that claim stays. "Leaves half a modern context window free" does not survive the new numbers (about 13% is free, not half), so it is replaced with the measured fraction. "A modern context window" became the explicit 200k so the CRM paragraph's "about 9%" and this paragraph's "about 87%" share one denominator. No other sentence on the page depends on the half-window claim; the rest of the post argues from the CRM example. ## Measured at this PR's base Base: `origin/main` 5870716 (the merge base, read in this PR's worktree). Method: bytes of the non-test `.ts` files under the example's `src/`, divided by 4, as recovered on objectstack-ai#22248 and applied in PR objectstack-ai#22267. No new method and no new gate (triage ruling on the card). ```text $ find examples/app-todo/src -name '*.ts' -not -name '*.test.ts' | wc -l 27 $ find examples/app-todo/src -name '*.ts' -not -name '*.test.ts' | xargs cat | wc -c 97653 $ find examples/app-todo/src -name '*.ts' -not -name '*.test.ts' | xargs cat | wc -l 2383 $ find examples/app-showcase/src -name '*.ts' -not -name '*.test.ts' | wc -l 91 $ find examples/app-showcase/src -name '*.ts' -not -name '*.test.ts' | xargs cat | wc -c 696149 $ find examples/app-showcase/src -name '*.ts' -not -name '*.test.ts' | xargs cat | wc -l 14229 ``` A second reading through git, not the filesystem, agrees on every file count and byte total: `git ls-tree -r -l 5870716 -- examples/APP/src`, summing the sizes of files ending `.ts` and not `.test.ts`. | example | files | bytes | bytes / 4 | the post now says | |:--|--:|--:|--:|:--| | `app-todo` | 27 | 97,653 | 24,413 | ~24.5k | | `app-showcase` | 91 | 696,149 | 174,037 | 91 files, ~174k | Rounding follows the precision the post already uses, as PR objectstack-ai#22267 recorded it: the Todo figure to the nearest 500 (the post's "~14.5k"), the showcase figure to the nearest 1k (the post's "~101k"). 174,037 / 200,000 = 87.0%, so about 87% used and about 13% (25,963 tokens) free. Control for the method: the same count over the tree the figures were first written on (`a6058728de`) gives `app-todo` 25 files / 58,255 bytes / 14,564 and `app-showcase` 83 files / 405,447 bytes / 101,362, i.e. the post's original "~14.5k", "83 files" and "~101k". The card's readings at `8cbe255ef` also reproduce exactly (Todo 95,787 bytes / 23,947; showcase 695,310 bytes / 173,828). Between that commit and this base `app-todo` grew by 1,866 bytes (+1.9%) and `app-showcase` by 839 (+0.1%): the figures move with `main`. The Todo rounding band (24,250 to 24,750 tokens, i.e. 97,000 to 99,000 bytes) holds the current reading at 97,653. The old sentence has no other copy: `git grep` for "half a modern context window" finds nothing in the tree after this change, and a grep of the tracked tree for the old figures ("~14.5k", "~101k", "83 files"; release notes and changelogs excluded) found them only in this paragraph; the other pattern hits are unrelated test-shard and vitest comments. ## Verification Docs-only. The gate list was derived from the actual diff with `node scripts/pm/dispatch-gates.mjs --commands --repo objectstack-ai/objectstack` (no path arguments; 1 path against merge base 5870716, 8 changed lines), every command run with its exit code recorded to a file before any pipe, and the record reconciled with `--ran`: ```text ✓ dispatch-gates --ran: 15 derived famil(ies) accounted for — 15 run, 0 NOT-MEASURED (a DERIVED zero — all 15 recorded an exit code and none of them is 3). ``` All 15 exit 0 at head bad1420 (including `check:doc-frontmatter` and its self-test, `check:doc-anchors`, `check:doc-authoring`, `check:nul-bytes`, `check:closing-keyword-parity` and its self-test). One of them, `pnpm --filter @objectstack/lint run check:doc-formula-expressions`, first refused on the fresh worktree with exit 3 (PREREQUISITE NOT MET: `@objectstack/formula` and `@objectstack/lint` were not built). That first pass measured nothing and is not counted as a pass; the two closures were built under the shared verify lock (`turbo run build --filter=@objectstack/formula --filter=@objectstack/lint --concurrency=2`, VERDICT command-exit 0) and the gate was re-run: exit 0. No changeset, `skip-changeset` on the PR: no published package ships `content/**`. Of the 82 tracked `package.json` files, 69 are non-private and all 69 declare a `files` array; none of those entries reaches `content/` or `blog/`, and the same scan counts 69 entries naming `dist` as its positive control. No governed surface is touched (`content/blog/**` is not in the register). ## Acceptance notes - **The numbers will drift again.** Every figure in the paragraph is a hard-coded reading of two example apps that change on `main` daily; the card names no gate and none is added. - **The CRM row beside it already reads 3 lines high.** The table's "Lines | 1,929" (set by objectstack-ai#22267, measured at its base `3513ac778`) answers 1,932 at this base by the post's own `find ... | xargs cat | wc -l` command (`app-crm`, 31 files, 74,182 bytes, ~18,545 tokens); `app-crm/src` changed between objectstack-ai#22267's base and its merge commit `c8bb3c8d9` (73,889 bytes then, 74,182 at the merge commit), and the bytes are identical at the merge commit and at this base. The tokens and "about 9%" figures still hold. Not edited here: the CRM row is carved out of this card. Same drift class the card already declares; noted, not filed. - **What the post now claims.** The card asked whether a numbers-only fix changes what the post claims: it does, from "half a window left over" to "about 13% left over". The triage ruling (rewrite the sentence to whatever the new numbers support) is what this PR does; whether the post wants a fuller rewrite of its framing is not decided here. --- _Generated by [Claude Code](https://claude.ai/code/session_0115N1oNnQS5WqofZ2DzaT3q)_ Co-authored-by: Claude <noreply@anthropic.com>

github-actions Bot added size/s documentation Improvements or additions to documentation labels Oct 8, 2026

objectstack-fleet Bot added the skip-changeset PR has no user-facing published change; bypasses the changeset gate label Oct 8, 2026

objectstack-fleet Bot assigned os-sales Oct 8, 2026

objectstack-fleet Bot marked this pull request as ready for review October 8, 2026 10:59

objectstack-fleet Bot enabled auto-merge October 8, 2026 10:59

objectstack-fleet Bot added this pull request to the merge queue Oct 8, 2026

Merged via the queue into main with commit c8bb3c8 Oct 8, 2026
38 checks passed

objectstack-fleet Bot deleted the claude/issue-22248-crm-size-drift branch October 8, 2026 11:17

objectstack-fleet Bot mentioned this pull request Oct 8, 2026

[PM seat] domain:devx @ objectstack — 🟢 zhuangjianguo · session_012zwsLMxQBwHUYJgPyrpEvx #6023

Open

github-actions Bot mentioned this pull request Oct 8, 2026

hourly full run: red on main (CI) #22292

Closed

objectstack-fleet Bot mentioned this pull request Oct 8, 2026

docs(blog): re-measure the Todo and showcase sizes in the context-window post and correct the "half a window free" sentence #22293

Merged

Navigation Menu

Repository navigation

docs: re-measure the example CRM size on the four pages that state it (31 files, 1,929 lines, ~18.5k tokens) - #22267

docs: re-measure the example CRM size on the four pages that state it (31 files, 1,929 lines, ~18.5k tokens)#22267
objectstack-fleet[bot] merged 1 commit into
mainfrom
claude/issue-22248-crm-size-drift

objectstack-fleet Bot commented Oct 8, 2026

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

Conversation

objectstack-fleet Bot commented Oct 8, 2026

What this changes

The fourth page

Measured at the PR's base

How the token and percentage figures were re-derived

Verification

Acceptance notes

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants