<- all tokdocs
2026.10.19@reporadarai

Jantt's end card says `npm install @jantt/core`, but no package named `jantt` has ever been published to npm, and the "zero runtime dependencies" core declares fflate in its package.json

Coding
Watch on TikTok

View on TikTok ->

A 35.83 s vertical clip at 1080x1920, MP4 container (mov,mp4,m4a,3gp,3g2,mj2), HEVC Main profile video at yuv420p and 506,152 bps over 1,075 frames at a constant 30 fps, HE-AACv2 audio at 44.1 kHz stereo and 48,113 bps, 561,203 bps overall, 2,513,724 bytes on disk, format id bytevc1_1080p_561193-1; posted at epoch 1792407600, which is 2026-10-19 11:00:00 UTC, by uploader reporadarai (uploader id 6692521761513571333) under the channel name "Repo Radar AI", audio credited as "original sound - Repo Radar AI", with a 308-character description that names the project, the MIT-adjacent pitch and five hashtags (#aitools #criticalpath #dag #visualization #gantt); at the metadata capture time of 2026-10-27 05:07:01 UTC the post showed 2,031 views, 88 likes, 1 comment and 59 reposts. I read ten of the eighteen extracted frames (001, 003, 005, 007, 009, 012, 013, 015, 017, 018) covering roughly 0 s, 4 s, 8 s, 12 s, 16 s, 22 s, 24 s, 28 s, 32 s and 34 s, against a transcript of 81 words in 6 segments. The video is a single continuous scroll of a rendered GitHub README inside a fake macOS browser chrome with a "Repo Radar AI" title-bar watermark, with burned-in karaoke captions overlaid. Frame 001 shows the Jantt wordmark and logo, a pink pill reading "★ 1.1k stars", and the caption "JSON → INTERACTIVE GANTT". Frame 003 shows the subtitle "The Declarative JSON Gantt Chart Engine", the tagline "Where human intuition meets declarative AI planning.", and six shields.io badges reading License MIT, TypeScript 5.4, Dependencies 0 Runtime, Tests 339/339 Passing, Node.js >=18.0.0 and PRs Welcome, above the line "Zero runtime dependencies. 100% pure TypeScript." Frame 005 (caption "GANTT CHARTS WITH") shows a screenshot of the playground: a JSON editor pane on the left with "$schema": "https://jantt.dev/schema/v1.json" and a "Metropolis High-Rise Construction" plan, a Gantt canvas on the right with right-angled connector lines, a "Critical Path (7)" toggle, and a hover card reading "Structural Steel Core (Floors 1-25) / Timeline Nov 14, 2026 → Jan 10, 2027 (57d) / Progress 15% / Depends On piling-pour". Frame 007 (caption "DEPENDENCIES AND WEIGHS") lists four "superpowers" including "The core math and layout engine (@jantt/core) is pure, zero-dependency TypeScript weighing under 14 KB gzipped." Frame 009 (caption "AI MODELS CAN") shows "Swiss Dark Mode" and a "Color Theme Gallery" with Cyber Emerald, Midnight Rose, Sunset Crimson and Master Specification Benchmark tiles. Frame 012 shows a four-way comparison table against DHTMLX Gantt, Frappe Gantt and Mermaid.js whose last row reads "Bundle Size (gzip) | < 14 KB | ~200 KB | ~35 KB | ~800 KB". Frames 013 and 015 (captions "AGAINST DARK BACKGROUNDS," and "CAN YOU SEE") show the quickstart blocks npm install @jantt/react @jantt/core and <script src="https://unpkg.com/@jantt/standalone"></script>, plus a stat strip reading "10× Fewer LLM Tokens vs JSX | 0 Runtime Dependencies | 100% Deterministic DAG Solver | 2-Way Bidirectional State Sync". Frames 017 and 018 are the end card: "Can you see an AI agent generating your next project schedule automatically with Jantt?", "FIND IT AT github.com/AhmadHassan-BTed/Jantt", "★ 1,118 STARS", "$ npm install @jantt/core" and a "@RepoRadarAI Follow for more open-source projects" bar. No terminal, no shell session, no npm install, no browser devtools and no bundle-size measurement appears in any frame I read; every interaction shown is a static screenshot embedded in the README, not a live demo. The spoken argument is that structured JSON replaces hand-written timeline code, that the core engine ships zero runtime dependencies under 14 KB compressed, and that LLMs emit timeline data in roughly a tenth of the tokens JSX would need. On the transcript-end check, the last segment ends at 33.54 s against a real audio duration of 35.79 s in audio.mp3 and 35.81 s in the muxed AAC stream, so every segment lands inside the audio and nothing in this transcript is a past-the-end Whisper fabrication.

The product name is Jantt, and Whisper dropped a letter the first time it heard it

The transcript is internally inconsistent about the name. At 5.82 s the second segment reads "Jant transforms structured JSON into interactive Gantt charts with draggable tasks and timelines", with one t. At 28.90 s the closing segment reads "Can you see an AI agent generating your next project schedule automatically with Jantt?", with two. The burned-in text settles it: frame 001 renders the wordmark as "Jantt", and the end card in frame 018 spells out the repository path as github.com/AhmadHassan-BTed/Jantt. The GitHub API confirms the canonical casing and spelling: the repo full_name is AhmadHassan-BTed/Jantt. The one-letter drop at 5.82 s is a Whisper artifact, not a second product. Anyone searching for "Jant" from the audio alone lands nowhere.

The repository itself checks out on every structural fact the video and its README assert about provenance. The GitHub API reports the description "The AI-native JSON Gantt chart engine. Dependency-free interactive timelines with React, script-tag, CLI, topological DAG solver, and formal JSON schema", language: TypeScript, license.spdx_id: MIT, created_at: 2026-08-28T07:32:35Z, archived: false, fork: false, and topics including gantt, json-schema, critical-path, dag and zero-dependencies. The MIT badge in frame 003 is accurate. The project is 60 days old as of the capture date and the latest push is 2026-10-26T09:16:19Z, so it is actively maintained. There is a second copy of the repo at gustavoml12/Jantt carrying the same description; the end card points at AhmadHassan-BTed/Jantt, which is the non-fork original by created_at and by the fork: false flag.

The install command on the end card does not resolve: npm has no jantt packages at all

Frame 018 displays $ npm install @jantt/core as the call to action, and the README quickstart in frames 013 and 015 gives npm install @jantt/react @jantt/core plus a CDN path at https://unpkg.com/@jantt/standalone. None of those work today. Querying the npm registry on 2026-10-27 returns HTTP 404 with body {"error":"Not found"} for @jantt/core, @jantt/react, @jantt/standalone and the unscoped jantt. A registry search for the text jantt returns {"total": 0}. The unpkg URL in the script-tag snippet returns 404 as well, because unpkg resolves against the same registry.

The repository explains why. The publish workflow at .github/workflows/release.yml is named "Publish & Release" and is triggered only on: push: tags: - "v*.*.*". The GitHub API reports zero releases and zero tags on the repository, and the release workflow's run list returns total_count: 0. The workflow has never fired, so nothing has ever been pushed to the registry. The CLI package at cli/package.json declares "name": "jantt" with "bin": {"jantt": "./bin/jantt.js"}, and the README advertises npx jantt open; that command fails for the same reason, since npx resolves jantt from npm and npm returns 404.

The practical effect is that a viewer who follows the video's instruction gets npm error 404 Not Found - GET https://registry.npmjs.org/@jantt%2fcore. The only working path to the code on 2026-10-27 is cloning the repository and building it from source, which the README does cover in its contributing section. The video never mentions that step.

"Zero runtime dependencies" is contradicted by @jantt/core's own package.json

At 11.08 s the video says "Its core engine has zero runtime dependencies and weighs under 14 kilobytes compressed." Frame 003 renders the matching badge Dependencies 0 Runtime, and frame 007 renders the README bullet "The core math and layout engine (@jantt/core) is pure, zero-dependency TypeScript weighing under 14 KB gzipped."

The package manifest says otherwise. packages/core/package.json on main ends with:

  "dependencies": {
    "fflate": "^0.8.3"
  }

That is a declared runtime dependency, not a devDependency. It is used, not vestigial: packages/core/src/share-coder.ts opens with import { deflateSync, inflateSync, strToU8, strFromU8 } from "fflate";, and packages/core/src/index.ts re-exports that module twice, once via export * from "./share-coder"; and again by name for compressPlanToUrlPayload, decompressPlanFromUrlPayload, cleanPlanForSharing and sanitizePlanForJson. Any consumer importing the package entry point pulls fflate into the graph.

The commit history dates the change precisely. git log on packages/core/package.json shows three commits: a377861d on 2026-08-28 ("feat: initial release of Jantt"), b1537d68 on 2026-08-29, and a349c2a9 on 2026-09-05 with the message "fix(share): compress shareable links with deflate-raw and enable dynamic live-update sync". The dependency was added 44 days before this video was posted, so the claim was already stale when the clip went up.

One nuance keeps this a partial correction rather than a flat one. packages/core/vite.config.ts builds the library with rollupOptions that set only output.assetFileNames and declare no external array, so fflate is inlined into dist/index.js and dist/index.cjs at build time. A consumer of the built bundle therefore loads a single artifact with no separate fetch. The accurate statement is that the core engine vendors one third-party compression library rather than shipping none, and the shipped bundle size includes it. The badge reading Dependencies 0 Runtime is a hardcoded shields.io image, not a generated count.

The "under 14 KB gzipped" figure cannot be checked from the repository, and nothing enforces it

The 14 KB number appears three times in the source material: the audio at 11.08 s, the README superpower bullet in frame 007, and the comparison table in frames 012 and 013, where the "Bundle Size (gzip)" row reads < 14 KB for Jantt against ~200 KB for DHTMLX Gantt, ~35 KB for Frappe Gantt and ~800 KB for Mermaid.js.

I could not verify it. The measurement would have to come from a published artifact or a size check in CI, and neither exists. The npm packages are unpublished, so there is no tarball to gzip. packages/core/package.json lists "files": ["dist"], and dist is a build output that is not committed to the repository. .github/workflows/ci.yml runs exactly four steps after setup, npm ci, npm run build, npm run typecheck and npm test, with no size-limit, bundlewatch or byte-count assertion anywhere in the job. Nothing in the project fails if the bundle crosses 14 KB.

The comparison figures for the three competitors are likewise uncited in the README; they are typed into a markdown table with tilde approximations and no link to a measurement. I am not disputing the rough magnitudes, which are plausible for those libraries, but none of the four numbers in that row traces to a source I can fetch. Treat the whole row as author-asserted. The claim that the Jantt core is small is believable for a dependency-light TypeScript layout engine, and the specific threshold of 14 KB gzipped is unverified.

"10x fewer tokens than JSX" is a slogan repeated six times with no benchmark behind it

At 16.76 s the video says "AI models can generate complete timeline data using 10 times fewer tokens than JSX." The README states the same ratio in six separate places: the audience bullet at line 29, the superpowers bullet at line 56, the stat strip at line 177 ("Fewer LLM Tokens vs JSX"), the Mermaid pipeline diagram at line 185 ("10x fewer tokens, type-safe, 0 hallucinations"), the pipeline prose at line 192 ("Uses 10× fewer tokens than JSX"), and the capability table at line 507, which renders it in LaTeX as "$10\times$ token reduction vs JSX/HTML, zero graphical hallucinations".

Searching the README for any token count attached to that ratio returns nothing. There is no before-and-after pair of prompts, no tokenizer named, no input and output token totals, no model specified for the measurement. The nearest thing to a methodology is line 507's reference to "Copy-paste benchmark prompts for Claude 3.5 Sonnet, GPT-4o, and Gemini 1.5 Pro", which supplies prompts rather than results. The repository does ship a real schema contract at schema/jantt.schema.json, a draft-07 JSON Schema titled "Jantt Plan Specification v1", so the underlying mechanism (feed the model a schema, get JSON back, let the library do layout) is implemented. The quantified 10x claim is the part with nothing behind it. Mark it unverified.

The schema file carries a second small problem. Its $id is https://jantt.dev/schema/v1.json, and every JSON example in the README and in the playground screenshot at frame 005 uses that URL as the $schema value. The domain jantt.dev does not resolve: dig jantt.dev A returns no records and dig jantt.dev NS returns no delegation, while a control query for github.com returns 140.82.114.4 from the same resolver. Any tooling that tries to dereference $schema from a Jantt plan fails. The schema is reachable only at its raw GitHub path, which the README does give alongside the broken one.

The green badges are static: CI has been red since 22 October and the live count is 341 of 343

Frame 003 shows a Tests 339/339 Passing badge linking to the Actions tab, and the README repeats "339/339 tests passing across 22 suites" at line 547 and again at line 817. The badge is a hardcoded img.shields.io/badge/Tests-339%2F339%20Passing-brightgreen URL, so it displays that text regardless of CI state.

The actual state on 2026-10-27 is a failing pipeline. CI run 40298259947, triggered by the 2026-10-26 push, reports conclusion: failure. Its job log shows Install Dependencies, Build All Packages and TypeScript Typecheck all green, then Run Test Suite failing with Test Files 1 failed | 21 passed (22) and Tests 2 failed | 341 passed (343). Both failures are in packages/core/tests/reconciler.test.ts, in the cases "cleanly prunes dangling dependencies when predecessor tasks are deleted" and "performs autonomous JSON maintenance: prunes expired tombstones, caps bloat, strips empty timestamps", each raising AssertionError: expected undefined not to be undefined. The log also carries two runtime errors from the fflate code path, Failed to decompress plan from URL payload: DOMException [InvalidCharacterError] and Failed to decompress plan from URL payload: Error: unexpected EOF.

The timeline matters for fairness to the video. Filtering CI runs to the main branch shows an unbroken sequence of success from 2026-09-06 through 2026-09-18, then nothing until 2026-10-22, which failed. The video posted on 2026-10-19, between those two runs, when the last recorded main-branch result was green. The badge number was wrong on 2026-10-19 (the suite had grown past 339 and the figure is frozen), but the pipeline was not yet red. It has been red since. The CI matrix also runs Node 20.x and 22.x only, while the Node.js >=18.0.0 badge in frame 003 promises 18; the root package.json sets "engines": {"node": ">=20.0.0"}, which contradicts the badge directly.

On stars, the video is a snapshot that has aged. Frame 001 shows "★ 1.1k stars" and frame 018 shows "★ 1,118 STARS" as of the 2026-10-19 post. The GitHub API on 2026-10-27 returns stargazers_count: 4620 with 15 forks, 13 open issues and 1 watcher. The repository gained roughly 3,500 stars in the eight days after this video, on a project that is 60 days old. I could not pull per-star timestamps to confirm the 1,118 figure at post time, since the stargazers endpoint returned 401 and 404 for my requests, so I report the 1,118 number as what the end card displayed rather than as independently verified.

Key Takeaways

  • Verified: The repository is real. AhmadHassan-BTed/Jantt is MIT licensed, written in TypeScript, created 2026-08-28, not archived, not a fork, and last pushed 2026-10-26, per the GitHub REST API.
  • Correction: The end card instructs npm install @jantt/core. The npm registry returns HTTP 404 for @jantt/core, @jantt/react, @jantt/standalone and jantt, and a registry search for "jantt" returns zero results. The publish workflow fires only on v*.*.* tags, and the repo has zero tags, zero releases and zero release-workflow runs. The packages have never been published.
  • Correction: At 11.08 s the video says "Its core engine has zero runtime dependencies." packages/core/package.json declares "dependencies": {"fflate": "^0.8.3"}, imported at the top of src/share-coder.ts and re-exported from src/index.ts. The dependency was added in commit a349c2a9 on 2026-09-05, 44 days before the post.
  • Partial correction: packages/core/vite.config.ts sets no external list, so fflate is inlined into the built bundle and a consumer of dist fetches one artifact. The accurate framing is one vendored compression library rather than none.
  • Unverified: "Under 14 kilobytes compressed" (11.08 s) and the < 14 KB comparison row. No package is published, dist is not committed, and ci.yml contains no bundle-size check. The competitor figures (~200 KB, ~35 KB, ~800 KB) carry no citation either.
  • Unverified: "10 times fewer tokens than JSX" (16.76 s). The README repeats the ratio six times and supplies no token counts, no tokenizer and no measured comparison.
  • Correction: The Tests 339/339 Passing badge is a hardcoded shields.io image. CI run 40298259947 on 2026-10-26 reports 2 failed and 341 passed across 22 suites, both failures in packages/core/tests/reconciler.test.ts. Main-branch CI was green through 2026-09-18 and first failed on 2026-10-22, three days after this video posted.
  • Clarification: The Node.js >=18.0.0 badge conflicts with the root package.json, which sets "engines": {"node": ">=20.0.0"}, and with ci.yml, which tests only Node 20.x and 22.x.
  • Clarification: The schema's $id and every README example point at https://jantt.dev/schema/v1.json. The domain jantt.dev has no A record and no NS delegation, so that URL cannot be dereferenced. The schema is reachable at its raw GitHub path.
  • Context: Whisper rendered the name as "Jant" at 5.82 s and "Jantt" at 28.90 s. The burned-in wordmark and the end-card repository path both read Jantt.
  • Context: The video shows no terminal, no install, no live drag interaction and no measurement. Every frame is a scroll of the rendered README with embedded screenshots.
  • Context: The end card showed 1,118 stars on 2026-10-19. The GitHub API returned 4,620 on 2026-10-27, roughly 3,500 added in eight days.

Resources

Published October 19, 2026. Writeup generated from a favorited TikTok.