中文
Dekita2 products

Small, honest tools. Built by one developer, in the open.

Also from Dekita

Early access New

InvoiceLens

Sort invoices and receipts on your own machine. No upload, no account, no cloud.

Read more →

Content Automation Pipeline — Verifiable Output

Everything on this page is real output from an unattended run at 2026-10-03 07:00. No beautifying, no padding, no invented numbers.

What makes this page different: every number below is labelled with its source file and the first 16 characters of the file's SHA256. You can recompute the hashes yourself, and you can ask to see the original files in full. A number that cannot be verified on the spot does not go on this page.

1 · What it actually does

Give it one industry topic. Within 24 hours, unattended, it produces one news brief, three platform-specific publishing cards, three cover images, and one plain-text delivery file. Every step passes five quality gates. If any gate fails, nothing is published.

400words in the brief
1601words in the long-form piece
23daily self-check items
5pre-publish quality gates

Word counts are measured by the scripts, not estimated.

2 · Real output

Cover images are rendered by code, not by a generative image model — because image models get CJK text wrong. That is a technical decision, not a stylistic one.
Long-form article cover, 760x427
Long-form cover 760×427
sha256:d15c82209e1e598d
Social cover, 560x746
Social cover 560×746
sha256:b10b1d081c02d789
Topic-article cover, 760x427
Topic-article cover 760×427
sha256:38935e898111d964

The three images are right here on this page — they are not placeholders. The dimensions and the first 16 characters of the SHA256 shown above are computed by a script from the exact bytes this page serves — download the image and recompute the hash yourself. Full SHA256 values ship with the handover.

⚠ Sample output — a fictional scenario, not a report
Read the sample →
One topic, delivered in Chinese / English / Japanese — this sample demonstrates the delivery format only.
Everything in the sample story — the shops, the dates, the dialogue — is invented to show the format. It is not a factual account, has not been verified, and has not been published on any platform. (Platform policies referenced inside it are taken from public policy text; the narrative around them is fiction.) The numbers on this page — word counts, image hashes, gate results — are the verifiable part and are measured by scripts.
The publishing cards also ship as a plain-text file (15,123 bytes), so you can paste them directly instead of retyping from an image.

3 ⭐ Five quality gates — the main difference from “an AI script”

Most people sell you a script that generates content. I sell a production line that refuses to skip quality checks. The scripts below all exist and you can run them.

GateWhat it blocksActual result this run
Length gateUnder/over the word-count band✓ pass 0 out of band
AI-flavor gateClichés, fake contrast, dash overuse✗ 3 over limit blocked, pending fix
Wording gateStock-tips, earnings promises● high-risk 0
Per-card gateMissing fields, length out of bounds✓ pass
Plain-language gateRun-on sentences, unanchored time words✗ 4 errors 13 warnings
The failures are on this page on purpose. Two gates caught real defects in this run, and I did not paint them green.
If a vendor hands you “all five gates green,” that usually means the gates are not wired into the step that actually matters — they run alongside for show. A gate that can report red is a gate.

What that looked like in the terminal

The five lines below are the same five results as the table above, printed by the run itself. Nothing is animated and nothing is re-enacted: this is a frozen frame.

$ python tools/daily_run.py --date 2026-10-03
[1/5] length gate        pass   0 out of band
[2/5] ai-flavor gate     FAIL   3 over limit → blocked
[3/5] wording gate       pass   high-risk 0
[4/5] per-card gate      pass
[5/5] plain-language     FAIL   4 errors, 13 warnings
exit 1 · nothing was published

That last line is the whole point. A run that fails two gates produces no publishing file — it does not produce a file with a warning attached.

The folder that run left behind

2026-10-03/
brief.md18,181 B
发布卡.txt15,123 B
covers/3 images, hashes in section 2
raw/4 files, unprocessed

The two byte counts are the same numbers as section 5, measured with wc -c on the files themselves. The cover folder holds the three images printed above, so you can hash those files and compare.

4 · What it does and does not do

Detail
DoesResearch and draft, rewrite per platform, render covers, run the quality gates, produce the plain-text delivery file, pull data on a schedule
Does notClick “publish” for you — platforms do not persist publishing settings with drafts, and some prohibit scripted publishing. I will not pretend otherwise.

5 · Verifiable deliverables (ask for whichever you want)

ArtifactSize
Full news brief (Markdown)18,181 bytes
Publishing cards, plain text15,123 bytes
Raw scraped data, unprocessed4 files
Quality-gate scripts (runnable)5

⭐ The gate scripts themselves are yours to read and run. If you do not trust the claims, feed them a deliberately broken draft and watch whether they fire.

6 · Pricing

Self-serve product · priced in USD · the only item in our catalog

ProductPriceWhat you get
Content Automation Pipeline (self-hosted, 5 pre-publish gates)$199 USDFull source code, gate specifications, deployment runbook, and a 20-minute handover call

Pay with any international card (Visa / Mastercard) in USD. Sales tax is calculated and remitted by the platform based on the buyer's location — you do not need a Chinese bank account or Chinese currency.

Two things stated up front, not buried: (1) zero testimonials, zero case studies (2) if the delivered software deviates from its written specification, that is my responsibility and I fix it at no cost.

Buy now — $199 USD

Checkout and payment are handled by Creem, our merchant of record. Digital delivery is instant; the source package and handover call are arranged by email.

7 · In one line

It is not “an AI that writes for you” — it is a production line with quality gates: it runs by itself, it reports its own errors, and you can verify the errors on the spot.
Every number on this page traces back to a file; two gate failures are deliberately left in.

8 · Who is behind this

I am a solo developer in Shanghai, China, operating as Dekita Automation. I sell as an individual — there is no team and no registered company entity.

My public work is verifiable rather than claimed: technical articles on dev.to (four pieces) and on Zenn (three Japanese articles). The strongest verifiable evidence is on this page itself, and it is scoped precisely: the pipeline metrics — word counts, cover-image SHA256 hashes, and the pass/fail result of each gate — are all measured by scripts and can be recomputed from the files. That is what "verifiable" means here. The fictional sample is a format demonstration and is not part of that verifiable set.

Produced 2026-10-04 | No generative images used | No beautified data | No hidden failures | This site is self-hosted on my own infrastructure — no third-party platform registration involved

Privacy Policy  ·  Terms of Service  ·  Acceptable Use  ·  Chinese version  ·  Contact: support@dekita-ai.com