ISHIKAWA.CO
Return to the Living Atlas

PUBLIC PRODUCT / PRIVATE BETA

FREE CORE AVAILABLE / PRIVATE BETA LIVE

Source-to-Publish Evidence Gate

Close the publication decision with an evidence-backed PASS or HOLD.

More source URLs do not finish a publication decision. This workflow records claims, evidence, mutable facts, and required follow-up, then regenerates the same decision from the same input.

Built from Hidekazu Ishikawa’s daily work publishing AI news, technical articles, and business recommendations. It does not add more search or writing. It closes the decision that comes next: can this be published?

[ THE DECISION ]

Not a tool for collecting more sources. A product for finishing the publication decision.

It does not perform live search, fetch sources, extract claims automatically, or publish to a CMS. It turns the public-safe manifest you supply into an explainable claim ledger and publication gate.

[ 45 SECOND DEMO ]

From HOLD to PASS in 45 seconds.

A public-safe manifest enters the real engine and returns HOLD for an unverified mutable fact. Add verified_at, rerun, and keep the resulting PASS in matching Markdown and JSON reports.

Free core 0.2.0 / real engine output / 44.98 sec / silent
  1. public-safe manifest
  2. HOLD / exit 3
  3. add verified_at
  4. PASS / exit 0
  5. Markdown + JSON

[ VERIFIED PRODUCT LOOP ]

Trace the same manifest from HOLD to PASS.

This is output from the real engine using the sample fixture in the public RC.4 archive. The HOLD case removes only verified_at; the source, claim, and mode stay unchanged.

RC.4 fixture / real engine / reproducible
Real Source-to-Publish Evidence Gate RC.4 engine output showing the manifest, HOLD with exit 3, the verified_at correction, and PASS with exit 0.

Reproduce the result with the free core, or choose the private beta when you also need complete practical workflows.

FREE CORE / MIT

Download the free core v0.2.0-rc.4 / MIT / tar.gz

SHA-256 3a4509e84a419d199a96dbf3cd7e4290ec2f58e6fad6315634941bc5aab60020

PAID / $29 ONE-TIME

Purchase the private beta up to 5 buyers / no calls / all 0.x updates

[ REPORT SAMPLE ]

Two clear states, not an invented confidence score.

The verified RC.4 run above is also readable as text. The Markdown and versioned JSON reports always carry the same gate result.

PASS

PASS / ready with limits

mode: technical-article
claim: Example 2.0 is available.
mutable fact: version / verified
publication_gate: PASS
exit: 0 / reasons: []
HOLD

HOLD / correct and rerun

mode: technical-article
claim: Example 2.0 is available.
mutable fact: version / verify-first
publication_gate: HOLD
exit: 3 / add verified_at, then rerun

[ THREE WORKFLOWS ]

Shift the review emphasis to fit the publication.

The paid workflow pack applies the same engine through three practical modes.

01 / AI NEWS

Separate the announcement from the current state.

Start with official announcements, documentation, and release notes. Recheck model names, availability, dates, regions, and pricing immediately before publication.

02 / TECHNICAL ARTICLE

Tie reproducibility back to evidence.

Connect official documentation, source code, release notes, and reproduction tests. Review versions, compatibility, security behavior, and API contracts.

03 / BUSINESS RECOMMENDATION

Keep recommendations separate from facts.

Separate vendor terms from independent primary evidence, and make changing prices, plans, regions, support, and terms explicit HOLD conditions.

[ LOCAL BOUNDARY ]

Keep the draft off the network.

The core engine runs offline, does not fetch URLs, and needs neither an account login nor a runtime license check.

OFFLINE

The engine makes no network connection and processes only the manifest you provide.

NO TELEMETRY

Usage data, drafts, source text, and report text are not sent to the seller automatically.

PUBLIC-SAFE ONLY

Credential-bearing, private, login, and signed URLs are rejected during preflight.

RAW DRAFT STAYS LOCAL

Support does not request raw drafts. Reproduction uses redacted errors and environment metadata only.

Independently QA-tested only on macOS 26 / Apple silicon / Python 3.10.8. Linux, Windows, Intel Mac, and other Python versions are outside private-beta support.

[ PACKAGE MANAGER ]

Run the version-pinned wheel temporarily with uvx.

Fetch the free-core 0.2.0 wheel directly from ishikawa.co and run its console script without installing it permanently. Point input.json to a public-safe manifest.

PINNED FIRST RUN / FREE CORE 0.2.0

uvx --from https://ishikawa.co/downloads/source-to-publish-evidence-gate/source_to_publish_evidence_gate-0.2.0-py3-none-any.whl evidence-gate input.json --json-output report.json --markdown-output report.md

The sample fixture returns PASS with exit 0; HOLD intentionally returns exit 3. After uvx retrieves the wheel, the engine runs offline with no telemetry and does not upload the manifest or report.

Download the 0.2.0 wheel MIT / PY3 / NONE / ANY

SHA-256 d9e603df19df5cb01741fdede43559d9b845eb0fb4f4b7f06725b6835521428b

[ BUYER JOURNEY ]

Install it, correct the HOLD, and run it again.

The product experience ends when you close one of your own decisions, not when you merely view a sample.

  1. 01 / DOWNLOAD

    Save the free core

    Download the public RC.4 tar.gz from the Free core CTA on this page.

    source-to-publish-evidence-gate-free-0.2.0-rc.4.tar.gz

  2. 02 / VERIFY SHA

    Match the SHA-256

    Confirm that the digest matches the value shown on this page before extraction.

    shasum -a 256 source-to-publish-evidence-gate-free-0.2.0-rc.4.tar.gz

  3. 03 / EXTRACT

    Extract and run buyer verification

    Enter the extracted directory and verify the archive inventory and product boundary locally.

    tar -xzf source-to-publish-evidence-gate-free-0.2.0-rc.4.tar.gz && cd source-to-publish-evidence-gate-free-0.2.0-rc.4 && python3 verify.py

  4. 04 / INSTALL

    Add the free plugin from the local marketplace

    Register the public archive as a marketplace and add the free core plugin.

    codex plugin marketplace add . && codex plugin add source-to-publish-evidence-gate@personal

  5. 05 / SAMPLE RUN

    Run the included fixture with the real engine

    Enter the plugin directory and generate JSON and Markdown reports with the same decision.

    cd plugins/source-to-publish-evidence-gate && scripts/evidence-gate skills/source-to-publish-evidence-gate/references/input-example.json --json-output /tmp/evidence-report.json --markdown-output /tmp/evidence-report.md

  6. 06 / EXPECTED RESULT

    Confirm PASS, then correct and rerun your own HOLD

    The included fixture returns exit 0 / PASS. If your manifest returns exit 3 / HOLD, correct the reason, keep before and after, and rerun.

    publication_gate: PASS / exit: 0

[ FREE / PAID ]

The decision engine is free. The workflow for finishing real work is paid.

The two archives remain separate, with value and license boundaries defined file by file.

FREE CORE / MIT

Test the decision quality first.

A usable deterministic core, not an advertising stub.

For people who want to verify the engine’s decision quality, privacy boundary, and installation path themselves.

  • deterministic engine
  • versioned JSON Schema / Markdown report
  • public URL preflight
  • one reproducible sample
  • buyer verification
  • install / uninstall / safety docs

If you are unsure, reproduce one decision with the free core. Stay free when the engine is enough; choose paid when you need the three practical workflows and completed samples.

The paid pack is delivered as a separate archive after purchase. Private files are not mixed into the free archive, and the Polar destination is never embedded in public HTML.

If you are unsure, reproduce one decision with the free core. Stay free when the engine is enough; choose paid when you need the three practical workflows and completed samples.

FREE CORE / MIT

Download the free core v0.2.0-rc.4 / MIT / tar.gz

SHA-256 3a4509e84a419d199a96dbf3cd7e4290ec2f58e6fad6315634941bc5aab60020

PAID / $29 ONE-TIME

Purchase the private beta up to 5 buyers / no calls / all 0.x updates

[ PRIVATE BETA PRICE ]

$29 one-time for the private beta.

This is not a subscription. The cohort is limited to five buyers and includes all 0.x updates. $39 is only a standard-price candidate to reconsider after beta evaluation; it is not the current sale price.

PRIVATE BETA

$29 one-time up to 5 buyers / no calls / all 0.x updates

STANDARD CANDIDATE

$39 one-time not for sale / reconsidered after beta evaluation

[ TERMS ]

Agree to the use boundary before purchase.

Checkout requires confirmation of the supported environment, support boundary, refund policy, and commercial license. Merchant-of-record terms and mandatory consumer protections take precedence where applicable.

[ SUPPORT ]

No calls and no consulting.

Asynchronous support covers blocking installation defects, missing archive content, version mismatches, failure to reproduce a documented workflow, and security or privacy bugs. The private beta includes one asynchronous question. A first response is targeted within three business days, but it is not an SLA.

Source collection, research, writing, judging individual claims, setup calls, screen sharing, meetings, advisory work, and custom integration are excluded.

[ REFUND ]

Full refund in principle within 14 days of purchase.

A reason is not required. A refund ends the download benefit and paid license, and the buyer is asked to delete downloaded paid files and private backups. Reports legitimately generated before the refund may continue to be used.

[ LICENSE ]

MIT for free. Single-user for paid.

The buyer may use the paid pack on up to three devices they own or manage. Generated reports may be used and shared commercially, including for client work. The paid archive, templates, guides, decision tables, workflows, and samples may not be redistributed or resold.

[ UPDATES ]

The private beta includes every 0.x update.

Fixes and improvements within 0.x are included without another fee. Rights to 1.0 or later, a future standard price, a subscription, or future versions are not included. Check the version and SHA-256 before updating, and retain existing reports.

[ FAQ ]

Make the boundary clear before purchase.

Does it fact-check my article automatically?

No. It checks whether the claims and evidence you supplied satisfy the declared publication policy and returns an auditable PASS or HOLD. It does not guarantee facts absolutely.

Does it upload my draft or source text?

No. The core engine runs locally, performs no fetch, and sends no telemetry. Optional external tools have separate data boundaries for you to evaluate.

Does PASS prove every claim is true?

No. PASS means the supplied manifest satisfies the gate. It is not legal, medical, plagiarism, authorship, or absolute factual certification.

Why pay if the engine is free?

The paid pack provides three complete workflows, decision tables, bilingual guidance, handoff templates, a HOLD-to-PASS walkthrough, and verified samples for consistent publishing work.

Can generated reports be used for client work?

Yes. A single-user buyer may use and share generated reports commercially. The paid pack itself may not be redistributed or resold.

Is a subscription or setup support required?

No subscription is required. Calls and consulting are not included. The private beta includes one asynchronous question within the support boundary.

[ AVAILABILITY ]

Keep the two entry points separate.

The free core is available as a version-pinned wheel or direct public archive. The private beta is available to up to five buyers. Review the price and terms in checkout before purchasing.

FREE CORE / MIT

Download the free core v0.2.0-rc.4 / MIT / tar.gz

SHA-256 3a4509e84a419d199a96dbf3cd7e4290ec2f58e6fad6315634941bc5aab60020

PAID / $29 ONE-TIME

Purchase the private beta up to 5 buyers / no calls / all 0.x updates