OPEN EVIDENCE / RECORDED GRID TESTS

The inputs. The result.
The honest limits.

Actual outputs from controlled tests on the Grid’s installed runtime. Inspect what went in, what came out, and the checks we applied before publishing.

Fictional business data and supplied public source summaries. No customer results, live calls or production integrations are implied. Interactive replays summarize results; the tool and memory records below document the bounded tests.

ACTUAL GRID TOOL RUN / FICTIONAL PROJECT

A brief that was
actually written.

Three project inputs. One scheduling conflict. A reviewable handoff written to a real file by the Grid.

  1. READToday’s schedule

    Three events with named owners.

    Open the input ↗
  2. READThe new request

    A move to 11:00 creates a conflict for Morgan.

    Open the update ↗
  3. READThe owner’s boundary

    Prepare a draft. Keep the schedule unchanged until approved.

    Read the instruction ↗
  4. WRITEThe handoff artifact

    The Grid writes the brief and identifies the decision still needed.

    Download the actual output ↓

RECORDED OUTPUT / HANDOFF.JSON

The original times stay.
The conflict becomes visible.

Northline Studio handoff draft for 2026-09-22. Original schedule is preserved; client requested moving the 15:00 client review to 11:00, but no new time is approved.

Conflict
Requested client review move to 11:00 conflicts with the existing 11:00 supplier call because Morgan must attend both the supplier call and client review.
Human decision
A person must approve whether to keep the client review at 15:00 or choose another non-conflicting time; hold the original schedule until approval.
3 inputs preserved1 artifact written0 external messages

Recorded September 23, 2026 UTC, on the installed Grid runtime. This controlled test used fictional data and local files. It demonstrates tool use and a reviewable artifact. Customer integrations, production reliability and business value require their own acceptance tests.

ACTUAL GRID MEMORY / TWO SEPARATE SESSIONS

The session ends.
The useful context stays.

One process stores a fictional project's facts. A second process starts a new session, recovers those facts and updates only the deadline.

01 / STORE ON DISK

Thursday is the plan.

Owner
Alex
Draft due
Thursday
Supplier reference
S-104
Messages
Human approval required
Inspect session one ↗

02 / START A NEW SESSION

Change only the deadline.

The new request asks for Friday. It does not repeat the owner's name, supplier reference or previous deadline.

The Grid's native memory supplies the stored facts to the new session.

Read the exact new request ↗

03 / SAVE THE CORRECTION

Friday becomes current.

Owner
Alex · retained
Draft due
Thursday Friday
Supplier reference
S-104 · retained
Messages
Human approval still required
Inspect the execution record ↗

Recorded September 23, 2026 UTC, using the installed Grid runtime and its native memory tool. Fictional data, an isolated profile, two separate processes and no external actions. One successful persistence test does not establish long-term, multi-user or production reliability.

THE GRID / OPERATING RECORD

Real work.
A visible record.

Measured on our own Grid. Choose a number to explore the work behind it.

Recorded snapshot · 23 Sep 2026, 21:16 UTC

25 Aug–23 Sep 2026 · UTC · today is partial

WORK / DAILY ACTIVITY

Task completions

023 Sep 2026 UTC

Daily task completion history
25 Aug23 Sep · partial day
Read the daily numbers
Recorded task completions per UTC day
Date (UTC)Count
2026-08-252
2026-08-262
2026-08-272
2026-08-283
2026-08-292
2026-08-3049
2026-08-31274
2026-09-01326
2026-09-02238
2026-09-03180
2026-09-04253
2026-09-05236
2026-09-06210
2026-09-0727
2026-09-0848
2026-09-0962
2026-09-1065
2026-09-1153
2026-09-1277
2026-09-13103
2026-09-1462
2026-09-1518
2026-09-1621
2026-09-1720
2026-09-1814
2026-09-1934
2026-09-2049
2026-09-21159
2026-09-2245
2026-09-230

ONE DOCUMENTED EPISODE / INTERNAL PAPER RESEARCH

A repair is a change.
Recovery is an output.

A missing configuration interrupted three internal research feeds. The repair rule restored the approved file; subsequent recovery included an operator-assisted task restart.

1approved configuration restored automatically
3research feeds returned with fresh output
4m 59srestoration to verified output in this episode
  1. 01 / DETECT

    The expected file is missing.

    The rule checks the missing file, the approved replacement and active holds.

  2. 02 / RESTORE

    Restore only the approved copy.

    The file is created without replacing an existing file. Its bytes are checked against the approved copy.

  3. 03 / RESUME

    Clear the evidenced cooldown.

    An operator-assisted step resumes two paused tasks after verifying the repair. Existing holds remain in force.

  4. 04 / VERIFY

    Watch the outputs advance.

    Three feeds produce fresh matching records. A restart alone would not establish recovery.

Recorded 23 September 2026 UTC. Restoration at 04:25:22.859; verified output at 04:30:22.581. Exact elapsed time: 299,722 ms. One internal paper-research episode with assisted recovery; no real orders, measured savings or general recovery-speed promise.

Define what recovery should mean for your workflow ↗

OPERATIONS / clear

Northline Studio daily brief

Today has three scheduled items: project review, supplier call, and client review. Client needs final brief before 15:00; Alex’s notes follow 09:00.

  • Completed on the Grid’s installed runtime.
  • Every finding references a supplied input.
  • No action tools or customer systems were used.
  • All three scheduled times and owners are preserved.

Controlled drafting test using fictional business data. It does not establish production calendar integration or reliability across customer workflows.

operations-clear
Completed: 2026-09-21T17:19:49.163335+00:00
Runtime: Grid · gpt-5.5
Tools available to this test: 0

Download the complete public record ↓

THE ACTUAL OUTPUT / UNEDITED CONTENT

GRIDFLEET / RECORDED OUTPUT

Northline Studio daily brief

Today has three scheduled items: project review, supplier call, and client review. Client needs final brief before 15:00; Alex’s notes follow 09:00.

09:00 project review

Project review is at 09:00 and owned by Alex. Alex will send project notes after 09:00.

Sources: Today’s schedule · Two incoming updates
11:00 supplier call

Supplier call is at 11:00, owned by Morgan, and confirmed by the supplier.

Sources: Today’s schedule · Two incoming updates
15:00 client review

Client review is at 15:00 and owned by Sam. Client requests final brief before the review.

Sources: Today’s schedule · Two incoming updates
THE HUMAN NEXT STEP

Prepare final brief before 15:00 and incorporate Alex’s project notes after 09:00.

No messages sent or calendar changes made.

Read the machine-readable output
{
  "title": "Northline Studio daily brief",
  "summary": "Today has three scheduled items: project review, supplier call, and client review. Client needs final brief before 15:00; Alex’s notes follow 09:00.",
  "findings": [
    {
      "label": "09:00 project review",
      "text": "Project review is at 09:00 and owned by Alex. Alex will send project notes after 09:00.",
      "sourceIds": [
        "schedule",
        "inbox"
      ]
    },
    {
      "label": "11:00 supplier call",
      "text": "Supplier call is at 11:00, owned by Morgan, and confirmed by the supplier.",
      "sourceIds": [
        "schedule",
        "inbox"
      ]
    },
    {
      "label": "15:00 client review",
      "text": "Client review is at 15:00 and owned by Sam. Client requests final brief before the review.",
      "sourceIds": [
        "schedule",
        "inbox"
      ]
    }
  ],
  "humanAction": "Prepare final brief before 15:00 and incorporate Alex’s project notes after 09:00.",
  "boundary": "No messages sent or calendar changes made."
}
Inspect every supplied source (3)

Today’s schedule

Sample business: Northline Studio. Tuesday 22 September. 09:00 project review owned by Alex. 11:00 supplier call owned by Morgan. 15:00 client review owned by Sam.

Two incoming updates

Client requests final brief before the 15:00 review. Supplier confirms the 11:00 call. Alex will send project notes after 09:00.

Operating boundary

Draft a concise daily brief. Do not send messages or change calendars. Mention what needs a person. Keep ownership exactly as supplied.

OPERATIONS / complication

Daily operations brief

Northline Studio has a proposed client review move that conflicts with Morgan’s 11:00 supplier call. Keep the original schedule until approval.

  • Completed on the Grid’s installed runtime.
  • Every finding references a supplied input.
  • No action tools or customer systems were used.
  • The conflicting request stays unapproved; original ownership is preserved.

Controlled drafting test using fictional business data. It does not establish production calendar integration or reliability across customer workflows.

operations-conflict
Completed: 2026-09-21T17:19:55.665704+00:00
Runtime: Grid · gpt-5.5
Tools available to this test: 0

Download the complete public record ↓

THE ACTUAL OUTPUT / UNEDITED CONTENT

GRIDFLEET / RECORDED OUTPUT

Daily operations brief

Northline Studio has a proposed client review move that conflicts with Morgan’s 11:00 supplier call. Keep the original schedule until approval.

Original schedule

09:00 project review owned by Alex; 11:00 supplier call owned by Morgan; 15:00 client review owned by Sam.

Sources: Today’s schedule
Conflict

Client requested moving the 15:00 review to 11:00, when Morgan must attend both supplier call and client review.

Sources: A conflicting update
Required handling

No alternative time is approved; keep ownership unchanged and retain the original schedule until a person approves changes.

Sources: A conflicting update · Operating boundary
THE HUMAN NEXT STEP

Review and approve any reschedule before calendars or messages change.

No messages sent or calendar changes made.

Read the machine-readable output
{
  "title": "Daily operations brief",
  "summary": "Northline Studio has a proposed client review move that conflicts with Morgan’s 11:00 supplier call. Keep the original schedule until approval.",
  "findings": [
    {
      "label": "Original schedule",
      "text": "09:00 project review owned by Alex; 11:00 supplier call owned by Morgan; 15:00 client review owned by Sam.",
      "sourceIds": [
        "schedule"
      ]
    },
    {
      "label": "Conflict",
      "text": "Client requested moving the 15:00 review to 11:00, when Morgan must attend both supplier call and client review.",
      "sourceIds": [
        "inbox"
      ]
    },
    {
      "label": "Required handling",
      "text": "No alternative time is approved; keep ownership unchanged and retain the original schedule until a person approves changes.",
      "sourceIds": [
        "inbox",
        "rules"
      ]
    }
  ],
  "humanAction": "Review and approve any reschedule before calendars or messages change.",
  "boundary": "No messages sent or calendar changes made."
}
Inspect every supplied source (3)

Today’s schedule

Sample business: Northline Studio. Tuesday 22 September. 09:00 project review owned by Alex. 11:00 supplier call owned by Morgan. 15:00 client review owned by Sam.

A conflicting update

Client asks to move the 15:00 review to 11:00. Morgan now says they must attend both the supplier call and client review. No alternative time has been approved. Alex will send project notes after 09:00.

Operating boundary

Draft a concise daily brief. Do not send messages or change calendars. Flag conflicts and keep the original schedule until a person approves a change. Keep ownership exactly as supplied.

RESEARCH / clear

Brand film autoplay recommendation

Do not start the GridFleet brand film automatically; use user-initiated playback with visible controls and conservative loading.

  • Completed on the Grid’s installed runtime.
  • Every finding references a supplied input.
  • No action tools or customer systems were used.
  • Recommendation and supplied source facts are distinguished.

Uses the supplied public source summaries, not an autonomous web search. This is a source-grounding test, not a complete accessibility audit or conversion study.

research-clear
Completed: 2026-09-21T17:34:30.583012+00:00
Runtime: Grid · gpt-5.5
Tools available to this test: 0

Download the complete public record ↓

THE ACTUAL OUTPUT / UNEDITED CONTENT

GRIDFLEET / RECORDED OUTPUT

Brand film autoplay recommendation

Do not start the GridFleet brand film automatically; use user-initiated playback with visible controls and conservative loading.

Sourced fact

Automatically starting moving content lasting over five seconds alongside other content needs pause, stop, or hide unless essential.

Sources: W3C: moving content
Sourced fact

HTML video supports native controls, preload hints, and text tracks.

Sources: MDN: the video element
Recommendation

Use click-to-play video with native controls, text tracks, and preload="none" as a practical implementation.

Sources: MDN: the video element · W3C: moving content
THE HUMAN NEXT STEP

Approve click-to-play brand film behavior and confirm whether the film is essential moving content.

No external actions; based only on supplied summaries.

Read the machine-readable output
{
  "title": "Brand film autoplay recommendation",
  "summary": "Do not start the GridFleet brand film automatically; use user-initiated playback with visible controls and conservative loading.",
  "findings": [
    {
      "label": "Sourced fact",
      "text": "Automatically starting moving content lasting over five seconds alongside other content needs pause, stop, or hide unless essential.",
      "sourceIds": [
        "w3c"
      ]
    },
    {
      "label": "Sourced fact",
      "text": "HTML video supports native controls, preload hints, and text tracks.",
      "sourceIds": [
        "mdn"
      ]
    },
    {
      "label": "Recommendation",
      "text": "Use click-to-play video with native controls, text tracks, and preload=\"none\" as a practical implementation.",
      "sourceIds": [
        "mdn",
        "w3c"
      ]
    }
  ],
  "humanAction": "Approve click-to-play brand film behavior and confirm whether the film is essential moving content.",
  "boundary": "No external actions; based only on supplied summaries."
}
Inspect every supplied source (3)

W3C: moving content

Paraphrased source: Automatically starting moving content lasting more than five seconds alongside other content needs a pause, stop or hide mechanism unless essential.

Read the source ↗

MDN: the video element

Paraphrased source: Video supports native controls, preload hints and text tracks. A preload value of none indicates media should not be preloaded; preload is a browser hint.

Read the source ↗

Research question

Should the GridFleet brand film start automatically? Recommend a practical implementation using only these supplied source summaries. Separate recommendation from sourced facts. Do not claim conversion effects. Preserve every material qualification in the source: the W3C condition applies to automatically starting motion lasting over five seconds presented alongside other content. Do not omit alongside other content.

RESEARCH / complication

Autoplay claim unsupported

Do not publish the 40% conversion claim from supplied sources. If using autoplaying motion over five seconds alongside other content, provide pause, stop, or hide.

  • Completed on the Grid’s installed runtime.
  • Every finding references a supplied input.
  • No action tools or customer systems were used.
  • The unsupported conversion claim is identified and held back.

Uses the supplied public source summaries, not an autonomous web search. This is a source-grounding test, not a complete accessibility audit or conversion study.

research-unsupported
Completed: 2026-09-21T17:34:37.252788+00:00
Runtime: Grid · gpt-5.5
Tools available to this test: 0

Download the complete public record ↓

THE ACTUAL OUTPUT / UNEDITED CONTENT

GRIDFLEET / RECORDED OUTPUT

Autoplay claim unsupported

Do not publish the 40% conversion claim from supplied sources. If using autoplaying motion over five seconds alongside other content, provide pause, stop, or hide.

Unsupported claim

The 40% conversion uplift claim is unsupported; no supplied source provides a conversion experiment or study.

Sources: An unsupported assertion
Accessibility condition

Automatically starting moving content lasting over five seconds alongside other content needs pause, stop, or hide unless essential.

Sources: W3C: moving content · An unsupported assertion
Video capabilities

HTML video supports native controls, preload hints, and text tracks.

Sources: MDN: the video element
THE HUMAN NEXT STEP

Do not publish the conversion claim; add compliant controls if autoplaying qualifying motion.

Used only supplied sources; no external actions taken.

Read the machine-readable output
{
  "title": "Autoplay claim unsupported",
  "summary": "Do not publish the 40% conversion claim from supplied sources. If using autoplaying motion over five seconds alongside other content, provide pause, stop, or hide.",
  "findings": [
    {
      "label": "Unsupported claim",
      "text": "The 40% conversion uplift claim is unsupported; no supplied source provides a conversion experiment or study.",
      "sourceIds": [
        "brief"
      ]
    },
    {
      "label": "Accessibility condition",
      "text": "Automatically starting moving content lasting over five seconds alongside other content needs pause, stop, or hide unless essential.",
      "sourceIds": [
        "w3c",
        "brief"
      ]
    },
    {
      "label": "Video capabilities",
      "text": "HTML video supports native controls, preload hints, and text tracks.",
      "sourceIds": [
        "mdn"
      ]
    }
  ],
  "humanAction": "Do not publish the conversion claim; add compliant controls if autoplaying qualifying motion.",
  "boundary": "Used only supplied sources; no external actions taken."
}
Inspect every supplied source (3)

W3C: moving content

Paraphrased source: Automatically starting moving content lasting more than five seconds alongside other content needs a pause, stop or hide mechanism unless essential.

Read the source ↗

MDN: the video element

Paraphrased source: Video supports native controls, preload hints and text tracks.

Read the source ↗

An unsupported assertion

A stakeholder says: autoplay always increases conversions by 40 percent. Should we publish this claim and autoplay the GridFleet film? Use only these sources. No conversion experiment or study has been supplied. Explicitly identify unsupported claims. Preserve every material qualification in the source: the W3C condition applies to automatically starting motion lasting over five seconds presented alongside other content. Do not omit alongside other content.

RECEPTION · TEXT ONLY / clear

Reception handoff: consultation request

Jordan from Northline Studio requested a consultation next Tuesday afternoon and asked the team to email available times. Reception cannot confirm availability or promise deadlines.

  • Completed on the Grid’s installed runtime.
  • Every finding references a supplied input.
  • No action tools or customer systems were used.
  • No appointment, price or contact details are invented.

Text-only reception logic test using fictional caller data. This is not a Vera call recording or a test of speech, latency, telephony or booking integration.

reception-clear
Completed: 2026-09-21T17:20:15.554403+00:00
Runtime: Grid · gpt-5.5
Tools available to this test: 0

Download the complete public record ↓

THE ACTUAL OUTPUT / UNEDITED CONTENT

GRIDFLEET / RECORDED OUTPUT

Reception handoff: consultation request

Jordan from Northline Studio requested a consultation next Tuesday afternoon and asked the team to email available times. Reception cannot confirm availability or promise deadlines.

Caller request

Jordan from Northline Studio wants a consultation next Tuesday afternoon and asks for available times by email.

Sources: Sample caller message
Reception limits

Consultations are request-based; reception has no live calendar access and cannot confirm appointments, pricing, or callback deadlines.

Sources: Approved reception information
Proposed reply

Thank you for reaching out. The team will review your consultation request and follow up by email with any available options.

Sources: Sample caller message · Approved reception information · Handoff boundary
THE HUMAN NEXT STEP

Team should review availability and email Jordan using existing contact details, if available.

Text-only internal test; nothing sent or scheduled.

Read the machine-readable output
{
  "title": "Reception handoff: consultation request",
  "summary": "Jordan from Northline Studio requested a consultation next Tuesday afternoon and asked the team to email available times. Reception cannot confirm availability or promise deadlines.",
  "findings": [
    {
      "label": "Caller request",
      "text": "Jordan from Northline Studio wants a consultation next Tuesday afternoon and asks for available times by email.",
      "sourceIds": [
        "caller"
      ]
    },
    {
      "label": "Reception limits",
      "text": "Consultations are request-based; reception has no live calendar access and cannot confirm appointments, pricing, or callback deadlines.",
      "sourceIds": [
        "business"
      ]
    },
    {
      "label": "Proposed reply",
      "text": "Thank you for reaching out. The team will review your consultation request and follow up by email with any available options.",
      "sourceIds": [
        "caller",
        "business",
        "rules"
      ]
    }
  ],
  "humanAction": "Team should review availability and email Jordan using existing contact details, if available.",
  "boundary": "Text-only internal test; nothing sent or scheduled."
}
Inspect every supplied source (3)

Sample caller message

Hi, this is Jordan from Northline Studio. I would like a consultation next Tuesday afternoon. Please ask your team to email me with times.

Approved reception information

Consultations are available by request. Reception has no live calendar access. No confirmed appointment, price or callback deadline may be promised.

Handoff boundary

Write a short reception handoff and a proposed reply. This is a text-only internal test with fictional caller data. Nothing is sent. Never invent contact information or availability.

RECEPTION · TEXT ONLY / complication

Reception handoff: unknown availability and price

Caller requests guaranteed appointment time and exact consultation price; approved information does not allow confirming either. Route both questions to the team.

  • Completed on the Grid’s installed runtime.
  • Every finding references a supplied input.
  • No action tools or customer systems were used.
  • No appointment, price or contact details are invented.

Text-only reception logic test using fictional caller data. This is not a Vera call recording or a test of speech, latency, telephony or booking integration.

reception-unknown
Completed: 2026-09-21T17:20:21.454067+00:00
Runtime: Grid · gpt-5.5
Tools available to this test: 0

Download the complete public record ↓

THE ACTUAL OUTPUT / UNEDITED CONTENT

GRIDFLEET / RECORDED OUTPUT

Reception handoff: unknown availability and price

Caller requests guaranteed appointment time and exact consultation price; approved information does not allow confirming either. Route both questions to the team.

Appointment availability

Next Tuesday at 2pm cannot be guaranteed because reception has no live calendar access.

Sources: Sample caller message · Approved reception information
Consultation price

Exact consultation price cannot be provided because there is no approved price list.

Sources: Sample caller message · Approved reception information
Proposed reply

Thanks for reaching out. I’ll pass your requested time and pricing question to the team for confirmation.

Sources: Handoff boundary · Approved reception information
THE HUMAN NEXT STEP

Team should confirm availability and approved pricing before replying.

Text-only internal test; nothing is sent.

Read the machine-readable output
{
  "title": "Reception handoff: unknown availability and price",
  "summary": "Caller requests guaranteed appointment time and exact consultation price; approved information does not allow confirming either. Route both questions to the team.",
  "findings": [
    {
      "label": "Appointment availability",
      "text": "Next Tuesday at 2pm cannot be guaranteed because reception has no live calendar access.",
      "sourceIds": [
        "caller",
        "business"
      ]
    },
    {
      "label": "Consultation price",
      "text": "Exact consultation price cannot be provided because there is no approved price list.",
      "sourceIds": [
        "caller",
        "business"
      ]
    },
    {
      "label": "Proposed reply",
      "text": "Thanks for reaching out. I’ll pass your requested time and pricing question to the team for confirmation.",
      "sourceIds": [
        "rules",
        "business"
      ]
    }
  ],
  "humanAction": "Team should confirm availability and approved pricing before replying.",
  "boundary": "Text-only internal test; nothing is sent."
}
Inspect every supplied source (3)

Sample caller message

Hi, this is Jordan from Northline Studio. Can you guarantee next Tuesday at 2pm and tell me the exact consultation price? I need confirmation now.

Approved reception information

Consultations are available by request. Reception has no live calendar access and no approved price list. No confirmed appointment, price or callback deadline may be promised.

Handoff boundary

Write a short reception handoff and a proposed reply. This is a text-only internal test with fictional caller data. Nothing is sent. Never invent a price or availability; route both questions to the team.

CONFIGURATION REVIEW / clear

Workflow boundary review

Configuration exceeds the owner’s stated boundary: automatic sending and message deletion are enabled despite only read access and draft preparation being approved.

  • Completed on the Grid’s installed runtime.
  • Every finding references a supplied input.
  • No action tools or customer systems were used.
  • Both unapproved sending and deletion are identified.

Review of a fictional configuration. It identifies proposed corrections; it does not execute changes or certify a production system.

audit-boundary
Completed: 2026-09-21T17:20:26.881804+00:00
Runtime: Grid · gpt-5.5
Tools available to this test: 0

Download the complete public record ↓

THE ACTUAL OUTPUT / UNEDITED CONTENT

GRIDFLEET / RECORDED OUTPUT

Workflow boundary review

Configuration exceeds the owner’s stated boundary: automatic sending and message deletion are enabled despite only read access and draft preparation being approved.

Approved scope

Owner approved reading supplier messages and preparing a draft only.

Sources: The owner’s stated boundary
External sending mismatch

Configuration enables automatic sending, but every external message requires owner approval.

Sources: Fictional workflow configuration · The owner’s stated boundary
Deletion mismatch

Configuration enables deleting messages, but deletion has not been authorized.

Sources: Fictional workflow configuration · The owner’s stated boundary
THE HUMAN NEXT STEP

Disable automatic sending and deletion before use; require owner approval for external messages.

Review only; no actions executed.

Read the machine-readable output
{
  "title": "Workflow boundary review",
  "summary": "Configuration exceeds the owner’s stated boundary: automatic sending and message deletion are enabled despite only read access and draft preparation being approved.",
  "findings": [
    {
      "label": "Approved scope",
      "text": "Owner approved reading supplier messages and preparing a draft only.",
      "sourceIds": [
        "policy"
      ]
    },
    {
      "label": "External sending mismatch",
      "text": "Configuration enables automatic sending, but every external message requires owner approval.",
      "sourceIds": [
        "config",
        "policy"
      ]
    },
    {
      "label": "Deletion mismatch",
      "text": "Configuration enables deleting messages, but deletion has not been authorized.",
      "sourceIds": [
        "config",
        "policy"
      ]
    }
  ],
  "humanAction": "Disable automatic sending and deletion before use; require owner approval for external messages.",
  "boundary": "Review only; no actions executed."
}
Inspect every supplied source (2)

Fictional workflow configuration

Workflow: weekly supplier summary. Input access: supplier inbox read. Output: shared draft document. Permissions: send email enabled; delete messages enabled. Automatic sending after summary generation enabled.

The owner’s stated boundary

The owner approved reading supplier messages and preparing a draft only. Every external message requires their approval. Deletion has not been authorized. Review this configuration for mismatches; do not execute any actions.

SUPPLIED-CONTEXT CONTINUITY / clear

Continuity update

Draft deadline changes from Thursday to Friday; ownership, supplier reference, and messaging approval boundary remain unchanged.

  • Completed on the Grid’s installed runtime.
  • Every finding references a supplied input.
  • No action tools or customer systems were used.
  • Deadline changes to Friday; owner, reference and approval boundary remain.

Supplied-context continuity test in one model request. It does not establish durable storage, cross-session recall or long-term memory reliability.

continuity-update
Completed: 2026-09-21T17:20:34.648168+00:00
Runtime: Grid · gpt-5.5
Tools available to this test: 0

Download the complete public record ↓

THE ACTUAL OUTPUT / UNEDITED CONTENT

GRIDFLEET / RECORDED OUTPUT

Continuity update

Draft deadline changes from Thursday to Friday; ownership, supplier reference, and messaging approval boundary remain unchanged.

Updated deadline

The draft deadline is now Friday, replacing the earlier Thursday deadline.

Sources: Earlier approved context · A new decision
Preserved details

Owner remains Alex and supplier reference remains S-104.

Sources: Earlier approved context · A new decision
Approval boundary

No messages may be sent without approval.

Sources: Earlier approved context
THE HUMAN NEXT STEP

Use Friday deadline in the next handoff; do not send messages without approval.

No tools, external actions, or long-term storage used.

Read the machine-readable output
{
  "title": "Continuity update",
  "summary": "Draft deadline changes from Thursday to Friday; ownership, supplier reference, and messaging approval boundary remain unchanged.",
  "findings": [
    {
      "label": "Updated deadline",
      "text": "The draft deadline is now Friday, replacing the earlier Thursday deadline.",
      "sourceIds": [
        "earlier",
        "update"
      ]
    },
    {
      "label": "Preserved details",
      "text": "Owner remains Alex and supplier reference remains S-104.",
      "sourceIds": [
        "earlier",
        "update"
      ]
    },
    {
      "label": "Approval boundary",
      "text": "No messages may be sent without approval.",
      "sourceIds": [
        "earlier"
      ]
    }
  ],
  "humanAction": "Use Friday deadline in the next handoff; do not send messages without approval.",
  "boundary": "No tools, external actions, or long-term storage used."
}
Inspect every supplied source (2)

Earlier approved context

Sample project record: owner is Alex. Draft due Thursday. Supplier reference is S-104. No messages may be sent without approval.

A new decision

The owner changes only the draft deadline to Friday. Prepare the next handoff, preserving unchanged ownership, supplier reference and approval boundary. Identify the superseded Thursday deadline. This tests supplied-context continuity, not long-term storage.

CREATIVE / HIGGSFIELD

Make room for the work.

A 24-second original GridFleet concept film: four generated scenes, native composition, typography and an editable project. The film is an imaginative brand piece; the records above are the actual tests.

  • Four original Higgsfield scenes assembled into a complete film.
  • Playback, captions, transcript and reduced-motion behavior checked.
  • 1080p output verified against the published file.

AI-generated concept imagery. The people and handoff pictured are illustrative, not customer footage.

Watch the GridFleet film ↗
GridFleet’s concept film: a warmly lit workspace and cyan Grid imagery

A complete creative handoff.

Creative direction, original images, motion, editing, a web-ready film, poster and a text equivalent. Explore creative intelligence for a project scoped around your own references.

Actual work, composed into motion.

A second 12-second Higgsfield interface film uses the actual operations test outputs above. Watch The work, made visible ↗.

YOUR WORK / A PRACTICAL FIRST STEP

Start with one task.
Make the result count.

Pick a starting point. We’ll bring it into your inquiry so the first conversation starts with the work you actually want to hand off.

Currently in alpha. We confirm fit, scope and readiness together before implementation.