# Portable Agents Pi research demonstration

## Evidence status

This is a manually approved live-provider and public-network demonstration recorded on 2026-10-22. It uses independently authored fictional Northstar fixtures. Provider output is nondeterministic, the retrieved FAA material is dated to this run, and the recording does not establish semantic equivalence, production suitability, or full Open Responses conformance.

## Operations shown

1. Ordered admission of an image, PDF and XLSX workbook.
2. Bounded local workbook computation with `code_interpreter`.
3. Generated image observations through the declared OpenAI image recipient.
4. OpenAI-hosted web discovery followed by bounded public-HTTPS source reading.
5. An evidence brief that distinguishes declarations, observations, measurements, calculations, retrieved evidence, inference and uncertainty.
6. Progressive loading of `business-tech-presentation`, bounded seven-slide PowerPoint construction and digest-verified local save.

## Complete visible transcript

The text below is the plain-text conversion of the approved terminal recording. Blank terminal chrome is omitted by the converter; wording is otherwise retained.

~~~text

 Pi v1.0.4
 escape interrupt · ctrl+c/ctrl+d clear/exit · / commands · ! bash · ctrl+o more
 Press ctrl+o to show full startup help and loaded resources.

 Pi can explain its own features and look up its docs. Ask it how to use or extend Pi.


[Context]
  AGENTS.md

[Extensions]
  document-inspection.ts, input-file-attachment.ts, portable-model-routing.ts, presentation-artifact.ts,
server-tool-progress.ts


 [Invocation-local file attached: northstar-console.png; resend required on a later turn] [Invocation-local
 file attached: northstar-specification.pdf; resend required on a later turn] [Invocation-local file
 attached: northstar-measurements.xlsx; resend required on a later turn] Build a short evidence brief. Use
 image_understanding for the visible
  product label, inspect the attached specification and measurements, use web_search to find current official
 US guidance for portable-device battery capacity, use read_source on the most relevant official URL,
  and use code_interpreter to calculate the mean measured runtime. Separate generated image observations,
 attached declarations, measured values, retrieved public evidence, calculations, and anything still
  unverified. Do not invent citations or treat this fictional product as real.


┌ ✓ code_interpreter · server · completed · 0.81s
│ Input: Parse the measurement workbook and calculate the mean measured runtime.
│ Output: Python completed in 797 ms; retained 322 stdout and 0 stderr UTF-16 units.
│ Preview: Code and interpreter output are omitted from progress; inspect the tool result in model execution
│ context.
│ ▸ Expand for full admitted details and call IDs
└────────────────────────────────────────────────────────────────

┌ ✓ image_understanding · server · completed · 2.40s
│ Input: Invocation-local image analysis
│ Output: Generated bounded image observations via openai/gpt-5.4; not verified truth.
│ ▸ Expand for full admitted details and call IDs
└────────────────────────────────────────────────────────────────

┌ ✓ web_search · server · completed · 14.41s
│ Input: FAA lithium battery watt-hour limit portable electronic devices passengers
│ Output: Returned 17 source URLs; factual support is not verified.
│ 1. Source — www.faa.gov
│ 2. Source — www.faa.gov
│ … 15 more sources
│ Preview: Concise research notes on **FAA lithium battery watt-hour limits for passengers’ portable
│ electronic devices (PEDs)**  **Most directly relevant FAA source** - *…
│ ▸ Expand for full admitted details and call IDs
└────────────────────────────────────────────────────────────────

┌ ✓ read_source · server · completed · 1.08s
│ Input: www.faa.gov/hazmat/packsafe/airline-passengers-and-batteries
│ Output: Read text/html representation (18353 UTF-16 units); factual support is not verified.
│ 1. www.faa.gov/hazmat/packsafe/airline-passengers-and-batteries — www.faa.gov
│ Preview: The displayed URL omits its query and fragment; the tool result retains the authorized retrieval
│ identity.
│ ▸ Expand for full admitted details and call IDs
└────────────────────────────────────────────────────────────────

 Northstar Field Console evidence brief

 Northstar is a fictional device used only for testing. Nothing below treats it as a real product.

 1. What the image model saw

 - The label says "NORTHSTAR" and "Synthetic Field Console · Prototype A."
 - The image model couldn't read a model number, battery capacity (Wh), mass or any other rating on the
   label.
 - These are machine-generated readings of the image. They don't prove what the product is or that it's
   genuine.

 2. What the specification declares (Prototype A)

 - Battery: 72 Wh. Mass: 1.8 kg. Display: 14 inches.
 - Target runtime: 8 to 10 hours under the synthetic office workload.
 - The document itself says these figures are "not third-party verification."

 3. What the workbook measured

 ┌──────────────────┬─────────────┬─────────────┬────────┐
 │ Metric           │ Prototype A │ Prototype B │ Unit   │
 ├──────────────────┼─────────────┼─────────────┼────────┤
 │ Battery capacity │ 72          │ 64          │ Wh     │
 ├──────────────────┼─────────────┼─────────────┼────────┤
 │ Measured runtime │ 9           │ 7.5         │ hours  │
 ├──────────────────┼─────────────┼─────────────┼────────┤
 │ Mass             │ 1.8         │ 1.6         │ kg     │
 ├──────────────────┼─────────────┼─────────────┼────────┤
 │ Display          │ 14          │ 13          │ inches │
 └──────────────────┴─────────────┴─────────────┴────────┘

 Prototype A's measured battery, mass and display match the specification. Its 9-hour runtime falls inside
 the declared 8–10 hour target. The workbook doesn't say how the measurements were taken, how many test runs
 there were, or what workload was used.

 4. Calculation

 - I calculated the mean measured runtime from the workbook: (9 + 7.5) / 2 = 8.25 hours.
 - This averages two different prototypes, with one value each. It isn't a repeated-trial average for either
   one.

 5. Official public guidance (FAA, read in full)

 I read the FAA's PackSafe page "Airline Passengers and Batteries." It covers carrying batteries on US
 passenger flights, based on 49 CFR §175.10. It's not a general limit on battery capacity.
 - Rechargeable batteries up to 100 Wh are allowed on passenger aircraft. Batteries from 101 to 160 Wh need
   the airline's approval, and anything over 160 Wh is forbidden. 1
   (https://www.faa.gov/hazmat/packsafe/airline-passengers-and-batteries)
 - Laptops and similar lithium-ion devices follow the same Wh limits as spare batteries. They can be checked
   only if they're fully powered off and protected from short circuits and accidental activation. 1
   (https://www.faa.gov/hazmat/packsafe/airline-passengers-and-batteries)
 - If a battery doesn't show its Wh rating, multiply volts by amp-hours to get it. Airlines and the TSA may
   set stricter rules. 1 (https://www.faa.gov/hazmat/packsafe/airline-passengers-and-batteries)
 - The page showed a "last updated" date of October 2, 2026, when I retrieved it.

 6. My inference

 If the declared and measured 72 Wh is correct, Prototype A would be in the FAA's 0–100 Wh band, which
 doesn't need airline approval. Prototype B's measured 64 Wh would be in the same band. This is my
 comparison, not an FAA ruling, and the product is fictional.

 7. Still unverified

 - The label doesn't show a Wh rating, so the image doesn't support the 72 Wh figure.
 - The workbook doesn't show how its values were measured, and no independent tests exist.
 - There's no evidence that the product exists or that any airline would accept it.
 - Rules from the TSA, individual airlines or other countries weren't checked.


 make a short pptx deck about those findings


┌ ✓ business-tech-presentation · skill · completed · 0.00s
│ Input: Load reviewed presentation design procedure
│ Output: Loaded business-tech-presentation guidance for this invocation.
│ Preview: Loading the skill grants no authority; its reviewed builder dependency remains separately bounded.
│ ▸ Expand for full admitted details and call IDs
└────────────────────────────────────────────────────────────────

┌ ✓ presentation_builder · server · completed · 0.09s
│ Input: Build bounded business-technology presentation
│ Output: Built 7 bounded slides (36120 bytes): 7 composed slides, 21 gallery component instances across 11
│ distinct admitted component types; artifact art_10715d217c61fb28dc4c922661efc315 is ready for automatic
│ local save.
│ Preview: The short-lived artifact ID is disclosed to the reviewed Pi extension; slide content, filename and
│ digest remain omitted from progress.
│ ▸ Expand for full admitted details and call IDs
└────────────────────────────────────────────────────────────────

 I've built a 7-slide deck, northstar-evidence-brief.pptx. I haven't rendered or looked at the slides, so
 check them in a preview before you use the deck. I also can't confirm it's saved in the project folder until
 your client shows that it was.

 Slides:
 1. Title: labelled as a fictional test product.
 2. Main message: the 72 Wh claim matches across the specification and the workbook, but no one has checked
    it independently.
 3. What the image model saw: the label names the prototype but shows no battery rating.
 4. Prototype A summary: its battery (72 Wh), runtime (9 h), mass (1.8 kg) and display (14 in) all match the
    specification.
 5. Runtime bar chart: Prototype A at 9 h and Prototype B at 7.5 h. The 8.25 h mean is labelled as an average
    of two prototypes, not of repeated tests.
 6. FAA comparison: both prototypes fall in the up-to-100 Wh band, which needs no airline approval. This is
    labelled as my inference, not an FAA ruling.
 7. Still unverified: four gaps remain before anyone can rely on these claims.

 Notes:
 - No new checks: the deck reuses findings from my last answer. Your attachments weren't sent again this
   turn, so I didn't re-read them.
 - FAA date: the FAA page wasn't re-read either, so its "last updated October 2, 2026" date is noted on the
   slide as coming from the earlier check.
 - No label photo: the deck doesn't include the label photo, because the image wasn't available this turn.
 - No recommendation slide: the deck builder pointed out that there's no recommendation or next-steps slide.
   I left one out because this is a findings brief. I can add one, for example "commission independent
   capacity testing," if you'd like.

 File details: 36,120 bytes · SHA-256 a802aa4a08747087e2fa9036a700fd946ef5825f63aae66ff623e02b2146f9c2

 Saved northstar-evidence-brief.pptx · sha256
 a802aa4a08747087e2fa9036a700fd946ef5825f63aae66ff623e02b2146f9c2.

──────────────────────────────────────────────────────────────────────────────────────────────────────────────

──────────────────────────────────────────────────────────────────────────────────────────────────────────────
~/var/portable-agents-reference (main)
↑94k ↓5.1k 4.7%/1.0M (auto)                                                 anthropic/claude-opus-5-5 • medium
~~~

## Limitations

- Northstar Field Console is fictional.
- Image observations are generated, not verified.
- Search results become retrieved evidence only after a successful source read.
- The workbook contains synthetic values, not an external benchmark.
- The PowerPoint was structurally generated but was not rendered or visually inspected during the recorded session.
- The evidence applies only to the recorded versions, routes, configuration and date.