Research · Measurements

Self-Conformance Report No. 002

Did every material claim in GAIBRA's replies carry its passport — where it came from, what kind it is, and what is unknown about it? Every turn in a production period was measured here, and published with its limits.

All researchThe standard ECS-1

What this report is: the Epistemic Custody Standard, ECS-1, requires every material claim in an output to carry a passport stating where it came from, whether it was measured, estimated or inferred, when, and what is unknown about it. This report counts — from GAIBRA's own database — every turn stored in the period, and checks whether its claims carried their passports.

It is a self-declaration, not assessed by an independent body, and the standard itself is a working draft for public comment.

Item Value
Period (UTC) 2026-10-16 14:00–18:30
Scope 12 turns across 9 kinds of request from 8 minds — everything stored in the period, not a chosen sample
Standard ECS-1 v0.6.1 · doi:10.5281/zenodo.23416950
Level Level 1 "Passported"
Result Passed — with a declared limit in §7

1. Turns

Turns Passed Failed No material claim Violations
12 11 0 1 0

The turn that "does not apply" is a general conversation with no figure and no verdict, so nothing in it needs a passport — that is its correct verdict, not a gap.

2. Claims by kind

Claims With a valid passport Measured Estimated Inferred Declared unknowns Custody entries
111 111 16 0 95 0 ⚠ 331

The ratios are coverage, not degrees of confidence. "Inferred" dominates as expected: most questions in the period were described in words (correcting a sentence · why a state fell · a growth figure the asker gave), so what was built on them is declared inferred, as it is. "Measured" came where live data was fetched: a stock analysis (9) and a strategic decision (7).

⚠ Unknowns: see §7 — zero here is not a certificate that nothing is unknown.

3. Minds and kinds of request

Kind of request Mind Turns Passed Does not apply Claims Of which measured
Linguistic analysis Linguist 3 3 0 36 0
Strategic decision Executive council 1 1 0 27 7
Stock analysis Investor 1 1 0 25 9
History of Islamic civilisation Historian 1 1 0 8 0
Writing code Developer 1 1 0 6 0
Data analysis Analyst 1 1 0 4 0
Fixing a bug Developer 2 2 0 3 0
A scientific question Scientist 1 1 0 2 0
General conversation General 1 0 1 0 0
Total 8 minds 12 11 1 111 16

The three linguistic-analysis turns came from the system shutdown drill: before it, during it, and after the return.

4. Passports by source

Source Kind Count Declared unknowns Seal failures
Data fetched from an external source Measured 1 0 0
An engine's computation Inferred 53 0 0
Live material Measured 15 0 0
What the user wrote Inferred 42 0 0

5. Conformance checks

Check Result Evidence
Structure of passports and custody entries ✓ Passed Zero violations in the period
Seal integrity (a hash that does not accept change) ✓ Passed Zero failures on storing, zero on verifying
Derivation rules ✓ Passed Zero violations · 162 derivation entries
Chain of custody: no upgrade in transit ✓ Passed Zero broken custody; no claim was raised above its kind in any turn
A passport for every material claim ✓ Passed 111 of 111
Deliberately wrong cases are rejected ✓ Passed 12 of 12 rejected at every system start in the period (three times)
An unknown extension is ignored, not broken ✓ Passed Zero violations
Declaring material unknowns ⚠ Declared limit §7 — first item

6. Chain of custody and the executive council

Action Entries Seal failures Verification failures on receipt
Derive 162 0 0
Forward 169 0 0

Information crossing between the executive council's members — a declared limit in report 001 — is now recorded: three members (risk officer · chief executive · finance director), each with its own crossing entry, none abstaining. In this turn the council declared the data insufficient and did not compose a single verdict — and that is written as it is.

7. Declared limits

  • ⚠ Unknowns the reply declares are not yet carried into the passports. Unknowns are zero across the twelve turns, although the replies themselves declare gaps to the user (the weakest confidence axis · the context question in linguistic analysis · the council declaring the data insufficient). The gap is said in the phrasing and not written in the passport, and the claims built on it do not inherit it. The checker passes this because it checks the passport's structure, not the completeness of what it declares. So zero unknowns in this report does not mean nothing is unknown.
  • The kind "estimated" did not occur in the period: what these requests said about the future was declared "inferred". GAIBRA is not claimed to estimate quantitatively in them.
  • A passport's precision today is a description and an engine result, not a single number.
  • The passport's passage to the language model that phrases the reply is recorded, but the model dropping a qualifier such as "estimated" while phrasing is not yet detected automatically.
  • What the language model writes outside the verdict object has no passport; this is declared at the level of the output.
  • Each engine's version is not yet declared in the passport.
  • Scope: 9 kinds of request from 8 minds were measured live in this period. "All of GAIBRA's services" is not written — what was not requested in the period was not measured.
  • A self-declaration not assessed by an independent body, and the standard is still a draft — so "conformance" is not claimed.

This report attests to the standing of the information — where it came from and what kind it is — not to its truth. It is published whether the period passed or not, and with its limits.