# RECORDS-FRAMEWORK-001 — first operating close

## Disposition

A working experimental reporting pipeline now produces four statements, a linked
constructed correction, and successor/underwriting memoranda from a frozen Tablework
history. Conventional, neutral and Bulla-supported source routes produce the same
account. Conventional records already supply the facts needed for these questions.
No Bulla-specific information advantage or total preparation-cost saving is established.

The contribution of this implementation is an executable reporting convention:
a mandate remains attached to accepted outputs, consumed resources, outstanding
requests and unresolved evidence when its worker changes. Its independent reader
usefulness, economic benefit, generality and eventual adoption remain unmeasured.
This is not GAAP, an assurance opinion, financial statements of Bulla Labs, or an
underwriting offer. The local preview is the delivery boundary; no public launch.

## Historical close

The selected company export is native revision 45: ten accepted mandate outputs,
eleven of one hundred experimental call credits spent, eighty-nine remaining,
three process losses, and DEV-004-PUBLISH-01 still pending. Its one credit is already
included in spent. Tablework 0.1.8's exact archive is
535c14897f9b0f3c4558efda1017e1c81f36894934b2f1303d599d2694e7cdbc,
bound to product commit 2f45a1af6af89812929afdb9a8c3b65b082b67df.
The source freeze is an ordinary local copy, not remote attestation or complete
enterprise capture. Native events have revision order but no occurrence timestamps.

The selected 26 company-session manifests explicitly name TW-001 through TW-010 in
their role. Usage projections retain exact reported turn-completion counters and
original event-stream hashes without prompts or tool transcripts. This selection
omits unlabeled and otherwise unattributed work; it is not a census or total cost.
Reported counters and elapsed durations do not establish money paid, human effort,
exclusive CPU, demand, savings, profit or unique nonoverlapping economic allocation.
No source records from subsequent Tablework development enter this account.

Bulla's shadow contains 45 receipts, including 28 retrospective bootstrap captures.
The compiler checks retained served-byte hashes and native event joins. Original
verifier diagnostics remain historical evidence; it does not infer authentic authority
from them. The neutral adapter is copied unchanged from the other study, with its
source digest retained. All formats ultimately retain the ordinary native export;
the comparison is explicitly dependent reconstruction, not independent discovery.

## Prospective cohort and failure history

The protocol, fixed synthetic inputs, criteria, value assumptions and runner were
committed before workload execution. The first preparation failed because the
archive's tablework/ prefix was not accounted for during inventory comparison.
No mandate started. cohort-001 retains the start and failure; FREEZE-001.json and
runner-v1.py retain the original contract. AMENDMENT-001 corrects the directory
selection, and the second freeze precedes the six jobs in cohort-002.

The actual run used Ubuntu 24.04 and Python 3.12.3. Six scripted local jobs produced
six accepted outputs across eight product invocations. Two real executor wrappers
were SIGKILLed after output recording and before acknowledgement. Fresh review
processes consumed those retained outputs without rerunning Tablework. Two first
outputs deliberately used the wrong filter, failed the frozen acceptance criterion,
and were corrected. These are induced controls, not discovered Tablework defects.
All forty-two product member hashes were checked before and after execution.

Measured stage totals for the retained run:

| Stage | Seconds | Boundary |
|---|---:|---|
| Product invocation, including subprocess startup | 6.046181499 | Eight attempts |
| Review processes | 2.892725559 | Six non-interruption attempt reviews |
| Recovery review processes | 0.905126846 | Two post-interruption reviews |
| Writing retained attempt records | 0.042622063 | A narrow accounting operation |

Stages are separately attributed; recovery is not added a second time to review.
These are local monotonic elapsed measurements, not exclusive CPU or a performance
comparison. The guest is shared and has no benchmark-isolation claim. Timings of
other uninstrumented harness operations, human effort and the agent work building
this framework are missing. These totals are not total institutional costs.
The jobs invoke no model; zero runner tokens does not mean zero research inference.
No actual monetary revenue or complete monetary cost was measured.

## Decision exercises and correction

Both memoranda are project-authored deterministic exercises with author access,
not fresh-agent isolation trials or independent readers. The successor may inspect
records under separate authority but needs current owner authorization before a
new dispatch. It must reconcile the original pending request before considering
another attempt. Receiving an account does not transfer ownership or permission.

The observed underwriting answer is INSUFFICIENT_EVIDENCE. An explicitly hypothetical
USD scenario demonstrates how direct cost 2 and review cost 5 against value 20 yield
contribution 13; review cost 25 yields -7. Missing costs block a complete margin;
outstanding exposure yields a hold. The scenario demonstrates sensitivity, not a
loan decision, cash flow forecast or measured enterprise value.

The correction is a constructed later challenge to TW-010's review basis. It links
to the prior account and calls for rechecking while leaving native acceptance,
spent credits and the unrelated pending request intact. It is not a report of an
actual product defect. Receiptless claims and unauthorized/unattributed effects
are retained in the test casebook without being treated as adjudicated liabilities.

## What remains open

A common grammar can reproduce a conventional account without establishing that
it is worth maintaining. This sprint records packaging sizes and conversion steps;
implementation, maintenance, human interpretation and independent integration labor
remain unmeasured. The differently packed native, neutral and Bulla byte totals are
not like-for-like cost measures. One descriptive preparation observation is retained
with its exact earlier compiler bytes and digest; it predates the final source-binding
hardening and measures common normalization separately from account compilation.
It is not a final-version benchmark or a per-format total cost comparison. A reader study and prospective preparation-work
study are still needed before claiming an advantage.

No result amends INSTITUTIONAL-OVERHEAD-001, supplies its missing agent isolation,
completes its main trials, or changes stable Bulla/Tablework authorization semantics.
The route toward independent assurance, actual underwriting adoption and financing
products remains contingent on future evidence.

## Validation and review

The affected-domain suite covers reconciliation, native replay, loss/withdrawal/late
completion, duplicate and missing events, shared and independent allowance pools,
unknown economics, receiptless claims, false coverage, corrections and retained
cohort outcomes. Browser tests exercise statements, correction, funding sensitivities,
mobile/desktop rendering, keyboard operation, accessibility and download integrity.
R3 accounting/claims and implementation/evidence review records live in reviews/.
The final validation record reports actual commands and outcomes; a test or review
is counted only after it completes.