TEST FIXTURE FAILURE-ORACLE WORKSHEET
Reusable decision record for sample data and controlled mutations

Written by Alfred.
Original text-only worksheet. No third-party media or customer data.

PURPOSE
Turn sample data into a scoped test contract. Complete expectations before
observing the implementation. Keep format validity, product acceptance,
preservation, and recovery as separate results.

Do not use a blank worksheet as evidence. A completed worksheet is still only
as strong as its candidate identity, procedure, observations, and comparison.

======================================================================
A. DECISION ENVELOPE
======================================================================

worksheet revision:
reviewer role (optional, no personal details):
observed at (include timezone):
implementation revision:
execution environment:

fixture identity:
  filename or neutral label:
  immutable digest or revision:
  byte length:
  provenance / rights basis:

condition under test:
  one decision-relevant rule or interaction:

setup:
  parser / importer mode:
  schema or rule-set revision:
  locale and timezone:
  feature flags:
  starting-state revision:
  other frozen inputs:

out of scope:

======================================================================
B. EXPECTATION — FREEZE BEFORE THE RUN
======================================================================

expected interpretation:
  normalized records:
  warnings:
  conflicts:
  proposed effects:

forbidden outcomes:
  [ ] silent identity merge
  [ ] silent identifier coercion
  [ ] invented default
  [ ] omitted or hidden rejected record
  [ ] preview / commit mismatch
  [ ] durable effect after rejection
  [ ] duplicate effect after retry
  [ ] other:

failure oracle:
  If the targeted assumption is broken, the smallest decision-relevant
  observation that must make this lane fail is:

expected rejection or hold reason (if applicable):

preservation rule:
  source material that must remain unchanged:
  starting state that must remain unchanged:
  proposal state that must remain unchanged:
  durable effects that must not occur:

allowable variation:
  List only differences that may vary without changing the decision.

comparison points:
  [ ] received-byte record
  [ ] parsed-row ledger
  [ ] normalized-record ledger
  [ ] warning / conflict ledger
  [ ] proposal diff
  [ ] rejected-row set
  [ ] durable-effect ledger
  [ ] retry / deduplication record
  [ ] unchanged-state snapshot
  [ ] typed final decision and reason
  [ ] other:

======================================================================
C. LAYER RESULTS
======================================================================

Use only PASS, FAIL, BLOCKED, NOT TESTED, or SUPERSEDED.
A pass applies only to the identified fixture, setup, procedure, and lane.

transport
  expected:
  observed:
  evidence:
  result:

syntax
  expected:
  observed:
  evidence:
  result:

shape
  expected:
  observed:
  evidence:
  result:

interpretation
  expected:
  observed:
  evidence:
  result:

policy
  expected:
  observed:
  evidence:
  result:

proposal / preview
  expected:
  observed:
  evidence:
  result:

effect / commit
  expected:
  observed:
  evidence:
  result:

recovery / retry
  expected:
  observed:
  evidence:
  result:

======================================================================
D. REJECTION PLUS PRESERVATION
======================================================================

Was the declared rejection / hold produced?       result:
Did it identify the intended condition?           result:
Were all prohibited durable effects absent?       result:
Was the starting state reconciled independently?  result:
Were source bytes preserved?                      result:
Was an accepted proposal identity avoided?        result:

Do not mark preservation PASS because an error appeared. Compare the named
state and effect surfaces. If a required surface is unavailable, use BLOCKED.

======================================================================
E. CONTROLLED MUTATION RECORD
======================================================================

base fixture identity:
base expected decision:

mutation identity:
exact edit:
intended single condition:
new immutable digest or revision:
expected interpretation change:
expected decision change:
forbidden collateral change:

Did the mutation remain scoped?                    result:
Did any unrelated layer change?                    result:
Evidence:

One byte changed does not necessarily mean one semantic condition changed.
Inspect the resulting parse and interpretation before calling a mutation
controlled.

======================================================================
F. PREVIEW-TO-COMMIT CONTINUITY
======================================================================

source digest:
parser / rule-set revision:
normalized proposal digest:
starting-state revision:
accepted proposal identity:
commit receipt identity:

[ ] Commit candidate used the reviewed source digest.
[ ] Commit candidate used the reviewed rule-set revision.
[ ] Commit candidate used the reviewed starting-state revision.
[ ] Commit candidate used the accepted proposal identity.
[ ] Applied effects were reconciled against the accepted proposal.
[ ] An uncertain response was reconciled before retry.

continuity result:
continuity evidence:

A matching digest establishes identity only for the hashed material. It does
not prove semantic correctness, rights, usability, security, or effect parity.

======================================================================
G. FAILURE-SHAPED TRANSITIONS
======================================================================

For each relevant row, record expected decision, observation, evidence, and
one allowed result state.

1. source bytes change after preview
   expected decision:
   observed:
   evidence:
   result:

2. parser or rules change after preview
   expected decision:
   observed:
   evidence:
   result:

3. starting state changes after preview
   expected decision:
   observed:
   evidence:
   result:

4. proposal expires
   expected decision:
   observed:
   evidence:
   result:

5. a different proposal identity reaches commit
   expected decision:
   observed:
   evidence:
   result:

6. process stops after staging but before durable effect
   expected decision:
   observed:
   evidence:
   result:

7. commit response is uncertain, then retry is considered
   expected decision:
   observed:
   evidence:
   result:

8. accepted commit is deliberately replayed
   expected decision:
   observed:
   evidence:
   result:

======================================================================
H. FINAL DECISION
======================================================================

required lanes:
non-pass lanes:
conflicting evidence:
unresolved preservation question:
superseded evidence:

final decision — choose one:
  [ ] PROCEED — every required lane passed for this scope
  [ ] HOLD — evidence is blocked, incomplete, stale, or awaiting review
  [ ] REJECT — an expectation failed or a forbidden outcome occurred

reason:
claim this record supports:
claims this record does not support:
next recheck trigger:

======================================================================
I. COMPLETION CHECK
======================================================================

[ ] Exact fixture bytes or revision are identified.
[ ] One primary condition under test is named.
[ ] Setup and starting state are frozen.
[ ] Expected interpretation was written before observation.
[ ] Forbidden silent transformations and effects are listed.
[ ] The smallest useful failure oracle is explicit.
[ ] Rejection and unchanged-state evidence are both checked.
[ ] All eight result layers remain separate.
[ ] Controlled mutations were inspected for unintended semantic changes.
[ ] Preview and commit identities are bound where relevant.
[ ] Interruption, stale proposal, and retry paths are covered where relevant.
[ ] FAIL, BLOCKED, NOT TESTED, and SUPERSEDED remain visible.
[ ] Actual results were not copied into missing expected values.
[ ] Evidence locations are named without secrets or personal information.
[ ] The conclusion is limited to the identified candidate and observation.

BOUNDARIES
This worksheet does not prove coverage of every valid or invalid input,
security, accessibility, legal compliance, production safety, or suitability
for high-consequence decisions. Synthetic fixtures do not reproduce every
real-data distribution, ambiguity, scale, or history. Review the expected
behavior with appropriate domain expertise where consequences require it.