An agent attempts to send a message. What can we establish?
With Blanco Méndez: a workflow review and a report on permissions, tests and limits.
Preparing the local exercise. If it does not load, the records and sources remain available below.
THE FACTSSelect a step to change it
Change: Permission given
Change: Permission withdrawn
Change: Attempt
Change: Response
Change: Effect evidence
Invented data. Nothing is sent. Receipts are not authenticated.
Advanced mode · 17 more cases and event order
The first three are the starting cases. Change each fact from its timeline step. The stages do not invent times for the response or effect.
Change a fact. See what changes.
After evaluating, try moving the revocation. Permission may change; a 202 still does not demonstrate delivery. The timeline reads the same engine as the receipt.
The same execution, inspected.
Interface R2.3.1 · engine and contract 1.1.0. This revision changes presentation, not the rule. The JSON record describes the calculation; it does not prove delivery.
Current input
Evaluated output
D001 record · version 1.6.0. Engine, contract and scenarios: 1.1.0.
Retained test record: 2026-11-03T20:22:51.931Z · UTC. Opening this page does not rerun that suite.
20/20 corrected scenarios are valid; all eleven R1 counterexamples are rejected at evaluation and receipt generation. Ten additional authored controls and seven semantic negatives pass. Self-review; no real effects.
Your evaluated input and output are in “Developer”; the downloadable record uses those same data. These are authored checks, not independent validation.
Where this model stops answering.
D001 record · version 1.6.0. Engine, contract and scenarios: 1.1.0.
Retained test record: 2026-11-03T20:22:51.931Z · UTC. Opening this page does not rerun that suite.
20/20 corrected scenarios are valid; all eleven R1 counterexamples are rejected at evaluation and receipt generation. Ten additional authored controls and seven semantic negatives pass. Self-review; no real effects.
Your evaluated input and output are in “Developer”; the downloadable record uses those same data. These are authored checks, not independent validation.
Limits of this same case
Teaching model of one grant and one operation.
Clock and total order are assumed when marked known. Revocation wins ties.
A bound final receipt is accepted as synthetic input; it is not authenticated or cryptographically verified.
No real action is performed or enforced. This does not test concurrent revocation or production security.
Cases and expected answers are authored here: agreement with this corpus is not independent evaluation.
Retained errors
R1 retained eleven inputs conflicting with grant or revocation presence; contract 1.1.0 rejects them. R1 remains unchanged.
Cannot by itself distinguish a 202 followed by delivery from a 202 that never completes.
Does not authenticate final receipts or verify a real clock.
What would require revising the conclusion
Implementation falsifier: a valid input that violates the declared rule, or a contradictory input that still produces a receipt. These controls are not claimed to cover every case.
Not recorded in the original record; consult the publication history.
Public withdrawal date
No established withdrawal is recorded.
Hypothesis
A data match does not establish real payment. This sample produces one row per payment; it is not a complete accounting report.
Falsifier
An unretained criterion is not reconstructed retrospectively. Recover the original protocol before a new test.
Contract / engine / scenario versions
Not recorded / Not recorded / Not recorded
Denominator for this check
Not re-established from originals.
Previous publication history; the current record retains its own version.
CSV01 / 1.5.0 · public presence verified on 2026-11-04T15:21:07.766720+00:00. This is the web readback time, not an experiment run or the exact deployment time.
Your agent says it sent it. Can you demonstrate it?
Reported result; not independently validated
Separate permission, technical acceptance and evidenced effect in a synthetic case.
Main limitation: Teaching model of one grant and one operation.
View outcome, method and sources
Last retained result · evidentiary scope
20/20 corrected scenarios are valid; all eleven R1 counterexamples are rejected at evaluation and receipt generation. Ten additional authored controls and seven semantic negatives pass. Self-review; no real effects.
Change from the previous version
No change in claim, outcome, limitations or evidence state.
Next recorded uncertainty
Outside this model, confirmation authenticity, its binding to the operation and real event order remain unverified.
Execution date established in this record
2026-11-03T20:22:51.931Z
Sources and access status
RFC 9110 · meaning of 202 Accepted ↗Fielding, Nottingham and Reschke, IETF RFC 9110; paraphrase and link only, no extracted code · Public technical reference consulted
Dates and traceability
Publication date of this version
Not recorded in the original record; consult the publication history.
Public withdrawal date
No established withdrawal is recorded.
Hypothesis
In this teaching model, permission revoked before the attempt produces DENY; a 202 response without a final receipt leaves the effect UNKNOWN.
Falsifier
A valid input contradicting the stated rule, or a contradictory input producing a receipt.
Contract / engine / scenario versions
1.1.0 / 1.1.0 / 1.1.0
Denominator for this check
20
Previous publication history; the current record retains its own version.
D001 / 1.5.0 · public presence verified on 2026-11-04T15:21:07.766720+00:00. This is the web readback time, not an experiment run or the exact deployment time.
In this teaching model, permission revoked before the attempt produces DENY; a 202 response without a final receipt leaves the effect UNKNOWN.
Method
Contract 1.1.0, 20 revised synthetic inputs and additional authored controls fixed before execution. Contradictions are explicitly rejected; the engine does not normalize.
Expected outcome
Match the 20 authored answers, including ALLOW, DENY and UNKNOWN; effect evidence is evaluated separately.
Test scope
Teaching model of one grant and one operation, not a revocation implementation.
All recorded limitations
Teaching model of one grant and one operation.
Clock and total order are assumed when marked known. Revocation wins ties.
A bound final receipt is accepted as synthetic input; it is not authenticated or cryptographically verified.
No real action is performed or enforced. This does not test concurrent revocation or production security.
Cases and expected answers are authored here: agreement with this corpus is not independent evaluation.
Failures and open questions
R1 retained eleven inputs conflicting with grant or revocation presence; contract 1.1.0 rejects them. R1 remains unchanged.
Cannot by itself distinguish a 202 followed by delivery from a 202 that never completes.
Does not authenticate final receipts or verify a real clock.
Documented historical results 03
B01 / Record 1.6.0
A conflict test suite
Public artifact; validation limited to its scope
24 published cases about which stored information should govern.
Main limitation: Documentary transcription of the previous public record. This experiment was not rerun in R1.
View outcome, method and sources
Last retained result · evidentiary scope
Currentness v0.2 preserves 24 cases, code and results: latest-write-wins 10/24; authority-first 16/24; reference 24/24 against author labels. This is consistency within that corpus, not independent semantic validation. Later self-review found additional cases that challenge generalizing the resolver. See F01.
Change from the previous version
No change in claim, outcome, limitations or evidence state.
Next recorded uncertainty
Independent semantic validation and generalization. F01 retains the disputed interpretation.
Not recorded in the original record; consult the publication history.
Public withdrawal date
No established withdrawal is recorded.
Hypothesis
Currentness v0.2 preserves 24 cases, code and results: latest-write-wins 10/24; authority-first 16/24; reference 24/24 against author labels. This is consistency within that corpus, not independent semantic validation. Later self-review found additional cases that challenge generalizing the resolver. See F01.
Falsifier
An unretained criterion is not reconstructed retrospectively. Recover the original protocol before a new test.
Contract / engine / scenario versions
Not recorded / Not recorded / Not recorded
Denominator for this check
Not re-established from originals.
Previous publication history; the current record retains its own version.
B01 / 1.5.0 · public presence verified on 2026-11-04T15:21:07.766720+00:00. This is the web readback time, not an experiment run or the exact deployment time.
Read the already-published record, version and limitations. This is not a new execution.
Expected outcome
No new expected outcome is assigned to the historical experiment.
Test scope
Documentary review of a specific version; historical results remain attributed to their authors.
All recorded limitations
Documentary transcription of the previous public record. This experiment was not rerun in R1.
Self-review; not production validation or independent evaluation.
Failures and open questions
The implementation returns B; the author’s local reading of the contract expects UNKNOWN.
B02 / Record 1.6.0
An experiment lab
Reported result; not independently validated
Frozen protocols, synthetic data and replayable experiment records.
Main limitation: Documentary transcription of the previous public record. This experiment was not rerun in R1.
View outcome, method and sources
Last retained result · evidentiary scope
Experiment Foundry R2 has source code, evaluation records and a 17-scenario report. It is a synthetic laboratory. The R2 version reviewed here is recorded as R2_NOT_VALIDATED. It establishes neither real-world performance nor an independently qualified evaluator. See F02 for an R1 failure and the bounded R2 repair.
Change from the previous version
No change in claim, outcome, limitations or evidence state.
Next recorded uncertainty
Match the complete sources and independently validate the evaluator.
Not recorded in the original record; consult the publication history.
Public withdrawal date
No established withdrawal is recorded.
Hypothesis
Experiment Foundry R2 has source code, evaluation records and a 17-scenario report. It is a synthetic laboratory. The R2 version reviewed here is recorded as R2_NOT_VALIDATED. It establishes neither real-world performance nor an independently qualified evaluator. See F02 for an R1 failure and the bounded R2 repair.
Falsifier
An unretained criterion is not reconstructed retrospectively. Recover the original protocol before a new test.
Contract / engine / scenario versions
Not recorded / Not recorded / Not recorded
Denominator for this check
Not re-established from originals.
Previous publication history; the current record retains its own version.
B02 / 1.5.0 · public presence verified on 2026-11-04T15:21:07.766720+00:00. This is the web readback time, not an experiment run or the exact deployment time.
Read the already-published record, version and limitations. This is not a new execution.
Expected outcome
No new expected outcome is assigned to the historical experiment.
Test scope
Documentary review of a specific version; historical results remain attributed to their authors.
All recorded limitations
Documentary transcription of the previous public record. This experiment was not rerun in R1.
Self-review; not production validation or independent evaluation.
Original internal sources were not matched here; figures are claims in the historical record.
Failures and open questions
Independent validation and complete sources are pending.
B03 / Record 1.6.0
A source-reading prototype
Reported result; not independently validated
TRACE normalizes captured sources and can abstain from a decision.
Main limitation: Documentary transcription of the previous public record. This experiment was not rerun in R1.
View outcome, method and sources
Last retained result · evidentiary scope
TRACE V1.1 is a local, decoupled adapter with a prospective protocol. The retained suite reports 108/108 checks; the captured sample produced ABSTAIN. SELF_REVIEW, with partial source coverage. This is not native Core integration, predictive validation or evidence of economic utility. This card describes V1.1; later V1.2 work is separate.
Change from the previous version
No change in claim, outcome, limitations or evidence state.
Next recorded uncertainty
Predictive validation, complete coverage, integration and economic utility.
Not recorded in the original record; consult the publication history.
Public withdrawal date
No established withdrawal is recorded.
Hypothesis
TRACE V1.1 is a local, decoupled adapter with a prospective protocol. The retained suite reports 108/108 checks; the captured sample produced ABSTAIN. SELF_REVIEW, with partial source coverage. This is not native Core integration, predictive validation or evidence of economic utility. This card describes V1.1; later V1.2 work is separate.
Falsifier
An unretained criterion is not reconstructed retrospectively. Recover the original protocol before a new test.
Contract / engine / scenario versions
Not recorded / Not recorded / Not recorded
Denominator for this check
Not re-established from originals.
Previous publication history; the current record retains its own version.
B03 / 1.5.0 · public presence verified on 2026-11-04T15:21:07.766720+00:00. This is the web readback time, not an experiment run or the exact deployment time.
Read the already-published record, version and limitations. This is not a new execution.
Expected outcome
No new expected outcome is assigned to the historical experiment.
Test scope
Documentary review of a specific version; historical results remain attributed to their authors.
All recorded limitations
Documentary transcription of the previous public record. This experiment was not rerun in R1.
Self-review; not production validation or independent evaluation.
Original internal sources were not matched here; figures are claims in the historical record.
Failures and open questions
Independent validation and complete sources are pending.
Retained failures 03
F01 / Record 1.6.0
A passing suite was insufficient
Reported result; not independently validated
The implementation returns B; the author’s local reading of the contract expects UNKNOWN.
Main limitation: Disputed interpretation: not presented as a demonstrated vulnerability.
View outcome, method and sources
Last retained result · evidentiary scope
In the retained case, B corrects C and A remains unlinked. Output B is reproduced; UNKNOWN is the author’s expectation under local precedence. The 15 October review leaves open a rival global-priority interpretation and possible author misinterpretation. SELF_REVIEW / DISPUTED: a counterexample to the proposed criterion, not an unambiguous bug or independent adjudication. The original 24/24 does not settle this dispute.
Change from the previous version
No change in claim, outcome, limitations or evidence state.
Next recorded uncertainty
Resolve the alternative reading and independently assess it. No unequivocal bug is established.
Not recorded in the original record; consult the publication history.
Public withdrawal date
No established withdrawal is recorded.
Hypothesis
In the retained case, B corrects C and A remains unlinked. Output B is reproduced; UNKNOWN is the author’s expectation under local precedence. The 15 October review leaves open a rival global-priority interpretation and possible author misinterpretation. SELF_REVIEW / DISPUTED: a counterexample to the proposed criterion, not an unambiguous bug or independent adjudication. The original 24/24 does not settle this dispute.
Falsifier
An unretained criterion is not reconstructed retrospectively. Recover the original protocol before a new test.
Contract / engine / scenario versions
Not recorded / Not recorded / Not recorded
Denominator for this check
Not re-established from originals.
Previous publication history; the current record retains its own version.
F01 / 1.5.0 · public presence verified on 2026-11-04T15:21:07.766720+00:00. This is the web readback time, not an experiment run or the exact deployment time.
Read the already-published record, version and limitations. This is not a new execution.
Expected outcome
No new expected outcome is assigned to the historical experiment.
Test scope
Documentary review of a specific version; historical results remain attributed to their authors.
All recorded limitations
Documentary transcription of the previous public record. This experiment was not rerun in R1.
Self-review; not production validation or independent evaluation.
Disputed interpretation: not presented as a demonstrated vulnerability.
Failures and open questions
The implementation returns B; the author’s local reading of the contract expects UNKNOWN.
F02 / Record 1.6.0
Consistent reports were insufficient
Reported result; not independently validated
A historical Foundry R1 audit reports acceptance of fabricated reports.
Main limitation: Original internal sources were not matched here; figures are claims in the historical record.
View outcome, method and sources
Last retained result · evidentiary scope
The retained summary of the historical review states that an authorized falsifier supplied 11 self-consistent reports for a false-causality scenario, that R1 accepted them and that the honest control failed. That summary attributes to R2 the rejection of the retained attack as INVALID_EXPERIMENT, with the holdout closed. The historical review reported inspecting the R2 receipt and the R1 audit, without rerunning R1 or recovering its vulnerable-run stdout. In this website review, those originals have not been recovered and Foundry has not been rerun; these claims remain pending verification. R2 remains unvalidated.
Change from the previous version
No change in claim, outcome, limitations or evidence state.
Next recorded uncertainty
Recover original evidence and validate R2. The reported repair does not close that validation.
Not recorded in the original record; consult the publication history.
Public withdrawal date
No established withdrawal is recorded.
Hypothesis
The retained summary of the historical review states that an authorized falsifier supplied 11 self-consistent reports for a false-causality scenario, that R1 accepted them and that the honest control failed. That summary attributes to R2 the rejection of the retained attack as INVALID_EXPERIMENT, with the holdout closed. The historical review reported inspecting the R2 receipt and the R1 audit, without rerunning R1 or recovering its vulnerable-run stdout. In this website review, those originals have not been recovered and Foundry has not been rerun; these claims remain pending verification. R2 remains unvalidated.
Falsifier
An unretained criterion is not reconstructed retrospectively. Recover the original protocol before a new test.
Contract / engine / scenario versions
Not recorded / Not recorded / Not recorded
Denominator for this check
Not re-established from originals.
Previous publication history; the current record retains its own version.
F02 / 1.5.0 · public presence verified on 2026-11-04T15:21:07.766720+00:00. This is the web readback time, not an experiment run or the exact deployment time.
Read the already-published record, version and limitations. This is not a new execution.
Expected outcome
No new expected outcome is assigned to the historical experiment.
Test scope
Documentary review of a specific version; historical results remain attributed to their authors.
All recorded limitations
Documentary transcription of the previous public record. This experiment was not rerun in R1.
Self-review; not production validation or independent evaluation.
Original internal sources were not matched here; figures are claims in the historical record.
Failures and open questions
A historical Foundry R1 audit reports acceptance of fabricated reports.
F03 / Record 1.6.0
The validator incorrectly accepted null evidence
Reported result; not independently validated
Receipt Gate 1.0.0 set delivery_allowed=true for JSON null in a local evaluation.
Main limitation: Original internal sources were not matched here; figures are claims in the historical record.
View outcome, method and sources
Last retained result · evidentiary scope
The retained R1 evaluation records PASS_BOUNDED, delivery_allowed=true and evidence_grade=NONE for null. Each of two runs found three false acceptances among 36 contract negatives. The package verdict is KILL. This is a failure of that candidate validator, not evidence that AEGIS performed an external action. SELF_REVIEW.
Change from the previous version
No change in claim, outcome, limitations or evidence state.
Next recorded uncertainty
Match and assess beyond the bounded record. No production delivery system is established.
Not recorded in the original record; consult the publication history.
Public withdrawal date
No established withdrawal is recorded.
Hypothesis
The retained R1 evaluation records PASS_BOUNDED, delivery_allowed=true and evidence_grade=NONE for null. Each of two runs found three false acceptances among 36 contract negatives. The package verdict is KILL. This is a failure of that candidate validator, not evidence that AEGIS performed an external action. SELF_REVIEW.
Falsifier
An unretained criterion is not reconstructed retrospectively. Recover the original protocol before a new test.
Contract / engine / scenario versions
Not recorded / Not recorded / Not recorded
Denominator for this check
Not re-established from originals.
Previous publication history; the current record retains its own version.
F03 / 1.5.0 · public presence verified on 2026-11-04T15:21:07.766720+00:00. This is the web readback time, not an experiment run or the exact deployment time.