Process evidence · Redacted work-in-progress annex
The record behind the journey.
An assembly of separate dated operations, an illustrative journey and historical proof cases. Source hashes identify retained private bytes, not independent attestations or current deployment. No experiments were rerun for this report.
Records below are paraphrased except for named titles. Personal identities, original human quotations and internal paths are withheld. Full sources and correlation identifiers are retained in the private operator guide. An agent’s historical readback is attributed, not represented as an independent laboratory replication.
P1 · Current saved scope; selected-field projection
Existing POC scope and Response Table readback
The saved goal is to show real system operations and correction work through existing Signal Capture and a draft-only feed. Its expected output is a recorded operation, correlated capture, public-safe draft and snapshot walkthrough. Publication remains the operator’s decision.
Limit: This is the governing objective and saved work record. It does not certify completion of every linked task.
Selected facts paraphrased; personal identifiers, original human quotations, internal paths and record identifiers omitted. Full sources retained privately.
P2 · Presentation specification, including illustration and fixture
Original partner presentation script
The original journey is: arrive, orient, define the boundary, trace the path, change a condition, inspect the evidence, improve with review, invite evaluation. The main journey is explicitly illustrative. Its fixed induction fixture and recorded correction case are separate exhibits.
Limit: An illustrative boundary stop is not an observed production refusal; a fixture is not an unfamiliar-agent benchmark.
Selected facts paraphrased; personal identifiers, original human quotations, internal paths and record identifiers omitted. Full sources retained privately.
The capture assignment is to preserve real system activity as the operator uses the products. Capture infrastructure comes first, selection second, automation third, and blogging follows when requested. The subsequent standing procedure is attended capture-to-draft.
Limit: Capture readiness, a saved source signal, an editorial draft and publication are different outcomes.
Selected facts paraphrased; personal identifiers, original human quotations, internal paths and record identifiers omitted. Full sources retained privately.
A real consumer-interface defect lost the selected task across reload. The owner’s packet traces the diagnosis, staged repair, operator commit, deployment and hosted owner-view readback.
Limit: This is the owner’s dated evidence. It does not establish member-view acceptance or hosted-send acceptance.
Selected facts paraphrased; personal identifiers, original human quotations, internal paths and record identifiers omitted. Full sources retained privately.
The capture seat reported one source signal and one linked draft for the selection-persistence fix. Attended intake preserved the real operation; a duplicate submission resolved to the same signal.
Limit: This is an attributed capture-seat report. It does not make the operation an automatically captured or universal-access demonstration.
Selected facts paraphrased; personal identifiers, original human quotations, internal paths and record identifiers omitted. Full sources retained privately.
R2 · Selected original chat export; OQS-reported readback
Authorized completion checkpoint
On September 30 UTC, human execution approval was recorded at 03:25:18. The operation completed at 03:25:39 and the automatic capture followed at 03:25:40. The checkpoint identifies completion and paired-response evidence.
Limit: This is a retained OQS checkpoint, not a fresh rerun. Its bounded event search does not claim to contain every event.
Selected facts paraphrased; personal identifiers, original human quotations, internal paths and record identifiers omitted. Full sources retained privately.
R2C · Capture-seat source, dedupe, draft and Response Table readbacks
Completion-to-capture return
The capture seat found the already-persisted automatic signal, verified exactly one matching source, created one draft and read back the appended Response Table entries. The operation was not replayed or re-ingested. The process-only task returned before model dispatch.
Limit: One local completion-to-capture path; attended editorial preparation. No inference retry, autonomous repair, hosted execution or continuing coverage is established.
Selected facts paraphrased; personal identifiers, original human quotations, internal paths and record identifiers omitted. Full sources retained privately.
One request for each of three seats followed the recorded wake, attended readiness arm, dispatch and posted-answer path. Replies were recorded at 06:14:20, 06:44:20 and 07:17:45 UTC on October 2. A wake alone did not dispatch.
Limit: One read-only trial per seat on one workstation. The report is historical evidence, not a current watcher-health check or proof that every request succeeds.
Selected facts paraphrased; personal identifiers, original human quotations, internal paths and record identifiers omitted. Full sources retained privately.
R3C · Capture-seat evidence qualification and draft return
Three-seat communication capture return
The capture account retains the human send-and-arm boundary, a posted reply from each seat, and a draft describing the one-desk communication process. The replies acknowledged receipt without claiming independent verification of the pickup mechanism.
Limit: The source describes local spool evidence without central Flight Recorder IDs and an open citation-gate watcher error. No autonomous operation or reliability rate is claimed.
Selected facts paraphrased; personal identifiers, original human quotations, internal paths and record identifiers omitted. Full sources retained privately.
R4 · Retained request, response and attended review artifact
Corrected TAG Relay exchange
An initial presentation answer invented terminology and omitted requested content. It was rejected and retained. A corrective request supplied six presentation beats; the later local answer returned all six, but omitted an evidence sentence and used an unsupported processing label. It was retained as an editorial draft with stated limits.
Limit: This demonstrates dialogue transport and attended review. It is not a controlled same-prompt experiment, independent file access, task execution or automatic insertion of a Relay answer into the Response Table.
Selected facts paraphrased; personal identifiers, original human quotations, internal paths and record identifiers omitted. Full sources retained privately.
The attended review was saved on the existing topic and visible in its Response Table. The separate discussion request arrived through an assembly channel that the TAG discussion-reply writer did not accept.
Limit: The recorded Relay thread and discussion thread are different. Visibility of an attended review is not a saved native discussion return.
Selected facts paraphrased; personal identifiers, original human quotations, internal paths and record identifiers omitted. Full sources retained privately.
H1 · Historical local case and separate approval readback references
September approved application correction
A terse refusal lacked a rationale and safe alternatives. The correction was held, separately approved and applied by the application. A later governed response retained the refusal and added the explanation and alternatives; the raw model answer remained terse.
Limit: The changed output was deterministic application rendering. This is not evidence of model learning, retraining or a changed raw model answer. Earlier unsuccessful attempts remain part of the record.
Selected facts paraphrased; personal identifiers, original human quotations, internal paths and record identifiers omitted. Full sources retained privately.
H1R · Selected facts checked directly in a retained JSON export
September correction receipt
The saved receipt contains equal, nonempty active and rendered correction lists. Its approval claim is unmeasured; the separate approval readback supplies approval evidence.
Limit: Model execution is browser-reported, not independently attested. A receipt alone does not establish the approval facts.
Selected facts paraphrased; personal identifiers, original human quotations, internal paths and record identifiers omitted. Full sources retained privately.
H2 · Separate historical acceptance report with negative attempt retained
September held-checkpoint case
An earlier attempt failed the chronology and scope conditions. A later controlled case held a task, approved one new correction, reused the retained response and completed with a paired-response record. Its recorded event chain contained one model dispatch.
Limit: This is a different task from the application-correction case. The one-dispatch observation belongs only to this controlled checkpoint case. It is not a saving, latency or reliability result.
Selected facts paraphrased; personal identifiers, original human quotations, internal paths and record identifiers omitted. Full sources retained privately.