Signal 23.
Verification record.
A public index of the retained evidence and documented gaps for Gravity's October 31, 2026 Work / Burnout scan. Compiled on November 2, 2026 from existing records. It does not alter the original forecast, the HIT classification, or any October artifact.
01 / Retained artifacts
PRE-REVIEW FORECAST
October 30 advance note ↗. Prior to review, P≈0.95 (moderate; operator judgment informed by 21 earlier scored signals) was logged. Not statistically calibrated.
REVIEWED SCAN
October 31 final report ↗. Assignment → Output → Fallout, contrary evidence, causal limitations and HIT under rubric v2.1.
DATED SOURCE LOG
Signal 23 source snapshot (text) ↗. Source URLs, publication dates, figures, scope limits, three-check decisions and stated method.
PUBLICATION RECEIPT
October 31 publication receipt ↗. Review cutoff, recorded classification, forecast status and SHA-256 of the source log.
d7260f7549827ad58d4fa3809e798eff8060fa4302e42a364623d3bfeb35ab00
This hash matches the SHA-256 printed in the retained October 31 receipt. It authenticates the present source-log bytes against the receipt; it does not authenticate the external studies, establish source independence, or prove the classification.
02 / Recorded checks and evidence boundaries
| Check | Public basis in retained source log | Limits an independent reviewer should test |
|---|---|---|
| ASSIGNMENT | Hubstaff, published October 27: 46% of surveyed AI users reported higher expectations after adoption. | Survey represents a selected population; perceptions are not causal attribution. Test whether the promised completion was specific and within window. |
| OUTPUT | Hubstaff: 53% encountered work needing correction; 45% spent at least one hour weekly correcting colleagues' AI work. ActivTrak, October 29: AI users showed longer workday spans. | Surveys and behavioral monitoring have different samples and measures. They do not directly establish one linked causal chain. |
| FALLOUT | Hubstaff: 30% seriously considered quitting over workload; 69% reported burnout in the preceding year. | Different recall periods; results are not proof AI caused burnout or that the same individuals satisfy all three checks. |
| CONTRARY | Hubstaff: 39% finished earlier; Gallup reports faster task completion; 2025 four-day workweek study and March 2026 ActivTrak report show favorable outcomes. | Some contrary studies fall outside the fixed October 24–30 publication window and remain contextual, not new weekly carriers. |
Original decision: HIT under the published three-check rubric; no causal conclusion about AI and burnout, and no claim that this scan independently proves Completion → Sequence. Read the scoring rules ↗
03 / Source-level inspection
- Hubstaff — State of the Workweek 2026 (27 Oct 2026; self-report survey; assignment, rework, burnout/quit considerations, productive-time contrary finding)
- ActivTrak Productivity Lab (29 Oct 2026; longer measured workday spans; different source sample and period)
- Axios — cybersecurity response reporting (30 Oct 2026; illustrative interview context, not population estimate)
- Gallup — benefits of AI at work (prior contextual data; productivity contrary evidence)
- Nature Human Behaviour — workweek intervention (2025; favorable burnout outcomes, outside scan window)
- ActivTrak — 2026 State of the Workplace (March 2026; burnout-risk metric; non-comparable to survey prevalence)
These entries are transcribed from the retained source log; their current live webpages may have changed since October 2026. The source log contains publication details and observations, not immutable copies of the underlying reports.
04 / Production mode and unretained controls
RETAINED / DISCLOSED
GPT-6 named for AI-assisted retrieval and drafting. Operator adjudication: manual. Source log, report, advance note, cutoff and publication receipt retained. Original P≈0.95 judgmental forecast preserved. Source-log SHA-256 verified.
NOT RETAINED / NOT ESTABLISHED
Exact model version and fixed inference settings; verbatim prompt and prompt hash; verbatim AI outputs; independent multi-model comparative run, responses, objections or disagreements for this scan; captured full source-page bytes; a completed independent blind reclassification; calibrated forecast probability.
The Gravity Array lists multi-model intelligence evaluation in Stage 1. This Signal 23 index documents only the evidence available for this scan; it does not establish that a cross-model verification run was independently reproduced, or that every declared Stage 1 module has a retained published run.
05 / Independent recheck and future scan controls
Independent recheck possible now: Open the original forecast, source log and final report. Verify the listed source figures and publication dates; distinguish in-window carriers from earlier context; try to reproduce each of the three checks without assuming the samples or individual subjects are linked. Reassess the contrary evidence and compare W1/W2/W4 MISS and W3 OBSERVING routes in Protocol v2.1 ↗.
Prospective verification packet for future scans: Retain a dated receiver assignment, exact admissible-source window and inclusion/exclusion log; captured source snapshots and hashes; full model/version/settings/prompt or a NOT RETAINED label; per-model independent evaluations, disagreements and counter-signal objections; independent or blind reclassification when available; and a final operator adjudication with recorded reasons. Keep the original prediction fixed and report any calibration method before use.
Negative-control question: Could Assignment, continued activity and Fallout also appear together in a comparable domain or sample without the proposed completion-assignment mechanism? Record such comparators and whether all checks occur in the same population, time window and sequence of evidence. If comparability is inadequate, preserve W3 OBSERVING rather than manufacture a definite result. These are prospective audit questions, not retroactive amendments to the scoring rules.