Certified into the record
The exact text we saved
This is the version exactly as it was saved, character for character. The SHA-256 hash below is a 64-character code worked out from that text. Run the same text through SHA-256 and you get the same code; change a single letter and the code comes out completely different. Each version also stores the previous version’s hash, which chains them together so no version can be swapped out quietly.
TRUVACE RECORD VERSION record: TRV-2026-3019 version: 1 kind: certified reason: Certified into the record timestamp: 2026-11-18T06:02:11.341326Z status: published lens: g_space sector: education headline: Human versus AI scoring of surgical mental practice in the FingerFix Smart® procedure dek: Introduction: Assessment of surgical procedural knowledge remains largely dependent on human expertise, which may limit reproducibility and scalability. This study evaluated whether generative artificial intelligence could score a surgical mental practice task from audio transcripts after video-based learning of FingerFix Smart® dynamic external fixator placement. Methods: Thirty-six medical students who had completed a rotation in a hand surgery department watched a reference video performed by an expert surgeo… gain_title: In a study of 36 medical students describing a hand-surgery fixator procedure, the fourth AI prompt scored their narrations within a mean bias of +0.92 points of an expert human grader, with an intraclass correlation of 0.940. problem_title: (none) trace_subject: (none) gain_reading: In a study of 36 medical students describing a hand-surgery fixator procedure, the fourth AI prompt scored their narrations within a mean bias of +0.92 points of an expert human grader, with an intraclass correlation of 0.940. gain_evidence: an absolute agreement intraclass correlation coefficient of 0.940 (95% CI, 0.884-0.969) | a mean bias of +0.92 points (95% CI, -0.09 to +1.93) problem_reading: (none) problem_evidence: (none) quick_read: Researchers tested whether generative AI could grade how well 36 medical students, all of whom had completed a hand surgery rotation, recalled placement of the FingerFix Smart dynamic external fixator. After watching an expert surgeon's video, the students described the procedure aloud, and the narrations were transcribed and scored against a 43-item checklist. The human grader gave a mean score of 28.42 out of 43. After the first two AI prompts under- and then overestimated those scores, the fourth prompt averaged 29.33 out of 43 and agreed closely with the human marks. Agreement was not perfect for individual students. The mean absolute error was 2.19 points, and the range in which most differences fell ran from -4.94 to +6.78 points. The prompts were refined and tested on the same transcripts, so the result does not show the method would hold on new students. limitation: Prompt development and scoring were done on the same transcripts, so the findings do not establish the reliability or generalizability of AI-based scoring; independent validation is required. tag: Evidence-backed gain key_points: The study involved 36 medical students who had finished a rotation in a hand surgery department. | Students watched an expert surgeon's reference video, then orally reproduced the FingerFix Smart placement procedure. | Narrations were transcribed and scored on a standardized 43-item checklist by an expert human and by AI. | The human mean score was 28.42 (SD 9.60) out of 43; the fourth AI prompt gave 29.33 (SD 8.20). | The first prompt underestimated human scores and the second overestimated them; two more prompts followed an item-by-item discrepancy analysis. | Mean absolute error was 2.19 points, and Bland-Altman limits of agreement ran from -4.94 to +6.78 points. rundown: The task tested surgical mental practice: after seeing an expert surgeon place a FingerFix Smart dynamic external fixator, each student described the steps aloud. Their transcribed narrations were scored on a 43-item checklist. An expert human evaluator's scores served as the benchmark. AI scoring used sequentially calibrated prompts. The first prompt scored too low and the second too high, so the team analysed disagreements item by item and wrote two further prompts. The fourth prompt came closest, at 29.33 versus 28.42 points on average, with an intraclass correlation of 0.940 (95% CI 0.884-0.969). The mean bias of +0.92 points had a confidence interval spanning zero (-0.09 to +1.93). Individual-level differences remained, with a mean absolute error of 2.19 points and limits of agreement from -4.94 to +6.78 points. The authors rate the level of evidence as III and say independent validation is required. sources: - peer_reviewed | Hand Surgery and Rehabilitation | https://doi.org/10.1016/j.hansur.2026.103056 | 2026-11-17 prev: 0000000000000000000000000000000000000000000000000000000000000000
- SHA-256 hash
- 383cbee08ebe24ed4c0bfbb1fa4723d046bc5aef3514ff4a04e475679ba25f71
- Previous version’s hash
- 0000000000000000000000000000000000000000000000000000000000000000
How to check it yourself, without trusting this page
Download any version from /api/record/TRV-2026-3019 and hash its canonical field yourself, for example with shasum -a 256. You should get its content_hash. Each version’s text also ends with prev:and the hash of the version before it (version 1 points to 64 zeros). If anyone had changed a single character of any version after it was saved, those hashes wouldn’t line up.
ace