SeriesFindings
Notes of record, numbered by year in order of release.
Each finding carries a release date, a version, the chapter it is filed under, and the SHA-256 digest of its Markdown source, which is published beside it. Corrections are recorded in the Errata rather than made silently. New findings are announced in the Atom feed.
| 2026.5 | 2026-09-03 | Unlock-path and verified-boot assessment of a locked MediaTek tablet, with a privacy-hardened stock configurationChapter 1 · Hardware Security · MEASUREDA hands-on assessment of a OnePlus Pad Go 2 (OPD2505, MediaTek Dimensity 7300, Android 16) established that no custom-ROM path exists — no build, no recovery, no functional OEM-unlock key, no host driver for the MediaTek fastboot interface — before any destructive command was issued. The device was instead factory-reset with Factory Reset Protection disarmed and configured with no vendor or Google account, a default-deny firewall, and a staged, reversible debloat plan. The stock firmware forces a network connection to activate and ships advertising-identifier and behavioural-tracing services that remain present with no account signed in; whether any of them transmits off-device is not established here and is left to a traffic-analysis follow-up. Every value was read from the device rather than from documentation. |
| 2026.4 | 2026-09-03 | A release gate for modelled results, and the provenance that backs itChapter 2 · Cryptographic Provenance · MODELLEDThe doctrine under which the research platform admits a result — a three-level claim status, an append-only Ed25519 ledger with four checks per entry, measurements signed at the instrument, a battery of release gates that must all pass, corpus pinning that reproduces everything from one commit, and a restore drill — together with the one thing provenance cannot do. |
| 2026.3 | 2026-09-03 | Learned inversion of modelled carbon Raman spectra, with a refusal gateChapter 3 · 2D Carbon Thin Films · MODELLEDOn the held-out split of the synthetic corpus of Finding 2026.2, an ensemble of gradient-boosted trees recovers crystallite size to 2.85% mean relative error and assigns the amorphisation stage with macro-F1 0.986, where the physics baseline fails at the turnover (901% mean relative error); a saturation detector used as a refusal signal lowers the error on corrupted rows without flagging a single clean one. Every number is a modelled result and supports no claim about real films. |
| 2026.2 | 2026-09-03 | A deterministic synthetic corpus for the amorphous-carbon Raman inverse problemChapter 3 · 2D Carbon Thin Films · MODELLEDA seeded, hash-pinned corpus of 3,400 synthetic Raman spectra of disordered and amorphous carbon, generated from a forward model after Ferrari and Robertson, with declared corruptions and five benchmark tasks. It exists to qualify an inversion pipeline before any measured spectrum is admitted; no physical film is represented. |
| 2026.1 | 2026-09-03 | Conventions for signed findings in this libraryChapter 2 · Cryptographic ProvenanceHow findings are numbered, versioned, digested, optionally signed, and corrected, and how a reader verifies that the finding they hold is the one that was released. |