{"entries":[{"lastCommit":{"author":{"email":"eth0net@pm.me","name":"eth0net","when":"2026-09-26T23:29:06.000000+01:00"},"committer":{"email":"eth0net@pm.me","name":"eth0net","when":"2026-09-26T23:29:06.000000+01:00"},"hash":"12950a7ce2542f48e8473fe630efa808b0589b10","message":"refactor(scanner): the art boxes move into the engine\n\nA browser cuts a card down to its artwork before it hashes one, so the\nrectangles have to be somewhere the wasm build reaches rather than in the\ntool that measured them.\n\nNothing it retrieves moves: 79 of 92 photographs, 65 of them answerable\nwithout asking, either side of the move.\n\nSigned-off-by: eth0net \n","tree":"251b2644c8e8eca5c67de9e3a12337bed35946e7"},"mode":"0100644","name":"art.rs","size":2296},{"lastCommit":{"author":{"email":"eth0net@pm.me","name":"eth0net","when":"2026-10-02T18:39:17.000000+01:00"},"committer":{"email":"eth0net@pm.me","name":"eth0net","when":"2026-10-04T10:58:24.000000+01:00"},"hash":"b187fef8275267acd7a266798fde52752221a0ee","message":"feat(scanner): a detection says which check turned a frame down\n\nOne pipeline rather than two: `card` is the quad a `Reading` answers with, so\na tally cannot drift from what the client runs. The measurements are kept\npast the check that refused them, which is what lets one pass say what a\nthreshold nobody shipped would have made of the same frames.\n\nTwo frames fail two checks each, because every other frame in the suite fails\nexactly one and the order the checks run in would otherwise be unpinned.\n`LEAST_FILL` had never been held to anything at all, and it is the only check\nthat refuses a photograph of a card on a table.\n\nSigned-off-by: eth0net \n","tree":"ad196fe39c9c793b986a76257f4332443e1f0337"},"mode":"0100644","name":"detect.rs","size":24219},{"lastCommit":{"author":{"email":"eth0net@pm.me","name":"eth0net","when":"2026-09-24T00:48:00.000000+01:00"},"committer":{"email":"eth0net@pm.me","name":"eth0net","when":"2026-09-24T13:49:31.000000+01:00"},"hash":"978df44ddbec38fc71217f9db8e666df76b4c619","message":"refactor(scanner): the hashing splits out of the build tool\n\nThe build fills the artwork index and a browser will query it. They have\nto produce the same bits from the same artwork, so the hash moves to a\ncrate with no image decoding, database or HTTP in it, and takes a luma\nplane and a stride rather than an `image` type. `manaweb-scan` keeps the\npull, the decoding and the measuring harness as `manaweb-artwork`.\n\nThat meant its own resampling and a cosine that does not vary by\nplatform, so the hashes are not the ones the old store holds. The store\ncarries what this build's hashing comes out as over a frame it states\nrather than ships, and refuses one another build filled — which matters\nbecause the tool keeps what a store already has and would otherwise end\nup holding half of each.\n\nSigned-off-by: eth0net \n","tree":"b8c7997f88f26c06f4b44f38f6583521fe966e34"},"mode":"0100644","name":"error.rs","size":293},{"lastCommit":{"author":{"email":"eth0net@pm.me","name":"eth0net","when":"2026-09-25T15:36:43.000000+01:00"},"committer":{"email":"eth0net@pm.me","name":"eth0net","when":"2026-09-26T22:08:40.000000+01:00"},"hash":"d65aea8659dfb93a77b609719bf0786d39bcdfff","message":"docs(scanner): the scanner is its own subject\n\nIt had grown to 341 lines of a 531-line roadmap, which is a file about\nphases carrying most of a file about one of them. Every pointer in the code\nwas aimed at the phase and now names the subject.\n\nSigned-off-by: eth0net \n","tree":"4d12ccb01da18e6625164f386c1697fbfa2b6569"},"mode":"0100644","name":"hash.rs","size":11912},{"lastCommit":{"author":{"email":"eth0net@pm.me","name":"eth0net","when":"2026-09-27T22:48:50.000000+01:00"},"committer":{"email":"eth0net@pm.me","name":"eth0net","when":"2026-09-27T22:48:50.000000+01:00"},"hash":"b8bc368b8ff363e2ac26e7b8d1eab46544da5169","message":"refactor(scanner): what a photograph asks the index moves into the engine\n\nWhich crops a scan is retrieved at — the card detection found read back both\nways up, or, where there is none, the framings it might have had — lived in\nthe tool that measured it. A browser has to ask the same ones, and two copies\nwould drift until the crops a scan used were the ones nobody had measured.\n\nRetrieval is unchanged, as with the boxes before it: 79 of 92 photographs,\n65 of them answerable without asking.\n\nSigned-off-by: eth0net \n","tree":"4f430cd87299323aa260845e8437f265d786d282"},"mode":"0100644","name":"lib.rs","size":805},{"lastCommit":{"author":{"email":"eth0net@pm.me","name":"eth0net","when":"2026-09-24T14:45:12.000000+01:00"},"committer":{"email":"eth0net@pm.me","name":"eth0net","when":"2026-09-24T16:33:29.000000+01:00"},"hash":"0953f8dc1ad8f6858a01a28e7e46855bd67ef8db","message":"fix(scanner): color reaches one channel inside the engine\n\nThe index was keyed by whatever the decoder crate's conversion happened to be, which is outside everything the fingerprint covers: a version bump moving it would have re-keyed newly hashed artworks into the same store as the old ones, silently, which is the failure the fingerprint exists to stop. The conversion is now the engine's own, stated and reachable from the browser, and the probe is in color so the word covers it.\n\nThe fingerprint moves to 1c1f64e991d4d1cc, so the store and the published index want rebuilding: just hash-art, then measure-art.\n\nSigned-off-by: eth0net \n","tree":"bfd49e6bf75a325ce90981d2d453c66256eeec7f"},"mode":"0100644","name":"luma.rs","size":1105},{"lastCommit":{"author":{"email":"eth0net@pm.me","name":"eth0net","when":"2026-10-02T18:39:30.000000+01:00"},"committer":{"email":"eth0net@pm.me","name":"eth0net","when":"2026-10-04T10:58:25.000000+01:00"},"hash":"eae68b54cec7ec1395b013fe069a79959cd97ff3","message":"docs(scanner): which stage drops a tilted card\n\nThe hull test, at every flood step from 4 to 14, and not the lean: a drawn\ncard clears every check out to 60 degrees, so what the refused frames have is\na soft outline rather than an angle. The open question moves to what\n`LEAST_FILL` at 0.86 would admit, which wants the pictures holding no card.\n\nA card past 44.3 degrees is read a quarter turn, and the detection having\nsucceeded is what stops the guessed framings being tried. What that costs has\nnot been measured, so it is written as what it is rather than as a verdict.\n\nSigned-off-by: eth0net \n","tree":"2f406e58425e9344b42b13702c0d4118f478fa89"},"mode":"0100644","name":"query.rs","size":3172},{"lastCommit":{"author":{"email":"eth0net@pm.me","name":"eth0net","when":"2026-09-24T00:48:00.000000+01:00"},"committer":{"email":"eth0net@pm.me","name":"eth0net","when":"2026-09-24T13:49:31.000000+01:00"},"hash":"978df44ddbec38fc71217f9db8e666df76b4c619","message":"refactor(scanner): the hashing splits out of the build tool\n\nThe build fills the artwork index and a browser will query it. They have\nto produce the same bits from the same artwork, so the hash moves to a\ncrate with no image decoding, database or HTTP in it, and takes a luma\nplane and a stride rather than an `image` type. `manaweb-scan` keeps the\npull, the decoding and the measuring harness as `manaweb-artwork`.\n\nThat meant its own resampling and a cosine that does not vary by\nplatform, so the hashes are not the ones the old store holds. The store\ncarries what this build's hashing comes out as over a frame it states\nrather than ships, and refuses one another build filled — which matters\nbecause the tool keeps what a store already has and would otherwise end\nup holding half of each.\n\nSigned-off-by: eth0net \n","tree":"b8c7997f88f26c06f4b44f38f6583521fe966e34"},"mode":"0100644","name":"store.rs","size":5057},{"lastCommit":{"author":{"email":"eth0net@pm.me","name":"eth0net","when":"2026-09-27T22:49:05.000000+01:00"},"committer":{"email":"eth0net@pm.me","name":"eth0net","when":"2026-09-27T22:49:05.000000+01:00"},"hash":"8b2e55a668085286b88a32132b2de46225f1d4c9","message":"feat(scanner): detection and rectification reach the browser\n\nThe engine exported color conversion and hashing and nothing else, so the\nwhole of what the photograph measurements are about stopped at the Rust\nboundary. `scan_detect` answers the four corners, `scan_rectify` reads a card\nback as a rectangle, and `scan_query` does the whole read in one call and\nsays whether a card was found or the framing was guessed at instead.\n\nHow big a rectified card comes back is the corners' to say rather than a\ncaller's, so `scan_rectify` writes the size before it asks for room, and\n`detect::size` makes that probe free rather than rectifying twice.\n\n`tools/scanner-check` now compares a drawn card as well as the probe frame,\nso detection, rectification and the query are held to what a native build\nanswers rather than only the hashing. The module goes from 45KB to 74KB.\n\nSigned-off-by: eth0net \n","tree":"aba701f1a5096dc461044e96818903e26a044e5f"},"mode":"0100644","name":"wasm.rs","size":9353}],"lastCommit":{"author":{"email":"eth0net@pm.me","name":"eth0net","when":"2026-10-02T18:39:30.000000+01:00"},"committer":{"email":"eth0net@pm.me","name":"eth0net","when":"2026-10-04T10:58:25.000000+01:00"},"hash":"eae68b54cec7ec1395b013fe069a79959cd97ff3","message":"docs(scanner): which stage drops a tilted card\n\nThe hull test, at every flood step from 4 to 14, and not the lean: a drawn\ncard clears every check out to 60 degrees, so what the refused frames have is\na soft outline rather than an angle. The open question moves to what\n`LEAST_FILL` at 0.86 would admit, which wants the pictures holding no card.\n\nA card past 44.3 degrees is read a quarter turn, and the detection having\nsucceeded is what stops the guessed framings being tried. What that costs has\nnot been measured, so it is written as what it is rather than as a verdict.\n\nSigned-off-by: eth0net \n","tree":"2f406e58425e9344b42b13702c0d4118f478fa89"}}