Skip to content

18-scoring.md

Latest commit

 

History

History
216 lines (172 loc) · 14.5 KB

18-scoring.md

File metadata and controls

216 lines (172 loc) · 14.5 KB
Raw

18 — The scoring: 64 bets, 71.09 %, and the hit count called exactly for the second session running

Measure: the verdicts below are counted out of the tables by python tools/checkscore.py docs/18-scoring.md, not added up by hand. The predictions are chapter 00 and have not been edited since they were written, which was before Mico02CD.bin was opened.

The totals

hits halves misses unresolved score
method (9) 7 1 1 0 83.33 %
content (55) 34 8 12 1 69.09 %
all (64) 41 9 13 1 71.09 %

A half counts 0.5. 41 + 4.5 = 45.5 of 64.

python tools/checkscore.py docs/18-scoring.md

  hit 41 · half 9 · miss 13 · unresolved 1      rows parsed: 64
  resolved 63 ; halves at 0.5 : 45.5 of 63 = 72.2 %

checkscore.py drops the unresolved clause from the denominator and reports 72.2 %. This chapter's headline keeps it in — 45.5 of 64 = 71.09 % — because a bet that cannot be settled is a bet that was written badly and the writer should pay for it. The difference is 1.1 points and it is one clause, C10.

Two known artefacts of the inherited tool, declared rather than tuned away: it was written for clauses numbered Pnn and its regular expression was widened in an earlier session to accept Cnn; and it prints a line for P93, scored elsewhere, a deferred clause belonging to a different repository, so its grand total reads 65 where this repository has 64. Chapter 16.

Method beats content by 14.2 points, which is the largest gap this branch has recorded, and it should be large: eight of the nine method clauses were bets about how a measurement would behave on an image whose format the tools had already been written for, and the fifty-five content clauses were bets about what was inside a disc nobody had opened.

The frame

# verdict what happened
C01 hit full census: 349,462 of 349,462 Mode 2 Form 1, zero Form 2, no mode byte but 02
C02 hit header MSF = LBA + 150 on all 349,462, no exception, 00:02:00 → 77:41:36
C03 hit the two subheader copies disagree on 0 sectors — 100 %, against a called 99 %
C04 half EOF-bit sectors 1,429, inside the called 1,400–1,650; but 1,429 against 1,527 extents is 6.42 % off, and the clause called 5 %. The better statement is the one the clause did not make: 1,422 sectors carry 0x88 and there are exactly 1,422 distinct file extents
C05 hit submode 0x00 on 155 sectors, called 100–200, and they are the tail run and nothing else
C06 hit EDC validated against the scalar reference on 128 sectors at three places, then 0 mismatches over 349,462
C07 half the tail's stored EDC does equal its computed EDC — but only because both are zero, and the ECC field is zero too. The writer computed it is disproved. The clause's testable half held and its reasoning did not
C08 hit 349,462 × 304 = 106,236,448 = 12.9252 %; × 2,352 = 821,934,624, remainder 0

The tail

# verdict what happened
C09 hit all 155: valid sync, correct progressive MSF, all-zero user area, censused not sampled
C10 unresolved 155 = 150 + 5 is available and consistent and nothing in the bytes demonstrates it. All 155 sectors are identical apart from their addresses. Carried to chapter 17 as Q7
C11 miss Q1 does not close. 150 is shown to be the medium's, but this disc is neither Toast nor a hybrid, so it separates nothing and the 2 stays unattributed. The clause claimed a closure it could not deliver
C12 hit no non-zero byte anywhere in the 317,440 bytes past LBA 349,306. (The clause's framing — "beyond volume space" — was wrong; there is no beyond. The measurement it named was right. Logged as M6)
C13 hit 157 → exactly 155, one run, after the root-extent fix

The two namespaces

# verdict what happened
C14 hit the mechanical Level 1 rule explains 1,423 of 1,423 = 100 %, against a called 90 % of the non-case differences
C15 hit 86 names carry a numeric disambiguator, inside the called 10–250
C16 miss 29 colliding groups, against a called ≥40. The clause over-estimated because it did not expect the collisions to be concentrated in two font directories
C17 hit 0 of 1,423 records violate Level 1 8.3; a DOS opens every file
C18 hit maximum depth 5 levels below the root, against a limit of 8
C19 hit both trees extracted: 1,423 files, 713,496,612 bytes, 1,400 distinct sha1 on each side, 0 hashes on one side only, 0 differing multiplicities
C20 hit the partition is exact: 779 case-only paths, 644 others, and zero exceptions — every case-only path has every component already fitting 8.3, and no other path does

CD-XA

# verdict what happened
C21 hit XA flags 0x0000, the 38 bytes after all zero: no Form 2, no interleaving
C22 hit 1,527 of 1,527 = 100 % carry the 14-byte XA system-use area, signature verified
C23 hit every file number 0, every owner id 0, 0x0D55/0x8D55 declaring mode 2 form 1; nothing claims audio, video, interleaved or CD-DA
C24 miss cdxa.py does not run on this at all. It reads Microsoft's RIFF/CDXA file container and there are none here; the XA layer of a disc is in the descriptor and the directory records, and reading it took a new 200-line tool. The clause's premise — that the briefing was right about which tool applied — was not checked

The slack

# verdict what happened
C25 hit 0 of 1,422 files with a non-zero last-sector tail, against a called ≤20
C26 hit Q5 closes with it was Toast
C27 hit 1,263,580 slack bytes, inside the called 500,000–3,000,000, and 100.0000 % zero

The declarations

# verdict what happened
C28 miss the sleeve's totals are exactly 7/11/4/10, so the clause's falsifiable half failed; and the mechanism it named was backwards. 28 is not the sleeve minus the runtimes, it is the directories including them: 11 + 4 + 13. The right answer was reachable and the clause pointed the other way
C29 hit 21 of 31 versions confirmed overall, and 11 of the 14 the clause named, against a called 10
C30 miss 6 of 13 Italian flags confirmed, against a called 9. None is refuted; seven are unverifiable from outside a compressed installer, which the clause did not allow for
C31 miss SpyBotSD/ is dated 2005-12-13 14:54:43, nine days before the bulk copy and eight seconds after Gtk/. It did not arrive late
C32 miss there is no GTK application on this disc and the disc's own page names no dependent. GIMP is not here
C33 half it is a Sun JRE (jre-1_5_0_06-windows-i586-p.exe) ✓; the OpenOffice dependency is unsupported — the disc says only «molti software gratuiti o a pagamento»
C34 hit Italian ✓, 4 pages ✓, /CreationDate D:20041021150643+02'00' ✓
C35 half the game is absent from the tree ✓ — but a screenshot of it ships and giochi.htm describes it as completo, which is more presence than the clause allowed
C36 hit 716 of 716 at 2005-12-22 11:16:10 are in Micodb/, against a called 600
C37 hit all 2001 records are in one product directory, completi/FineReader5. (There are 61, not the briefing's 33; 33 is the count at one timestamp)

The shell

# verdict what happened
C38 hit 50 bytes, open=IERunner.exe, icon=CD-ROM\mico.ico, and Mico.ico is a real 766-byte icon
C39 miss _ierr.dll is not resource-only: a 44,544-byte CODE section is 60 % of it. And there are zero res:// URLs on the whole disc — the clause repeated the briefing instead of checking it
C40 hit LegalCopyright: (c) 2001-2002 Pavel Kilovatiy, Comments: TWebBrowser wrapper, US-English translation, a Borland linker. The magazine did not write it. The best clause of the session, and it was written against the briefing
C41 half no server, no CGI ✓ — but the form has no action and no method at all, the eleven fields are packed by hand in JavaScript rather than by the browser, and there is validation making email mandatory
C42 hit both personal addresses, one as To: and one as CC:; abbonamenti@ is not involved
C43 miss 87 <input> tags, 40 distinct names, 65 radio buttons in 18 groups, 18 text boxes, 2 hidden fields. The clause imagined a form with eleven fields; the personal block is eleven of eighteen text boxes inside a survey
C44 hit no submitted response anywhere on the disc

The suppliers

# verdict what happened
C45 miss 93 binaries: 85 PE32, 7 NE16 and one PE32+ x64 (truecrypt-x64.sys). The clause said all PE32 and none 64-bit, and missed a 16-bit installer and a 64-bit driver on the same disc
C46 hit 35 distinct CompanyName strings, inside the called 20–45
C47 miss 50.00 % of COFF timestamps in 2004–2005, against a called 70 %. Sixteen files carry a 1992 constant and thirty-three are from 1998–2003
C48 hit the 1992 date is not a date: 16 files carry it and all 16 and only 16 report linker 2.25, which is Borland's constant
C49 hit both Macrovision hits are InstallShield stubs with CompanyName: Macrovision Corporation and the same COFF timestamp; real protection zero
C50 hit 106 of 106 .hmc inflate and the declared length matches the inflated length every time
C51 miss the inflated payload is binary in 106 of 106 — uncompressed 24-bit bitmap rows in a nested container — against a called 90 % text
C52 hit all 36 .sld in one product; the stream vocabulary (Design1.GalleryItems, Gallery.ExportItems, Preview1_56x56) identifies it
C53 half they parse ✓ and they are one product family — but they are split across two directories in two categories, WebSite 2 and WebSite 5, which the clause explicitly bet against
C54 hit .clx and .car are in the same directory. More than called: the .clx header spells its own language and a sibling DLL names the engine

The media

# verdict what happened
C55 hit 0 files, 0 bytes, 0.000000 %, against a called ≤0.5 %
C56 hit the column is filled with a measured 0 % and the chapter says in the same breath that the method cannot see inside a compressor
C57 hit zero loose .wav, against a called ≤5. Q3 closed as not applicable

The paths

# verdict what happened
C58 miss 646 DOS-shaped paths, below the called 800–4,000. The clause did not expect compressed installers to leak nothing: 568 of the 646 come from the one product that shipped expanded
C59 hit 52 distinct build roots, against a called 15
C60 half 66 of 66 Macintosh-shaped hits are not paths ✓ — but the diagnosis is wrong. Most are in plain string tables in uncompressed DLLs and font name tables: timestamps, error prefixes, Cc: labels, an ATIP format string. The heuristic is not confused by entropy, it is matching what it was written to match on a disc where that shape means a clock

Against the collection

# verdict what happened
C61 half zero real crossings over 20 hash lists ✓, and ≤2 ✓ — but the two hits are da39a3ee…, the sha1 of the empty file, and they do cross with both CLIC discs. The clause said zero with those two and said any crossing would be a Microsoft or Apple redistributable. Neither is true of an empty file

The shape of the session

# verdict what happened
C62 hit 20 documents, 00–19
C63 hit abbonamenti@ quoted, the two individuals counted and never written, in docs/ and notes/ alike, checked before the first commit
C64 half called 41 hits, 12 halves, 9 misses, 2 unresolved, 73.4 %. The hit count is exactly right; the distribution is not, and the headline is 2.3 points high

What the misses have in common

Thirteen misses, and nine of them share one shape: the clause believed something the briefing said instead of checking it.

  • C24 accepted that cdxa.py applied here. It reads a different container.
  • C39 accepted that the pages fetch images through res://_ierr.dll/gif/…. There are zero such URLs on the disc.
  • C11 accepted that this disc was the missing control for CLIC 11's Q1. It is neither of the two things being told apart.
  • C31 accepted that Spybot arrived after the sleeve was printed. Its directory predates the bulk copy.
  • C32 and C33 accepted that Gtk and Java must have a dependent on the disc. Neither does.
  • C28 accepted the briefing's reading of the sleeve and then guessed the arithmetic backwards.
  • C43 accepted the briefing's eleven fields as the size of the form. It is eighty-seven inputs.
  • C51 accepted that a compressed blob in a help-shaped folder would be text.

The one clause written explicitly against the briefing — C40, that the magazine did not write IERunner.exe — is a hit, and it is the finding this repository's second chapter is built on.

The rule is not distrust the briefing. The briefing measured this object carefully and most of what it said held. The rule is the one this branch already had and did not apply evenly: §A is worth no points, and a clause that merely extends §A inherits §A's errors without inheriting its verification. Nine of thirteen misses are that.

And what the halves have in common

Five of the nine halves (C07, C41, C53, C60, C61) are the same failure in a gentler form: the number was right and the mechanism was wrong. The tail's EDC does match, for the wrong reason. The Macintosh hits are all false, for the wrong reason. There are no real crossings, but the two that appear are not what was predicted. A clause that names a count and an explanation can only be half right, and this session wrote nine of them.

Yesterday's session called 41 hits and got 42. This one called 41 and got 41, and was 2.3 points high on the score for the same reason as yesterday — the distribution. Two sessions running, the hit count has been callable and the shape of the misses has not. That is worth carrying forward: the next session should predict its hit count with confidence and its halves with none.