{
  "paper": {
    "id": "sliwerski-msr-2005",
    "title": "When Do Changes Induce Fixes? (On Fridays.)",
    "pdf": "sliwerski-msr-2005.pdf",
    "doi": "10.1145/1083142.1083147",
    "pages": 5
  },
  "main_results": [
    {
      "id": "R1",
      "quote": "In a first investigation of the MOZILLA and ECLIPSE history, it turns out that fix-inducing changes show distinct patterns with respect to their size and the day of week they were applied.",
      "states": "it turns out that fix-inducing changes show distinct patterns with respect to the day of week they were applied",
      "page": 1,
      "section": "Abstract",
      "source": "abstract",
      "note": "The sentence states two main results: the pattern by the day of week, which this entry is recorded for, and the pattern by size, which R2, in the conclusion (page 5), states more specifically. No summary sentence names the day. The title's \"(On Fridays.)\" (page 1) and the heading of Section 5.2, \"Don’t Program on Fridays\" (page 4), do."
    },
    {
      "id": "R2",
      "quote": "In this paper, we examined the properties of fix-inducing changes in the ECLIPSE and MOZILLA projects and found, among others, that the larger a change, the more likely it is to induce a fix; checking for other correlated properties is straight-forward.",
      "page": 5,
      "section": "7 Conclusion",
      "source": "conclusion"
    },
    {
      "id": "R3",
      "quote": "We also found that in the ECLIPSE project, fixes are three times as likely to induce a later change than ordinary enhancements.",
      "page": 5,
      "section": "7 Conclusion",
      "source": "conclusion"
    }
  ],
  "claims": [
    {
      "id": "C1",
      "quote": "Additionally, Table 3 shows that fix-inducing transactions are roughly three times larger than non fix-inducing transactions.",
      "states": "Additionally, Table 3 shows that fix-inducing transactions are roughly three times larger than non fix-inducing transactions.",
      "page": 4,
      "section": "5.1 Fix-Inducing Transactions are Large",
      "serves": [
        "R2"
      ],
      "split_from": null,
      "selection_reason": "R2 says that in the ECLIPSE and MOZILLA projects the larger a change, the more likely it is to induce a fix, and rests on the size results for the two projects together; this is the one for ECLIPSE. If fix-inducing transactions in ECLIPSE were not larger than the others, R2 would hold for MOZILLA only and would need substantial revision.",
      "note": "Table 3 (page 4) prints the two averages and not their ratio: in its all row, 7.49±44.37 under fix-inducing and 2.61±13.66 under ¬fix-inducing. The arithmetic is left to a later step."
    },
    {
      "id": "C2",
      "quote": "Table 4 presents the same breakdown for MOZILLA which shows a similar trend.",
      "states": "Table 4 presents the same breakdown for MOZILLA which shows a similar trend.",
      "page": 4,
      "section": "5.1 Fix-Inducing Transactions are Large",
      "serves": [
        "R2"
      ],
      "split_from": null,
      "selection_reason": "R2 rests on the size results for ECLIPSE and MOZILLA together, and this is the one for MOZILLA. If fix-inducing transactions in MOZILLA were not larger than the others, R2 would hold for ECLIPSE only and would need substantial revision.",
      "note": "The sentence gives no number, and no sentence gives one for MOZILLA. The numbers are in Table 4 (page 4), whose all row prints 5.19±34.12 under fix-inducing and 1.97±10.13 under ¬fix-inducing. \"The same breakdown\" is the one Table 3 gives for ECLIPSE, and \"a similar trend\" refers to C1, that fix-inducing transactions are roughly three times larger than non fix-inducing transactions."
    },
    {
      "id": "C3",
      "quote": "Table 5 presents the results for ECLIPSE. The likelihood P (bug) that a change will induce a fix is highest on Friday.",
      "states": "Table 5 presents the results for ECLIPSE. The likelihood P (bug) that a change will induce a fix is highest on Friday.",
      "page": 4,
      "section": "5.2 Don’t Program on Fridays",
      "serves": [
        "R1"
      ],
      "split_from": null,
      "selection_reason": "R1 says that fix-inducing changes show distinct patterns with respect to the day of week they were applied, in MOZILLA and ECLIPSE, and the title gives that pattern as \"(On Fridays.)\". R1 rests on the results for the two projects together, and this is the one for ECLIPSE. If the likelihood that a change induces a fix were not highest on Friday in ECLIPSE, the pattern would not hold for ECLIPSE and R1 would need substantial revision.",
      "note": "The first sentence is quoted because the second does not name the project. The sentence gives no number. Table 5 (page 4) prints the likelihood for each day in its P (bug) row."
    },
    {
      "id": "C4",
      "quote": "The same holds [...] for MOZILLA (see Table 6).",
      "states": "The same holds for MOZILLA (see Table 6).",
      "page": 4,
      "section": "5.2 Don’t Program on Fridays",
      "serves": [
        "R1"
      ],
      "split_from": null,
      "selection_reason": "R1 rests on the results for MOZILLA and ECLIPSE together, and this is the one for MOZILLA. If the likelihood that a change induces a fix were not highest on Friday in MOZILLA, the pattern by day of week would not hold for MOZILLA and R1 would need substantial revision.",
      "note": "\"The same\" refers to C3, that the likelihood P (bug) that a change will induce a fix is highest on Friday. The sentence runs from the foot of the left column to the right column, which opens with Tables 5 and 6 and their captions; that is what the [...] marks. The sentence gives no number. Table 6 (page 4) prints the likelihood for each day in its P (bug) row."
    },
    {
      "id": "C5",
      "quote": "Table 5 shows that for ECLIPSE, the average likelihood of introducing a fix-inducing change is almost three times higher for fixes, indicated by P (bug | fix), than for regular changes, indicated by P (bug | ¬fix).",
      "states": "Table 5 shows that for ECLIPSE, the average likelihood of introducing a fix-inducing change is almost three times higher for fixes, indicated by P (bug | fix), than for regular changes, indicated by P (bug | ¬fix).",
      "page": 4,
      "section": "5.2 Don’t Program on Fridays",
      "serves": [
        "R3"
      ],
      "split_from": null,
      "selection_reason": "R3 says that in ECLIPSE fixes are three times as likely to induce a later change as ordinary enhancements, and this sentence gives that comparison from Table 5. If fixes in ECLIPSE were not more likely than regular changes to be fix-inducing, R3 would fail.",
      "note": "Table 5 (page 4) prints the two likelihoods and not their ratio: in its avg column, P (bug | fix) 23.7 and P (bug | ¬fix) 8.1. The arithmetic is left to a later step. R3, in the conclusion (page 5), gives the ratio as \"three times as likely\", and this sentence as \"almost three times higher\"."
    }
  ],
  "excluded": [
    {
      "id": "E1",
      "quote": "Section 5 shows the results of our investigation of the MOZILLA and ECLIPSE: It turns out that fix-inducing changes show distinct patterns with respect to their size and the day of week they were applied.",
      "page": 1,
      "section": "1 Introduction",
      "duplicate_of": [
        "R1",
        "R2"
      ],
      "reason": "Repeats the abstract's sentence, R1, and with it the pattern by day of week that R1 is recorded for and the pattern by size that R2 states."
    },
    {
      "id": "E2",
      "quote": "Notice that only 27% of all bugs in the MOZILLA project are FIXED (47% for ECLIPSE).",
      "page": 2,
      "section": "3.2 Semantic Analysis, footnote 1",
      "reason": "Its numbers describe the study: they give the share of the bugs in each project's bug database that are resolved as FIXED, which describes the data the linking works on."
    },
    {
      "id": "E3",
      "quote": "We identified 25,317 links for ECLIPSE, connecting 47% of fixed bugs with 29% of transactions and 53,574 links for MOZILLA, connecting 55.30% of fixed bugs with 43.91% of transactions.",
      "page": 3,
      "section": "3.3 Results",
      "reason": "Its numbers describe the study: they count the links the linking found in each project and the share of fixed bugs and of transactions those links connect, the material that the results in Section 5 rest on.",
      "note": "No main result states how far the linking reaches: the abstract presents the linking only as the way to locate fix-inducing changes. Tables 1 and 2 (page 4) print the link totals, 25,317 for ECLIPSE and 53,574 for MOZILLA, and not the shares of fixed bugs or of transactions, which no table prints."
    },
    {
      "id": "E4",
      "quote": "Notice that we disregard less than 10% of links for both projects.",
      "page": 3,
      "section": "3.3 Results",
      "reason": "Its number describes the study: it says how many of the links the confidence condition leaves out of the data the results rest on.",
      "note": "Tables 1 and 2 (page 4) print the links for each pair of confidence levels, and no cell gives the share that the condition disregards."
    },
    {
      "id": "E5",
      "quote": "We considered all changes and bugs until January 20, 2005; our database contains 78,954 transactions for ECLIPSE and 109,658 transactions for MOZILLA. They account for 278,010 and 392,972 individual revisions for both projects, respectively.",
      "page": 3,
      "section": "5 First Results",
      "reason": "Its numbers describe the study: they give the transactions and revisions in the database that the results in Section 5 rest on.",
      "note": "\"They\" refers to the transactions. The second sentence pairs its numbers with \"respectively\": 278,010 revisions for ECLIPSE and 392,972 for MOZILLA."
    },
    {
      "id": "E6",
      "quote": "For instance, the top-left cell means that [...] the average size of transactions which are fixes and induce later on a fix is 3.82 (with a standard deviation “±” of 26.32).",
      "page": "3-4",
      "section": "5.1 Fix-Inducing Transactions are Large",
      "reason": "R2 would still stand, because this sentence explains how to read one cell of Table 3, the size of the transactions that are both fixes and fix-inducing, and R2 rests on the comparison of fix-inducing with other transactions that C1 and C2 give.",
      "note": "Table 3 (page 4) prints 3.82±26.32 in its fix row under fix-inducing."
    },
    {
      "id": "E7",
      "quote": "Friday is the day where most ECLIPSE developers do fixes, for MOZILLA this is Sunday.",
      "page": 4,
      "section": "5.2 Don’t Program on Fridays",
      "reason": "R1 would still stand, because this sentence gives the day on which the most fixes are made, and R1 is about the day on which fix-inducing changes are applied, which C3 and C4 give.",
      "note": "\"This\" refers to the day where most developers do fixes. Tables 5 and 6 (page 4) print the fixes for each day in their P (fix) rows, under the heading \"% of revisions\", which is a share of the revisions of each day rather than a count of developers."
    },
    {
      "id": "E8",
      "quote": "This does not hold for MOZILLA (see Table 6).",
      "page": 4,
      "section": "5.2 Don’t Program on Fridays",
      "reason": "R3 would still stand, because it states the comparison for ECLIPSE only, and this sentence gives it for MOZILLA.",
      "note": "\"This\" refers to the sentence before it, C5: that for ECLIPSE the average likelihood of introducing a fix-inducing change is almost three times higher for fixes than for regular changes. Table 6 (page 4) prints, in its avg column, P (bug | fix) 45.2 and P (bug | ¬fix) 38.1."
    },
    {
      "id": "E9",
      "quote": "The risk that a fix will be later undone is highest for ECLIPSE on Saturdays, and for MOZILLA on Fridays.",
      "page": 4,
      "section": "5.2 Don’t Program on Fridays",
      "reason": "R1 would still stand, because this sentence gives the pattern by day of week for the fixes only, a subset of the changes, and R1's pattern rests on the likelihood for all changes that C3 and C4 give.",
      "note": "Tables 5 and 6 (page 4) print the risk for each day in their P (bug | fix) rows."
    },
    {
      "id": "E10",
      "quote": "Almost every second change in MOZILLA is a fix and two out of five changes are fix-inducing.",
      "page": 4,
      "section": "5.2 Don’t Program on Fridays",
      "reason": "R1 would still stand, because this sentence gives the share of MOZILLA's changes that are fixes and that are fix-inducing over the whole week, and R1 is about how the likelihood of inducing a fix varies from one day of the week to another.",
      "note": "No summary sentence states how many of MOZILLA's changes are fixes or fix-inducing, and the sentence after it builds on it: \"In the future we will investigate MOZILLA to find out what makes MOZILLA risky.\" Table 6 (page 4) prints, in its avg column, P (fix) 48.5 and P (bug) 41.5."
    }
  ]
}