add two missing ones

theses/html/2017-11-17-Alexis.md

@@ -0,0 +1,261 @@
+<p>TL;DR (aka abstract): Conditional votes work, but with two caveats.
+First, they operate prospectively and, when cast, their value is
+determined by looking into the future to the end of the voting period.
+Hence, they are indeterminate. Second, the default of PRESENT does not
+work at all, meaning that the validity of a conditional vote is also
+indeterminate. Also, CFJ 3569 is FALSE because Trust Token endorsement
+must be direct.</p>
+<p>This thesis and judgment is written as a reconsideration of CFJ 3569.
+The statement of the CFJ is "If grok had not deregistered, e would have
+issued trust tokens to both Aris and G. by eir vote on Proposal 7899."</p>
+<p>The context of the CFJ is as follows: grok voted :Endorse Aris" on
+proposal 7899, but Aris's vote was "Endorse the Arbitor". G. was the
+Arbitor at the end of the voting period. In effect, the CFJ is asking
+whether grok's endorsement was effective of endorsing G. as well for the
+purpose of Rule 2452.</p>
+<p>I originally judged CFJ 3569 to be FALSE, finding that Rule 683's
+condition requiring that a "valid vote" be clearly indicate meant that
+conditional votes failed altogether; since a conditional vote does not
+clearly indicate a single valid vote. A number of arguments were raised
+in response, leading to a Motion to Reconsider. I will summarize the
+critical arguments below:</p>
+<blockquote>
+<p>From G.:</p>
+</blockquote>
+<ul>
+<li>The judgment cited precedent that the correctness of a ballot is
+evaluated at the time of its submission, but did not specify which
+precedent, most precedent being around informally conditional actions
+(e.g. "if I own an X, I do Y with it")</li>
+<li>Rule 2127 is explicitly intended to define the effect of conditionals
+and get around the usual definition, by defining what "clearly
+specified" means in the context of a vote. This could operate either by
+deferral of the determination of clearly specified, or by operating
+retroactively.</li>
+<li>Even in absence of the above effects, Rule 2127's default of PRESENT
+would apply to a ballot cast conditionally, so that such a ballot is at
+least evaluated in the instant to PRESENT.</li>
+</ul>
+<blockquote>
+<p>From myself:
+- If Rule 2127 had retroactive effect, then it would create the
+possibility of retroactively changing the outcome of another proposal,
+by making a vote conditional on a vote on a different proposal; if the
+latter was a conditional then retroactively changing it could also
+retroactively change the first vote. From this, paradox could arise.</p>
+<p>From Aris:
+- Present conditionals are well-understood to succeed, under the general
+built-up framework of interpreting conditional actions, as such a
+conditional always identifies a vote.
+- A ballot is a textual entity, containing a conditional that should be
+evaluated when the ballot is evaluated, i.e. when the decision is
+resolved. The ballot is merely a notice of something in the future.</p>
+</blockquote>
+<p>To begin, I will first point to Rule 217, which states "When
+interpreting and applying the rules, the text of the rules takes
+precedence. Where the text is silent, inconsistent, or unclear, it is to
+be augmented by game custom, common sense, past judgments, and
+consideration of the best interests of the game." In regards to
+conditionals votes, it is clear that all of the augmenting factors point
+to conditional votes functioning in some fashion: game custom certainly
+establishes that they do, as do the best interests of the game. Past
+judgments on conditional voting are in line with them functioning at
+all, and common sense (into which one can likely subsume legal
+principles of statutory interpretation) imply that Rule 2127 (the rule
+defining conditional votes) would not exist if it did not have some purpose.</p>
+<p>So the question then is, what does the text of the rules state? CFJ 3465
+is the strongest precedent I am aware of regarding ballot evaluation. In
+it, I successfully argued to the H. Judge nichdel that the correctness
+(note: this is distinct from validity) of a ballot was evaluated at the
+time of its submission.</p>
+<p>I will reiterate the critical arguments here. Rule 683 states "An entity
+submits a ballot on an Agoran decision by publishing a notice satisfying
+the following conditions:" followed by a list of conditions. The term
+"notice" is, in this context, undefined by the rules. Per usual English
+interpretation, in this context, "notice" simply means a document.</p>
+<p>Rule 683 is clear that the mechanism to submit a ballot is to publish
+the notice. Publishing is defined by Rule 478 as sending a public
+message. It follows that if, at the time of publication, the notice does
+not satisfy the conditions set out in Rule 683, no ballot is submitted.
+The fact that the notice may later satisfy the conditions is immaterial,
+as the notice is not being published at that point.</p>
+<p>So can conditional votes generally satisfy the conditions? The condition
+at issue is "4. The ballot clearly identifies a valid vote, as
+determined by the voting method." The definition of valid vote was
+discussed at length in my previous reasons, and I will not discuss them
+here. Absent confounding factors, then, it would be clear that the
+published notice must indeed identify a valid vote. I do agree with
+Aris's arguments that the general rules regarding conditional actions
+apply here. There is nothing disallowing a notice which is conditional,
+at the time of its publication; assuming the conditional is effective
+per the precedents on conditional actions, such a notice does identify a
+valid vote once the conditional is resolved.</p>
+<p>Rule 2127 is, however, a confounding factor. As G. argued, it is clearly
+intended to get around the usual way that conditionals can act only in
+the present, and instead creates a form of future-conditional vote. Note
+that I must at this point reject Aris's argument that a conditional vote
+is evaluated at the time of the decision's resolution. Rule 2127 is
+explicit that the conditional is evaluated at the end of the voting
+period, and there is nothing in the rules or game custom to contend
+otherwise.</p>
+<p>As G. noted in eir arguments, Rule 2127 states "If a vote on an Agoran
+decision is submitted conditionally (e.g. "FOR if <X> is true, otherwise
+AGAINST"), then the selected option is evaluated based on the value of
+the condition(s) at the end of the voting period, and, rules to the
+contrary notwithstanding, is clearly specified if and only if the value
+of the condition(s) is/are determinate at the end of the voting period.
+If the option cannot be clearly identified, a vote of PRESENT is cast."
+Of note is the fact that it explicitly attempts to take precedence over
+other rules (including 683) to define that "the selected option... is
+clearly specified" under the circumstances it provides when a
+conditional vote is made.</p>
+<p>Does this apply to condition 4? After reexamining the circumstances, it
+appears to me that the answer is yes. While the language used is
+slightly different (option vs vote, clearly indicated vs clearly
+specified), and while some of those terms are terms of art within the
+rules, none is explicitly defined, so there is some leeway in
+interpretation here. Note also that 2127 itself uses "clearly
+identified" in the following sentence, indicating some looseness in
+terminology. In my original judgment, I had not considered this line of
+text in the way that G.'s arguments encouraged, and this represents the
+critical change in my reasoning.</p>
+<p>Unfortunately, this still presents some difficulty, because the
+definition is prospective. The direct consequence of interpreting
+condition 4 using the definition specified in Rule 2127 is that the
+correctness of the ballot at the time of its submission is determined by
+looking into the future. The only potential save could be, as G. argued,
+the following sentence: "If the option cannot be clearly identified, a
+vote of PRESENT is cast." This does not directly bear on the outcome of
+CFJ 3569, but in the interests of completeness and avoiding simply
+postponing the controversy, I will analyze how to evaluate conditional
+votes below.</p>
+<p>Having determined that a conditional vote is indeed valid provided its
+condition is determinate, the only remaining questions required to
+resolve this case is whether grok's vote had a determinate condition and
+whether Aris and G., who both cast valid votes on the decision, were
+"endorsed by [grok]'s valid vote on the Decision" (Rule 2452). It is
+clear that grok's vote had a determinate condition, making it valid, and
+it clearly endorsed Aris (regardless of how such it would actually be
+evaluated). So grok would have issued Aris a trust token. However, did
+grok's vote endorse G.?</p>
+<p>Per Rule 2127, "Casting a vote endorsing another voter is equivalent to
+conditionally casting a vote whose value is the same as the most common
+value (if any) among that voter's valid votes on that decision." While
+this definition technically applies only to the point when the vote
+would cast, it would be unreasonable to interpret the meaning of "a vote
+endorsing that voter" as meaning anything other than "a conditional vote
+whose value is the same as the most common value (if any) among that
+voter's valid votes on that decision." There are two reasonable ways to
+evaluate this definition on grok's vote then: either it is an inherent
+property of the vote, or it is evaluated, like all conditionals, at the
+end of the voting period.</p>
+<p>A quirk in the wording of Rule 2127 gives a possible hint. It defines
+endorsing votes as an equivalence with a specific class of votes, rather
+than merely defining what it means to endorse a player. Consequently, if
+"equivalent" is interpreted as it would be in a mathematical context,
+any conditional vote whose value is as described in Rule 2127 is an
+endorsing vote, regardless of whether or not the vote explicitly uses
+the word "endorse". The terms are interchangeable.</p>
+<p>Substituting this into Rule 2452, that would mean that grok issues a
+Trust Token to G. if "grok had a valid conditional vote whose value is
+the same as the most common value (if any) among G's valid votes on the
+decision". If we evaluate this at the end of the voting period, then a
+thoroughly absurd result ensues: not only would G. be issued a trust
+token, but also every other player who cast the same vote. As such, we
+must interpret the definition as applying to an intrinsic property of
+the conditional. Thus, we cannot consider Aris's vote when we decide
+whether grok's vote endorsed, G. and so it did not. grok would not have
+issued G. a Trust Token.</p>
+<p>It's not necessarily clear, however, that "equivalent" in the Agoran
+ruleset should be interpreted this way (an old rule that mathematical
+definitions take precedence having been repealed). If we interpret this
+wording as being merely an odd way of defining the term endorsement,
+then the outcome is the same, however. The definition of endorsement
+specifically applies to valid votes; thus, in order to resolve grok's
+conditional vote, we must first resolve Aris's. A different definition
+of endorsement, where grok's vote copied the "Endorse the Arbitor" from
+Aris's vote, might lead to a conclusion where grok did endorse G., but
+we clearly must resolve Aris's vote first, so grok's vote is never
+directly dependent on G.'s.</p>
+<p>Thus, either way, CFJ 3569 is FALSE.</p>
+<p>With that resolved, how then should a conditional vote be evaluated?
+This is unfortunately a thorny issue. As I wrote above, whether or not
+the vote is valid is effectively determined prospectively, by looking
+into the future. Before digging deeper, however, I will draw attention
+to one key point: the question of a ballot's correctness (i.e. whether
+it meets the requirements of Rule 683) is independent from the question
+of how it is evaluated in an Agoran decision. Thus, I see two remaining
+questions: first, is a conditional vote with an indeterminate condition
+correctly submitted, or does it fail to clearly identify a valid vote?
+and second, how does one evaluate a conditional vote?</p>
+<p>To address the first question, I quote again Rule 2127: "If a vote on an
+Agoran decision is submitted conditionally (e.g. "FOR if <X> is true,
+otherwise AGAINST"), then the selected option is evaluated based on the
+value of the condition(s) at the end of the voting period, and, rules to
+the contrary notwithstanding, is clearly specified if and only if the
+value of the condition(s) is/are determinate at the end of the voting
+period. If the option cannot be clearly identified, a vote of PRESENT is
+cast." The first sentence provides very clear conditions about when the
+option is clearly specified. What does the second sentence do in this
+context? Does it indicate that, when the condition is indeterminate, the
+selected option is clearly specified as PRESENT?</p>
+<p>I think that this cannot be the case, because to do so would completely
+obviate the definition of "clearly specified" in the first place. Why
+would the rule state that an option is not clearly specified if the
+conditional is indeterminate, only to change its mind in the next
+sentence? This does not actually matter, however, because the first
+sentence contains a precedence claim and the second sentence does not.
+Per Rule 2240, this means that the first sentence takes precedence and,
+consequently, the second sentence does not modify it. If the condition
+is indeterminate at the end of the voting period, the vote is not
+clearly specified. Per the reasoning above, this means that the
+conditional ballot is not submitted.</p>
+<p>What, then, is the meaning of the second sentence? The most reasonable
+interpretation, in context, is that it is trying to say that if the
+condition is indeterminate, then at the end of the voting period, we
+resolve it to PRESENT rather than attempting to evaluate it. If this is
+the correct interpretation, it is meaningless, because by the above we
+can never get to that point. There is an alternative interpretation: the
+second sentence is not providing for a way to evaluate the original
+conditional ballot but instead provided for an altogether new,
+unconditional ballot of PRESENT to be cast.</p>
+<p>Unfortunately, this interpretation also fails to have any meaningful
+effect. The vote counting procedure is defined in Rule 955 in terms of
+valid ballots, and Rule 683 provides a clear definition: "A valid ballot
+is a ballot, correctly submitted, that has not been withdrawn. During
+the voting period of an Agoran decision, a player CAN by announcement
+withdraw (syn. retract) a ballot that e submitted on that decision." As
+Rule 683 has a lower ID number than Rule 2127, by Rule 1030, Rule 683
+takes precedence. There is thus no possibility of Rule 2127's "vote of
+PRESENT" actually being a valid ballot.</p>
+<p>So in conclusion, the answer is that a conditional vote is valid if and
+only if the condition ends up being determinate. If it is not, the
+entire vote fails; PRESENT is not an alternative.</p>
+<p>Finally, the question comes of evaluating a conditional vote. Happily, I
+see no reason why, for a valid vote, Rule 2127 should be anything other
+than what it straightforwardly says. Provided that the conditional is
+valid (i.e. determinate at the end of the voting period), it is
+evaluated as one would expect, at the end of the voting period. Prior to
+then, however, just like with the validity of the vote in the first
+place, the actual value of the vote is indeterminate.</p>
+<p>What are the practical implications of this? First, it is important to
+distinguish between prospective and retroactive effects here. The
+validity of a conditional vote is not retroactively changed at the end
+of the voting period, when we finally are able to decide if it is
+determinate or not. Rather, the validity <em>in the first place</em> is
+dependent on the future. In some cases, this can create
+effectively-retroactive effects. For instance, a subsequent, otherwise
+valid ballot cast by the same voter, without withdrawing the first one,
+would enter the same indeterminate state of validity, since it is valid
+if and only if the first ballot is invalid.</p>
+<p>This may present situations where the game may be difficult to play, but
+the concept of indeterminacy provides a shield in some cases. Rule 1023
+means that the validity of the vote is indeterminate before the voting
+period ends, since it cannot be determined from information reasonably
+available. Rule 2127 itself provides that if a conditional vote is
+indeterminate, it doesn't work. So, taking from the example in my
+arguments, if a vote on one proposal is conditional on a vote on another
+proposal ending later, then it necessarily fails because the condition
+will be indeterminate. Likewise, Rule 2162 protects against
+indeterminately-valued switches.</p>
+<p>In conclusion, rules are hard.</p>