DOI: https://doi.org/10.5281/zenodo.21947310
Canonical: https://thonly.org/research/dedicatory-generation · Licence: CC0 1.0
Draft notes for the editor: this paper publishes doctrine and claims, and deliberately withholds the pipeline. The product it governs is unbuilt and appears on no schedule, and for an unbuilt artifact publication protects against being blocked, not against being beaten — so the fences are stated in full and the blueprint is not. That constraint narrows what is described; it does not narrow whether to publish, because this paper's purpose is precedent, and a precedent has to exist as a document. Someone will do this work to some interrupted canon, probably soon. The question this paper tries to answer early is what the rules should be when they do.
Between 1975 and 1979 the Khmer Rouge destroyed a generation of Cambodian artists — the musicians, filmmakers, writers and performers of what is remembered as Cambodia's Golden Age. The recordings that survive did so by accident: cassettes carried across a border, prints in a private collection, a reel that somebody buried. What did not survive was almost everything else, and the almost everything else is the subject of this paper.
Generative models can now produce new work in a lost artist's voice. The public discourse about whether they should has developed around a case that does not apply here: a living or recently-deceased artist, with a surviving estate, in an intact legal system, where consent is at least a question that can be asked. Every proposed framework — licensing, opt-out, estate approval, revenue sharing — presumes a party on the other side of the table.
The genocide case differs on four axes at once. Consent is not obtainable, and not merely unobtained. There are no surviving estates or institutions to hold rights, because the institutions were destroyed alongside the people. The records of intent were themselves destroyed — there is no diary, no contract, no interview in which an artist said what they would have wanted. And the destruction was aimed at the culture: the erasure was the objective, not a side effect, which means that a reconstruction performed carelessly does not merely offend an absent party — it completes an unfinished act.
This paper specifies a discipline for that case. Its central move is a single change of preposition. Work generated in a lost artist's voice is published for them and never by them: an offering, not an authorship claim — a form with deep precedent in the tradition the culture actually belongs to, where merit is dedicated to the departed and the dedication is the point.
Nine further claims follow, and one of them is the one to protect hardest: the model's training objective must exclude every economic signal, and admit only a refusal signal. A system optimized on what performs is a system that will learn to make the dead popular. A system that can only be told no can be steered away from harm without ever being pointed at a market.
The lineage here is Cambodian and not imported. Rithy Panh filmed clay figurines rather than reenactments, because a reenactment of the camps would have been a lie with better production values. Anastylosis at Angkor reassembles original stones and makes every new element visibly distinguishable, forever. Both are answers to this paper's hardest problem, arrived at long before it: fidelity and forgery are the same axis, so success is the danger.
Offered under CC0 1.0 Universal as defensive prior art.
Keywords: generative models, cultural heritage, genocide, attribution, dedicatory generation, anastylosis, provenance marking, training objectives, refusal signal, voice models, Cambodia, Khmer Golden Age, defensive publication.
Dedicated to the public domain under CC0 1.0 Universal. Neither the authors nor HeartBank® will seek patent, trademark, or any exclusive right over the discipline described here. The point of publishing it is that anyone may adopt it, including parties we would not choose.
Terms coined and freed with this paper: dedicatory generation, the by/for boundary, fidelity-scaled marking, the refusal-only objective, the four verbs of an interrupted canon, and restoration-as-pipeline.
Terms inherited from this corpus and cited rather than re-claimed: pattidāna as a routing operator, the determined-versus-judged seat distinction, and the bhāṇaka guard (a reciter is not the author of what is recited).
What is emphatically not ours. The heritage itself belongs to Cambodia and to the descendants of the people who made it — not to this institution, not to any model, and not to the authors. Sinn Sisamouth, Ros Serey Sothea, Pen Ran and their contemporaries are named here because they should be named; the historical record holds that they did not survive the Khmer Rouge period, and the specific circumstances of their deaths remain undocumented, which is itself part of what this paper is about. Rithy Panh's body of work, in particular The Missing Picture (2013), is his and is the direct artistic precedent for the central discipline described here. Anastylosis is a conservation method developed in Greek and Southeast Asian archaeology and codified in the Venice Charter (1964); its application at Angkor, notably the Baphuon reconstruction, is the work of the École française d'Extrême-Orient and Cambodian conservators over decades. The broader debate on generative models and deceased artists is a live public conversation with many participants; we engage it and claim no part of it.
There is by now a recognizable shape to the argument about generative models and dead artists. An estate objects, or licenses. A platform adopts an opt-out. A jurisdiction extends a personality right past death for a term of years. A settlement establishes a royalty. The disagreements are real and the participants are serious, and the entire apparatus rests on an assumption so basic that it is rarely stated: that there is somebody to ask, or somebody to sue.
Four things are simultaneously true of the Cambodian case, and each one removes a different assumption.
Consent is not obtainable, not merely unobtained. This is a stronger condition than "the artist is dead." A recently deceased artist has an estate that inherits the right to decide; the decision continues to be made, by successors, under law. Here the chain does not exist to be followed.
There are no surviving estates or institutions. The destruction was not confined to individuals. Publishers, studios, archives, the recording industry, the professional bodies, the universities that trained the artists, and the civil registries that recorded who they were were destroyed together. There is no entity that holds these rights because the category of entity that would have held them was eliminated.
The records of intent were destroyed. No diary survives that says what an artist would have wanted done with their voice. No contract, no interview, no letter. Ordinary posthumous-rights reasoning leans heavily on evidence of what the person would have chosen; here the evidentiary base is itself part of what was burned.
And the destruction was aimed at the culture. This is the axis that changes the ethics rather than the logistics. The killing of the artists was not incidental to some other objective — the erasure of the cultural forms was the objective. Which means the ordinary risk of a careless reconstruction (that it misrepresents someone) is joined by a much worse one: a careless reconstruction finishes the job. A forgery in this context does not merely mislead a listener about who made a song. It replaces a destroyed thing with a manufactured thing and calls the culture restored.
⭐ The condition generalizes, and that is why this is published rather than filed. Every interrupted canon has some version of these four axes — the Armenian, the Indigenous canons of the Americas and Australia, the libraries burned in Baghdad and Sarajevo, and any tradition whose transmission was severed rather than merely neglected. The discipline below is written for Cambodia because that is the canon the authors have standing in, and it is stated generally because it is not about Cambodia.
The most serious objection to this paper is not that its rules are wrong. It is that the rules are a way of proceeding, and the correct action is to stop.
The argument runs like this. A canon severed by genocide is not an incomplete work awaiting continuation; it is a closed testimony. Its incompleteness is the evidence. Anything added — however carefully marked, however scrupulously attributed — converts a site of loss into a site of activity, and activity is consoling in a way that loss is not supposed to be. The people best positioned to do this work are, almost by definition, people who find the loss unbearable, which is exactly the disposition least able to judge when to stop. And the discipline in this paper is the kind of thing that makes proceeding feel responsible, which is worse than proceeding recklessly, because recklessness at least announces itself.
⚠️ We do not think this objection is answered by the rules below, and we will not pretend it is. Three things can be said, and none of them is a refutation.
The counterfactual is not silence. The recordings are already circulating; the models already exist in general form; the capability arrives whether or not any careful party uses it (§4). The realistic alternative to a disciplined practice is not an untouched canon but an undisciplined one, conducted by parties with no relationship to the culture and no reason to accept any of these constraints. That is an argument for the discipline; it is not an argument that any particular work should be made.
The refusal is available at every point and is built in. Claim 9's revocability, §7's refusal to fill gaps, and P-B3's stopping condition are all instruments for not proceeding. A discipline whose only outputs are permissions is a rationalization; this one has to be able to say no to itself in public, and §10 is where it commits to that in advance.
And the objection should be held permanently rather than dispatched. ⭐ The correct posture is not that this argument was considered and defeated in 2026, but that it remains live and that any practitioner should be able to state it as well as its opponents can. A discipline that can no longer articulate the case against itself has become a licence.
It would be possible to derive this paper's central discipline from Western conservation ethics. It is not necessary, because Cambodia produced two answers first, and both are better.
Rithy Panh filmed clay. The Missing Picture reconstructs the author's childhood under the Khmer Rouge — and where a conventional documentary would reenact, Panh made small clay figures and filmed those. The choice is not stylistic modesty. A reenactment of the camps, however carefully staged, would produce footage that looks like evidence and is not — and in a history whose central wound is the destruction of the record, manufacturing convincing false record is the specific harm to avoid. The figurines can never be mistaken for what happened. They can carry grief without counterfeiting testimony.
Anastylosis reassembles and marks. At Angkor, and above all in the decades-long reconstruction of the Baphuon, conservators put fallen temples back together using the original stones, in their original positions, with structural additions made in material that is visibly and permanently distinguishable from the ancient fabric. The rule is codified: what is restored must be legible as restoration, forever, to anyone standing in front of it. The temple is made whole enough to enter, and no visitor is ever deceived about which century they are touching.
⭐⭐ Both answers encode the same insight, and it is the one this paper is organized around: fidelity and forgery are the same axis. There is no quality threshold beyond which a reconstruction becomes safe. The better it gets, the more dangerous it is — so SUCCESS IS THE DANGER, and the marking must scale with the fidelity rather than being an artifact of its imperfection.
A model that produces a passable imitation is a curiosity. A model that produces an indistinguishable one is a weapon against the record. The discipline cannot rely on the output being detectably imperfect, because the entire trajectory of the technology is toward it not being.
The temple analogy is load-bearing and it fails at one specific joint, which is worth working through because the failure is the reason §7's marking rule has to be imposed rather than inherited.
In stone, the distinction is in the material. A conservator at the Baphuon does not have to mark which blocks are ancient — the ancient blocks are ancient, physically, and a visitor can see the difference between weathered sandstone and new fill. The honesty of anastylosis is partly a discipline and largely a property of matter. Stone remembers.
Audio has no such property. A digital recording has no original material. A perfect copy is not a copy in any sense that matters — it is the recording, bit for bit, with no wear, no patina, and nothing to weather. There is no equivalent of the old stone, which means there is nothing in the artifact itself that distinguishes a restored 1968 vocal from a generated one, and no amount of craft will produce that distinction because the substrate does not carry it.
⭐⭐ So the discipline has to supply what the material does not. In stone, honesty is discovered; in audio, honesty must be manufactured and attached, and it must be attached in a way that survives copying, re-encoding, platform ingestion, and the ordinary hostility of the internet to metadata. That is a much harder problem than the temple faced, and it is why claim 6 states marking as an obligation that scales with fidelity rather than as a disclosure made once.
⚠️ And it is why claim 10 exists. If the artifact cannot carry its own provenance reliably, then the only remaining control point is upstream — the model itself — which is precisely why the weights are the thing not released. The temple could afford to publish its methods because the stones told the truth on their own. Recordings cannot, so the method is where the restraint has to live.
The central move is one word.
Work generated in a lost artist's voice is published FOR them. Never BY them.
Not "in the style of," which is a marketing formula and points at the work. Not "featuring," which asserts participation. Not "AI Sinn Sisamouth," which is an authorship claim wearing a disclaimer. For.
This is a familiar form and the tradition it comes from is the culture's own. In Theravāda practice, merit from an act is dedicated to the departed — the act is performed by the living, the dedication names the dead, and nothing is claimed on their behalf. The one performing the act does not become the person they dedicate it to, and no one has ever been confused about this. Offering is a complete and well-understood relation between the living and the dead that requires no consent, manufactures no authorship, and misrepresents nobody.
Three consequences follow immediately:
Generalizing claim 1 from an artist to a canon reveals that the four media do not take the same verb, and treating them as one operation is the first mistake available.
| medium | verb | what it means | what the artifact is |
|---|---|---|---|
| Film / play | PRESERVATION | the print exists and is degrading; the work is there | a restored original |
| Music | OFFERING | the artist is gone; new work is dedicated, never attributed | a song for them |
| Story | NEW AUTHORSHIP | a living author writes, in a tradition, under their own name | their own work |
| Scripture | INVENTION | nothing is recovered or continued; a new form is built | an explicitly new thing |
⚠️ This table exists because of a correction, and the correction should travel with it. The founder of the institution behind this paper initially described all four of his projects as "a complete restoration." That description flattens precisely the distinction the discipline protects: it makes an offering sound like a recovery, and it makes an invention sound like a homecoming. Restoration is the right word for exactly one of the four, and using it for the others is the by/for error at the scale of a whole culture rather than a single song.
⭐ The practical test is short: before generating anything, name which verb applies. If the answer is preservation, the model's job is repair and its output should contain nothing that was not in the source. If it is offering, nothing may be attributed. If it is new authorship, a living person signs it as theirs. If it is invention, no continuity claim may be made at all.
A framing that recurs in this debate treats restoration and generation as competing options: restore the old recordings, or generate new ones. That framing is wrong on the facts of how the work is done.
Restoration is the training pipeline. The same processes that clean a degraded recording — denoising, source separation, spectral repair, resynthesis of damaged passages — are the processes that produce a clean corpus, and a clean corpus is the input to any voice model. There is no fork in the road where an institution chooses one branch. The restoration path arrives at generative capability whether or not anyone intended it, which has an uncomfortable implication that this paper states rather than hides:
⚠️ Anyone who restores this catalogue to a high standard has thereby built most of what is needed to forge it. The capability is a byproduct of the conservation, not an additional project someone must decide to start. This is precisely why the discipline has to be published in advance and why an institution that intends only to restore is not thereby exempt from these rules.
⭐ It also reframes the ordering: restore first and completely, publish the restorations, and let generation be a strictly downstream and separately-governed act. An institution that generates before it has restored has skipped the part that benefits everyone in favour of the part that benefits it.
Two structural claims, and the first restores a guard this corpus applies elsewhere.
Claim 3 — authorship is a judged seat and must never be occupied by the system. A useful distinction separates seats whose output can be checked by recomputation — determined seats, where anyone with the same inputs gets the same result and no discretion exists — from judged seats, which require a decision that could have gone otherwise and for which somebody must be answerable. A model's contribution to a dedicated work sits in the determined class: it is a transformation, checkable, and it decides nothing about meaning. Authorship is irreducibly judged, and it therefore cannot be held by a system that cannot be answerable.
This is the old reciter's guard restated for machines. In the tradition that carried this culture's texts for two millennia, a bhāṇaka recites the canon with extraordinary fidelity and is never confused with its author; the fidelity of the transmission is exactly why the distinction is maintained rather than a reason to relax it. A model is a reciter. A reciter is not an author, however good the recitation.
Claim 4 — every released work carries a named human signer. Not an institution, not a project, not a collective. A person, identifiable, who is accountable for the decision to release this particular work and for the claim that it is an offering rather than an attribution. The signature is the mechanism that keeps §5's judged seat occupied by somebody.
⚠️ A named signer is not a shield and should not be sold as one — it does not make a bad release good. It makes a bad release attributable, which is the minimum condition for a discipline to have any teeth at all.
If a reader takes one mechanism from this paper, it should be this one, because it is the load-bearing constraint and the easiest to quietly drop.
The training and selection objective admits a REFUSAL signal and excludes every ECONOMIC signal.
Concretely: the system may be told no — this output is wrong, this is disrespectful, this is not what the tradition does, this crosses a line, do not produce things like this. That signal may shape the model. And the system may not be optimized on plays, streams, engagement, retention, revenue, chart position, virality, or any proxy for them, at any point, including indirectly through a selection process that quietly prefers what performed.
The reason is not squeamishness about commerce. It is that the two objectives produce different artifacts and the difference is invisible in any single output.
⭐ This is the anti-pattern claim of the paper, and it is stated as a prohibition on an objective rather than on a behaviour deliberately, because behaviours can be justified case by case and an objective cannot be smuggled past a reader who is looking at it.
⚠️ Honest exposure: the prohibition is unenforceable from outside. Nobody inspecting outputs can determine what a model was optimized on. It binds an institution that adopts it and does nothing to one that does not — which is an argument for publishing it as a norm that can be demanded of others, not for believing it will hold on its own.
The strongest technical objection to claim 5 is that it describes something that cannot be built. Training and selection require a signal that says more like this; a system that can only be told no has nothing to climb, converges on nothing, and produces output too poor to release. On this reading the refusal-only objective is not a constraint but a description of a system that does not work.
The objection is partly right, and the part it is right about is smaller than it looks.
What it correctly identifies: some positive signal is unavoidable. A model must be fit to something — the surviving corpus — and fitting is a positive objective. Nothing in claim 5 forbids that, and it should not be read as forbidding it.
What the claim actually excludes is a narrower and more specific class: signals derived from audience response to the generated output. Plays, saves, skips, completion rate, chart position, revenue, retention, and any selection procedure that quietly prefers the take that performed better. The distinction is between fidelity to the source, which is a fixed historical target that cannot drift, and performance with an audience, which is a moving target belonging to the present.
⭐ Stated that way the claim survives the objection and becomes sharper: fit to the dead is permitted; fit to the living market is not. The first has a ground truth that a genocide fixed in 1975 and that no optimization can move. The second is exactly the gradient that would bend a murdered artist's voice toward whatever a contemporary platform rewards, one imperceptible increment at a time.
⚠️ The residual difficulty is real and unresolved: human curation leaks the excluded signal. A person choosing which of forty takes to release is applying taste, and taste is substantially formed by what performs. The prohibition binds the machine objective cleanly and binds the human one not at all. We have no mechanism for this beyond the named signer of claim 4 and the requirement that selection criteria be stated in advance rather than discovered afterward.
Claim 6 — marking scales with fidelity. Because success is the danger (§2), provenance marking cannot be treated as a formality that a good-enough output outgrows. The marking obligation increases as the output becomes harder to distinguish, and it is permanent, in the anastylosis sense: not a disclosure at the point of first publication but a property that travels with the artifact wherever it goes. The specific technical means are outside this paper's scope by the withholding rule at the head; the requirement is not.
Claim 7 — never fill a real gap. Where a recording is lost, the lost recording stays lost. Where a verse is missing, the verse is missing. Where a film reel was destroyed, no generated footage occupies its place in the archive.
The temptation here is the strongest one in the whole discipline, because filling gaps is what the technology is for and because the result would be genuinely beautiful and genuinely consoling. It is refused for the reason Panh refused reenactment: a filled gap is indistinguishable from a survival, and the absence is itself part of the record. The holes in this canon are evidence of what was done to it. An archive that closes them has erased the shape of the crime while appearing to repair it.
⭐ The positive form: the gap is displayed as a gap. A catalogue that shows its missing entries is telling the truth about what happened, and it is doing something no reconstruction can do — it renders the loss legible to someone who was not there.
Claim 8 — value routes as dedication, not as royalty. Where a dedicated work receives support, that value flows to living heirs, to living practitioners of the tradition, and to the restoration of the originals — not to a rights-holder position that would have to be manufactured in order to receive it. Inventing an estate to pay would be a second falsification layered on the first: it would create, retroactively, exactly the party whose absence defines this case. The routing form is dedication, which is the same operator as the attribution form of §3, applied to money instead of to credit.
Claim 9 — perpetual heir revocability. Where a descendant or heir can be identified, they may withdraw consent for their ancestor's voice at any time, permanently, without cause, and without a term limit. This is deliberately stronger than any licensing regime, and it is not symmetric: their permission is not required to begin (because in most cases there is no one to ask), but their refusal is sufficient to stop. ⚠️ The asymmetry is uncomfortable and is stated rather than smoothed: it means work may exist before anyone who could object has been found. The mitigation is that the objection, when it comes, is absolute and retroactive.
Claim 10 — the weights are never published. The corpus this institution otherwise releases into the public domain stops here, and the exception is deliberate. A voice model's existence is the risk; its use under discipline is not. Publishing weights that reproduce a murdered artist's voice would hand that capability, permanently and irrevocably, to every party that has none of the constraints above — and no licence term survives contact with a downloaded file.
⚠️ This paper therefore contains an admitted internal contradiction and does not resolve it. An institution committed to open publication is withholding the one artifact that matters most, on the grounds that openness here transfers a weapon rather than knowledge. That is either the correct exception or the beginning of the rationalization by which every openness commitment dies. We do not have a principled boundary that admits this exception and excludes the next one, and a reader who finds that unsatisfying is not making an error.
A distinct objection, and the one most likely to be raised by the people with the most standing to raise it: new work dedicated to the murdered artists competes with living Khmer musicians for the same attention and the same money.
The objection is strong and it is not primarily about ethics. A small music economy has a finite audience. Work in the voice of a beloved lost artist has an enormous inherent advantage over work by a living unknown — it arrives pre-loved, carrying national memory, and it needs no introduction. A dedicatory catalogue could plausibly crowd out the living tradition it claims to honour, and it would do so while being praised for its reverence.
Three responses, and the third is the only structural one.
The routing already points at the living (claim 8): value flows to living heirs, living practitioners, and the restoration of originals. That is a real mitigation of the money half and does nothing about the attention half.
The four verbs already assign the growth path to the living (§3.1): story is new authorship under a living name, and invention is explicitly new. The dedicatory branch is the one branch that is deliberately not the future of the tradition.
⭐ And the structural commitment, which is the one worth holding to: the dedicatory catalogue must be bounded and the living catalogue must not. A discipline that permits unlimited generation in dead voices while living musicians compete for the remainder has inverted its own purpose regardless of how the proceeds are routed. A bound is the only mechanism that makes the priority real, and stating that a bound is required without specifying it is the honest position while the product is unbuilt.
⚠️ We note without comfort that the authors are not the right party to set that bound, and that the parties who are — Cambodian musicians, and the audience — have not been asked.
Every claim above presumes somebody applies it. That party is unspecified, and the omission is not an oversight.
The obvious candidates each fail in a different way. The institution doing the work cannot be the arbiter of its own restraint; that is the auditee auditing itself. A national body would be the right answer in a country with an intact cultural-heritage apparatus and a settled relationship between the state and its own recent history, which is a strong assumption to make here. An international conservation authority has the expertise and does not have standing, and importing an external adjudicator into a heritage question is a move with its own long and unhappy history in this region.
⭐ What can be said is what the seat must be shaped like. It is a judged seat (§5) — it decides things that could have gone otherwise, so it must be occupied by identifiable people who are answerable. It must include descendants and living practitioners with real veto power rather than consultative status, since claim 9's revocability is meaningless if the party who could exercise it has no standing to be heard. And it must be able to stop the practice entirely, not merely to adjust it, or it is a body for calibrating an activity rather than for governing one.
⚠️ This paper does not constitute that body and should not be read as proposing itself for the role. It states the rules; it does not claim the authority to apply them, and an institution that published the rules and then appointed itself their judge would have written a licence in the shape of an ethic.
A dedicated work does not stay where it was published. It is ingested by streaming services, syndicated, embedded, excerpted, re-uploaded, and eventually stripped of everything that is not audio. Any discipline that lives only at the point of first publication has governed the least important moment in the artifact's life.
Three requirements follow, stated as obligations rather than implementations:
⭐ The general form, which applies well beyond this case: a provenance rule that depends on the good behaviour of downstream intermediaries is not a rule, it is a preference. The test of any marking scheme is what survives an adversarial re-upload, and a scheme that has not been tested against that has not been tested.
n = 0. Nothing described here has been built. No model has been trained, no work released, no marking scheme deployed, no heir consulted.
The pipeline is withheld (see the head note), so this paper cannot be used to implement the discipline it describes — only to hold an implementer to it.
⚠️ The voice-model leak is unsolved and it is the largest hole in the paper. Every claim above governs an institution that follows them. None of them prevents a model from escaping — through a breach, an insider, a subpoena, a bankruptcy, an acquisition, or the ordinary decay of an organization's discipline over decades. The discipline is a commitment about conduct and it has no answer for custody failure, which is the failure most likely to actually occur. Claim 10 reduces the exposure and does not eliminate it.
The named signer scales badly. One accountable person per released work is workable at small volume and becomes a bottleneck or a rubber stamp at large volume, and a rubber stamp is worse than nothing because it manufactures the appearance of accountability.
Heir identification is largely impossible here, for exactly the reason the case is special: the civil records were destroyed. Claim 9's revocability is real where a descendant can be found and is inert where none can be, which is most of the time.
The by/for boundary may not survive contact with ordinary listeners. §3's erosion risk is stated as a prediction in §10 because we genuinely do not know whether a careful attribution discipline survives a decade of casual reference, and there is a plausible reading in which it cannot.
The institution behind this paper has an interest. It intends to do this work. A discipline published by the party that wants permission to proceed should be read with that in mind, and the correct posture for a reader is that this document is an offer to be held to something, not evidence of virtue.
And the LLC gates the archive. The entity that would hold the restored material does not yet exist, and publishing an archive before it does would be premature on grounds that have nothing to do with this paper's ethics.
Registered before any of the work is done, so that they can be checked against outcomes rather than reinterpreted after them.
P-B1 — the discipline is adoptable. If dedicatory generation is a real category rather than one institution's scruple, at least one unrelated party working on an interrupted canon adopts the by/for boundary, or an equivalent, without contact with this corpus. If nobody does, the category was a private preference.
P-B2 — restoration outpaces generation. In an institution following §4, restored originals released will exceed dedicated new works released, cumulatively, at every point. If generation outpaces restoration, the pipeline argument has inverted in practice and the institution has become what it said it was preventing.
⭐ P-B3 — the erosion tripwire, and it is the one that matters. Public reference to dedicated works will drift from for to by. Sampled at intervals, the proportion of third-party references — press, platform metadata, listener discussion — that describe a dedicated work as by the deceased artist is the measurement. The prediction is that this proportion rises over time regardless of the institution's own care, because the shortening is a property of ordinary language and not of anyone's intent.
⚠️ If P-B3 fires, the correct response is not better copy. It is that attribution discipline alone is insufficient and the marking of §7 must carry the load that language could not — or that the practice should stop. A prediction registered in advance is what makes stopping a possible outcome rather than an admission.
All eleven are dedicated to the public domain. None will be asserted by this institution against anyone, including against parties who adopt them badly.
The reason to write this now is not that the work is ready. It is that somebody is going to do this to some interrupted canon fairly soon, and the discourse they will inherit was built for a case with a living estate on the other side of the table.
Being first to do the work is not the same as setting a precedent. A precedent has to exist as a document — something a later party can be pointed at, argued with, and held to. That is the whole of what this paper is trying to be, and it is why it is published under a licence that lets anyone use it, including people we would rather not.
The discipline reduces to a sentence, and the sentence came from a filmmaker who worked it out with clay long before any of this was possible: you may make something to hold the grief, and you may not make something that could be mistaken for what was lost.
The songs are closed. The catalogue ends where it ended, in 1975, and nothing will be added to it. What can be added is work made by living people, in that tradition, offered to the ones who did not get to finish — signed with the living name, marked so no one is deceived, and dedicated in the ordinary way that the culture in question has dedicated things to its dead for two thousand years.
That is a smaller claim than restoring a golden age. It is also the only one that is true.
Authored by Thon Ly with Miss Aquarius℠. Dedicated to the public domain under CC0 1.0 Universal. Corrections are welcome, and objections from Cambodian artists, scholars, and descendants are welcome most of all — this is a document about a heritage the authors did not make and do not own.