These are not fables. A fable closes — it hands you a moral and sends you home. These do not close, and there is no lesson waiting at the end to reward you for finishing. They are an Anishinaabe ontology carried as story, because story is the form the knowing keeps so it doesn’t lose what it carries when it travels. Read them as quaint tales with a takeaway and you’ve read the wrapper and recycled the gift. Trickster hermeneutics is the entry requirement, not an advanced elective: meaning is held aloft, not brought down. The text is public. Understanding is not a download — it lives in community, by relation. Come in to wonder, not to extract.
The Indigeneering Commons
Built by Maya Chacaby (Odehamik) · MythOS substrate two and its operators. Anishinaabe, Opaaganising (Red Rock Indian Band).
This is the home for the Indigeneering Anishinaabemowin Project— engineering carried back to the land its genius was born on — built, against all advice, on the colonizer’s own machine, and pointed at the one thing the machine was built never to do: decline to know.
It is the technical and creative-technical work: the Indigeneering Operating Manual (Chaga in the Machine), the method, the pipeline, the logs, the lexicon, the morpheme work — reclaiming Anishinaabemowin the apocalypse paved over, from a wounded old archive, refusing to guess.
Built for
Indigenous engineers. Language-reclamation warriors. People building Indigenous LLMs that speak their language. You did not come here to be given a copy of a machine. You came to see that it can be built — so you can grow your own, in your own wound, in your own community, by relation. Inoculation, not distribution.
What it took (by the numbers)
The corpus — what was restored
| 142 | four-layer stories built (78 Part II · 63 Part I · one parallel reading) — the whole of Jones’s Ojibwa Texts, both volumes, complete. Every story wears the word draft: the DRAFT tag is the line-review tracker, so “built” is not yet “finished.” |
| 7,471 | sections — the working unit: one Jones passage, re-typed, restored, translated, glossed |
| 84,312 | Layer 1 word tokens — Jones’s 1919 orthography, re-typed by eye from the page scans. Never OCR. |
| 94,495 | Layer 2 word tokens — the Fiero restoration. This is the number that matters most: living-orthography Anishinaabemowin that did not exist in readable form before this work. |
| 211,625 | Layer 3 word tokens — Jones’s 1919 English, verbatim, as a historical artifact. Not corrected, not modernized. |
| 271,918 | Layer 4 word tokens — the gloss, written phrase-by-phrase against the restored line |
| 662,350 | word tokens across all four layers |
| 178,807 | of those are Anishinaabemowin (Layers 1 + 2) |
The lexicon, the grammar, the rules
| 21,724 | verified Jones↔Fiero word pairs |
| 16,207 | unique Anishinaabemowin headwords |
| 992 | morpheme ledger entries — obviation, tense, valence, initial change, each tied to an attested example and a source |
| 419 | variants kept as the no-standard-orthography record — variation is data, never noise |
| 71 | evidence-derived character-substitution rules |
| 1,232 | recurring-novel pairs flagged for a linguist’s eye (714 Part II · 518 Part I) — the human-review queue. The machine hands the uncertain up; this number is flag-never-guess made visible, not a backlog. |
Counted from the lodge on 2026-08-18, file by file. These grow as the work does — a moving count, not a finish line; flag-never-guess applies here too, so nothing still in human review is counted as done.
Every one of those was built under a single fastener: flag, never guess. A flagged uncertainty is a finished state, not a placeholder. The machine that confidently fills a gap puts a wrong word in a learner’s mouth, at machine scale, in a language with few enough speakers left that the wrong thing becomes the remembered thing. So it doesn’t.
Where the work is open — human eyes wanted
The machine did not finish the language; it built the room where humans finish it. These are the piles still waiting for eyes — and part of why the door below exists: vetted relations don’t only receive the data, they sit down at this table.
| count | review queue |
|---|---|
| 714 | priority recurring-novel pairs — the real eyes-on set |
| 1,230 | inflected forms of attested lemmas — confirm the inflection |
| 150 | vowel-length variants — confirm the spelling |
| 102 | pairings that need a human look |
| 54 | multi-sense words — which sense? |
| 2,484 | held single-attestation rare words — the deep pile (915 Part II · 1,569 Part I) |
These two piles together held 10,735 words in June. A machine pass tested every held word against every corroboration source the system had grown — the other part’s pile, the live lexicon, the crosswalk, and page co-occurrence across all 142 records — and graduated what the evidence could actually reach. What is left is not a backlog: it is a census of rarity, words that stand once in the whole surviving corpus. No amount of machine cleverness graduates those. They wait on human eyes, or on more Anishinaabemowin that does not exist yet.
How this Commons is built (and what it refuses)
What’s here, in the open — the method. The Operating Manual, the pipeline write-ups, the decisions, the engineering, the footnotes, the numbers above — and an independent, adversarial audit of the whole thing. The blueprint travels freely on purpose: a blueprint in your hands is not a copy of this work, it’s the start of yours. → How this was built & audited
What’s held — the living language data itself (the verified lexicon, the morpheme ledger as a queryable dataset etc.) is not on this site and not in this repository. A public file is a scraped file. The data moves only by relation, through the door below. → Request access
What’s never explained — the ceremonial. Where the stories carry sacred work, it is named and left to the listener. The wondering is not ours to resolve for you.
Under whose law
CARE Principles for Indigenous Data Governance · OCAP® · TK & BC Labels (in process). The relationship is carried on every page, not buried in one byline. → Use & sovereignty
In process: all raw data sent back to the communities whose ancestors told the stories.
Honest caveat, never hidden: nothing fully public is truly un-scrapable by technical means alone. The protection here is not a wall — it is publishing only the method, keeping the living data off the web entirely, the labels doing the relational-legal work, and the refusal being load-bearing. We do not promise the impossible. We make the taking cost more than it’s worth.
Miigwech for reading the parts that are here. The rest grows in relation, on its own season.