core/docs/decisions/ADR-0073a-anchor-lens-content-phase.md

ADR-0073a — Anchor lens content phase (Plan Phase L1.1)

Status: Accepted Date: 2026-05-19 Ratified: 2026-05-19 Author: Shay Phase: Plan Phase L1.1 (content prerequisites — pack enrichment, no code) Parent: ADR-0073 (umbrella anchor-lens architecture)

---

Context

ADR-0073 identifies three honest gaps blocking anchor-lens implementation, the first being distinction-bearing breadth absent from cognition-tier packs. Greek's interesting move is splitting one English concept into multiple lemmas; the cognition packs carry one lemma per concept. Without those distinctions on disk, no later phase (L1.2–L1.4) can render "knowing-as-experience" vs "knowing-as-system" deterministically — the substrate atoms simply do not exist to pivot on.

L1.1 fixes this with pure content enrichment. No runtime code, no composer, no realizer, no test of behaviour — only:

• lexicon.jsonl additions
• matching morphology.jsonl additions
• alignment.jsonl creation on the cognition-tier packs (currently only the micro packs carry alignment)
• manifest.json checksum refresh + re-verification

This is the highest-leverage step in the L1 sequence because it unblocks every later phase without code risk.

---

Decision

Greek additions (grc_logos_cognition_v1)

Nine new lemmas covering three distinction-bearing families that English collapses:

Knowledge family (English collapses to knowledge):

entry_id	lemma	distinguishing atom
grc-core-cog-021	ἐπιστήμη	logos.episteme.systematic_knowledge
grc-core-cog-022	σύνεσις	logos.synesis.insight

(γνῶσις at grc-core-cog-007 retains logos.epignosis.knowledge — the L1.3 lens reads it as the experiential variant.)

Love family (English collapses to love):

entry_id	lemma	distinguishing atom
grc-core-cog-023	ἀγάπη	logos.agape.covenant_love
grc-core-cog-024	φιλία	logos.philia.companion_love
grc-core-cog-025	ἔρως	logos.eros.passionate_love
grc-core-cog-026	στοργή	logos.storge.familial_love

Time family (English collapses to time):

entry_id	lemma	distinguishing atom
grc-core-cog-027	αἰών	logos.aion.age_era
grc-core-cog-028	χρόνος	logos.chronos.clock_time
grc-core-cog-029	καιρός	logos.kairos.opportune_moment

Hebrew additions (he_core_cognition_v1)

Three new lemmas from Hebrew's load-bearing covenant / wholeness / righteousness distinctions:

entry_id	lemma	distinguishing atom
he-core-cog-021	חסד	logos.chesed.covenant_loyalty
he-core-cog-022	שלום	logos.shalom.wholeness_peace
he-core-cog-023	צדק	logos.tzedek.right_order

Alignment.jsonl on both cognition-tier packs

The cognition-tier packs (grc_logos_cognition_v1, he_core_cognition_v1) gain a companion alignment.jsonl mirroring the micro packs' format. Edges fall in three categories:

• cross_lang.<atom> — three-way alignment on the shared logos.* atoms where all three substrates have the lemma (word / truth / light / life / beginning / pneuma / sophia).
• cross_lang.<family> — within-family alignment where two substrates have distinguishing lemmas but the third collapses or lacks the family (e.g. ἀγάπη ↔ חסד on the covenant-love axis; English has no covenant-love lemma).
• cross_lang.no_english_collapse — annotation edges marking families English does not split (love, time). These are metadata: target_id is en-collapse-<family>, a sentinel pointing at no real lexicon entry, with weight=0.0 and an evidence_ids list naming the collapsed English term.

Manifest checksum refresh

Per CLAUDE.md doctrine:

checksum = hashlib.sha256(Path(lexicon_path).read_bytes()).hexdigest()

grc_logos_cognition_v1/manifest.json and he_core_cognition_v1/manifest.json are updated with the new checksums of their post-enrichment lexicon.jsonl files. python -m language_packs verify <pack_id> is the canonical gate.

---

What L1.1 deliberately does NOT do

• No new AnchorLens class. That's L1.2.
• No composer wiring. Composers continue to render English by default.
• No --anchor-lens CLI flag. That's L1.4.
• No teaching corpus in non-English. Teaching chains in grc/he are a later phase (L2+).
• No modification of existing lemma atoms. The 20 grc + 20 he existing entries are untouched so downstream tests / composers / teaching chains keep referencing the same atoms they always have. Only new lemmas carry the distinguishing atoms.

These deferrals are deliberate: L1.1 keeps the diff to pure JSONL/JSON, so the substrate is ratified independently of any runtime change.

---

Trust boundary

L1.1 touches user-influenceable content (pack files) but the gate remains the same as for every other language pack: manifest checksum + python -m language_packs verify. The new entries are authored by hand here, not ingested from an external source.

No dynamic imports, no filesystem traversal, no shell passthrough.

---

Verification

python -m language_packs verify grc_logos_cognition_v1   → OK
python -m language_packs verify he_core_cognition_v1     → OK
python -m language_packs list                            → both packs listed,
                                                            entry counts 29 / 23
python -m core.cli test --suite packs -q                 → green
python -m core.cli eval cognition                        → public 100/100/91.7/100
                                                            (byte-identical;
                                                             new lemmas not yet
                                                             consumed by any
                                                             composer)

The cognition eval byte-identity holds because the new lemmas sit on disk but no composer references them yet — composers will start consuming them in L1.3.

---

What this unlocks

L1.2 (AnchorLens class + loader + default_unanchored_v1 sentinel) can now start. Without L1.1's substrate, L1.2 would have nothing to lens onto: the unanchored-vs-anchored distinction would be academic because both branches would render the same single lemma. L1.1 makes anchor lens possible; the later phases make it operator-visible.