The research programme, 1977–2024

Status: machine-drafted, unreviewed. This narrative was written from data/theses/timelines.json, which is itself built on the occurrence→proposition mapping — a single uncalibrated LLM pass. Nothing here has been checked against the books. Where the data is ambiguous or the reading is contestable, it says so in the text rather than in a footnote. Treat every count as what the mapping believes, not as a fact about Hirsch.

The framing is Lakatos's, taken from the plan's §5.1: a research programme has a hard core its author will not give up, a belt of auxiliary hypotheses added to meet particular anomalies, a record of responses to critics, and a justification that can shift underneath a proposition that never changes its wording.

How to read the counts

Two different measures appear throughout, and they disagree in interesting places.

• Stated (same / narrower / broader, stance hirsch_asserts) — Hirsch asserts the proposition at or near its own scope. This is the measure the timeline's first_stated uses.
• Supported (supports) — an occurrence offered as a reason for the proposition. supports is 43% of all mappings and the loosest relation in the scheme; a proposition with many supports and few statements is one the corpus surrounds but rarely voices.

The distinction matters most at the edges. P16 is stated in three books but only ever restated as itself in 1977. P08 is stated from 1987 but its earliest same is 1996.

(a) The hard core

Applied structurally — first stated in 1987 (Cultural Literacy) or earlier, still stated in one of the last two books — 19 of the 25 propositions qualify. That is itself the first finding, and it cuts two ways:

1. Hirsch's programme really is unusually static. He publishes the same argument for thirty-seven years, and the late books are re-argued rather than revised.
2. Or same is too generous, and a looser paraphrase is being counted as a restatement. The calibration sample exists precisely to settle this and has not been rated yet.

The strongest members — first stated in 1977 or 1987, stated in at least nine of the ten books, and never contradicted by a concession that gives up the core:

Four more meet the structural test but with weaker evidence of continuous assertion: P13 (ideas over funding/demography — stated in 7 books, absent from HtEC 2020 and AE 2022), P25 (the "whose knowledge?" objection is answerable — 8 books but only 11 same in total, and its first same is 2006 rather than 1987), P19 (the functional criterion — see §e), and P04 (decoding versus comprehension — 8 books but only 13 same across the whole corpus; it is a distinction Hirsch uses far more than he asserts).

The centre of gravity is 1987. Of the nineteen propositions that meet the structural test, seventeen have their first genuine restatement (same) in Cultural Literacy. Only three propositions in the entire list are restated as themselves in 1977 — P02 (writers leave shared knowledge unstated), P16 (relative readability) and P20 (the grapholect) — and all three are about writing and language rather than schooling. The Philosophy of Composition supplies the linguistics; the education argument arrives whole in 1987.

(b) Auxiliary hypotheses: what was added, when, and against what

Four propositions enter after 1987. Each answers a specific occasion, and three of the four are methodological or evidential rather than substantive — they are additions to the protective belt, not to the core.

Two further additions are reframings inside existing propositions rather than new propositions, and the plan's list treats them as auxiliaries:

• Ethnicity-as-shared-knowledge (2022). P22 is stated from 1987 (cl:chpreface_C18, "Americans are different from Germans… because each group possesses specifically different cultural knowledge"). What 2022 adds is the term: "American ethnicity" explicitly names the shared literate language taught in schools (ae:ch8_C2). A vocabulary change on a thirty-five-year-old claim, not a new hypothesis.
• The ratchet (2024). P05's cumulative-learning claim is broadened into cultural evolution (re:ch1_C89), and P17's anti-developmentalism is given an evolutionary warrant (re:ch1_C15). Both events carry new sources that appear once, in 2024 only: S1373 (Tennie, cumulative culture) and S0692 (Henrich, The Secret of Our Success).

The France/Sweden case is where the programme is most obviously Lakatosian. It enters in 2016 as an auxiliary that turns a negative result into confirmation, and by 2022 it has been promoted: ae:ch6_C4 calls it "the most decisive educational experiment in history". The French data itself (S0004, French National Soundings 1987–2007) had already been in the corpus since 2006 — the evidence precedes the hypothesis by a decade.

(c) Anomalies and critics the corpus shows him answering

The corpus records 892 objections, spread across every book (1996 is the peak, with 204). Its serious limitation for adjudication is that almost none of them are named: the most common attributed source is "unattributed" (30), then "Progressive educators" (8). Named critics appear only sporadically (Dewey, Bourdieu, Huntington, Rothstein, Monty Neill, Ravitch). Whatever the adjudication layer needs, it will not get named critics from this extraction.

Five threads are nevertheless legible.

1. The 1987 "whose knowledge?" backlash → the equity reframing, and a rebrand. Forty-two objections across the corpus turn on elitism, canon, Eurocentrism or indoctrination, and they start immediately: cl:ch1_OBJ14 ("providing children with traditional information indoctrinates them in a conservative point of view"), cl:ch4_OBJ7 ("literate national culture is inherently elitist"). Two responses are visible in the data:

• A rebrand. In 1996 Hirsch records that the term "Cultural Literacy" "raises too many extraneous questions and controversies", and renames both foundation and movement to Core Knowledge (swn:ch1_OBJ9). This is the single clearest programme-level concession to critics in the whole corpus, and it is a concession about packaging.
• A justification switch. P23 (equity requires commonality) is reframed in 1996 from general literacy to a direct argument that privileged children get the knowledge at home and only school can supply it to the rest (swn:ch4_C265), with developmentalism recast as the mechanism of class injustice (swn:ch4_C132). The equity argument is present in 1987 (cl:chpreface_C4) but becomes the primary warrant after the backlash. The same move appears in P20, where "standard grammar is white speech" is answered with "it is standard educated speech, necessary for economic success" (swn:ch1_OBJ11).

The thread does not close. In 2024 Hirsch is still answering it, now with Bourdieu named: refusing transmission is "the true violence", because the wealthy survive the shipwreck privately (re:chappendix-i_OBJ11).

2. Testing → consequential validity. P14 begins in 2006 as a fairness complaint and becomes, in 2016, a technical claim about validity theory backed by an outside source (S0370, Sireci, Validity Theory and Test Validation) and by item-level analysis of PARCC. The 2024 move is a narrowing: the demand that tests be curriculum-based is restricted to the early grades (re:ch6_C63).

3. Common Core. Hirsch's position is not simple opposition. In 2016 he argues Common Core "can only succeed if its knowledge-centric statements are prioritised over its skills-based ones" (wkm:ch6_C160) and concedes that the content vacuum comes from standards-makers avoiding political controversy, not from test-makers alone (wkm:ch1_OBJ5). By 2024 the vagueness is reframed as structurally inevitable — a "political necessity" produced by the ratchet (re:ch6_C37).

4. The charter-effect objection to the lottery study. This is the cleanest objection→reply→counter-objection chain in the corpus, and it runs across three books:

• 2010 — "the success of these students is due to the schools being charter schools rather than the curriculum" (the-making-of-americans:ch5_OBJ9). Reply: the same results appear in "unheralded" regular public schools using the same curriculum.
• 2022 — two selection objections. Creaming (ae:ch1_OBJ9): answered by New York's blind-lottery mandate. Motivated parents (ae:ch1_OBJ10): answered by the Grissmer Colorado Core Knowledge lottery study, in which both arms came from families motivated enough to apply.
• 2024 — differential attrition (re:chappendix-iii_OBJ2): answered by excluding high-attrition lotteries and controlling for redshirting.

The dependency worth flagging: the Grissmer study (S0005) enters in 2016 and appears in every book after, 38 occurrences in all — and the registry flags it ckf_affiliated. The single strongest causal evidence in the late programme is not independent of the programme.

5. The demographic anomaly. The decline-caused-by-demography objection recurs from 1996 to 2024 and gets a different answer each time: the 64% drop in high scorers cannot be a composition effect (swn:ch2_OBJ17, 1996); Jencks's Iowa data show decline in a demographically stable population (S1047, 2010); French scores fell among executives' and professionals' children too (wkm:chprologue_OBJ2, 2016); the decline began before desegregation (re:chpreface_OBJ3, 2024). Alongside it sits the one large concession on the historical claim: there was no golden age, only a "silver age" (wkm:chprologue_OBJ3, 2016).

(d) How the justification shifts

The per-proposition justification_shift records are machine-drafted, and the shape they converge on is consistent enough to be worth stating — and consistent enough to be worth suspecting, since a model asked twenty-five times whether a warrant shifted will tend to find that it did.

Across the propositions, the recurring arc is:

cognitive mechanism (1977–1996) → equity and economic mobility (1996–2010) → civic cohesion (2010–2020) → national identity and evolutionary necessity (2020–2024)

Instances, with the propositions that carry them:

• Cognitive mechanism. P02's warrant in 1977 is Miller's memory limits and chunking; P01's in 1987 is de Groot's chess studies and Hirsch's own Grant-and-Lee experiment; P17's is the logical refutation of the acorn metaphor.
• Equity and mobility. From 1996: P23 becomes an argument about what privileged homes supply (swn:ch4_C265); P20 shifts from "a classless society needs a trans-local language" (1977) to the minority earnings gap; P07 acquires Hart & Risley's word gap (2006, S0031).
• Civic cohesion. From 2010: P21 moves from individual enfranchisement to preventing political fragmentation, and by 2020 is carried by Gallup polling on patriotism and secession (S0051). P18's aim of schooling becomes a civic duty (the-making-of-americans:ch1_C139).
• National identity, then biology. From 2020: P18's register turns communal and biological — "induction into the tribe" (ae:ch9_C29); P17 becomes a neocortical blank-slate claim (S0083, Kalisman) and then an evolutionary one (S0692, Henrich); P22 names the object "American ethnicity".

Two counter-cases keep this honest. P12's warrant moves in a different direction entirely — from "conflicting ideologies" (1977) through institutional history (Teachers College, 1996) to a theological register: providentialism, education schools as "sectarian churches" (2010). And P16's justification moves backwards into the laboratory: from a universal law of least effort (1977) to processing-speed measurements in Hirsch's own UVa experiments (S0316, 2024).

The register shift is the best-evidenced part of this section: of 39 reframing events, 18 are tagged register and 13 justification, against 7 scope and 1 evidence_base. What changes over the decades is mostly how Hirsch talks, and only secondarily what he claims.

(e) What dropped out

P16 — relative readability (1977). The clearest casualty. Stated 18 times in The Philosophy of Composition with 180 supporting occurrences, and then never restated as itself again: 0 same in every subsequent book, three narrower occurrences in 1987, one in 2024. Its foundational sources go too — Spencer's Philosophy of Style (S0118) is dropped after 1987. The proposition is drafted as hard core; on this evidence it is a first-book thesis Hirsch abandoned as a stated position while keeping its conclusion (prose quality is objective) as a background assumption. In 2024 it returns in a different form — subsumed into expertise, justified by the efficiency of expert rewrites (re:chappendix-ii_C33).

P19 — the functional criterion (1987). Stated 9 times in Cultural Literacy and never again at that volume: 1 in 2006, 3 in 2010, 2 in 2016, 2 in 2020, 1 in 2022, 0 in 2023 and 2024. This is the proposition that says the content list is empirically discoverable and revisable rather than a canon — the answer to the canon-war charge. It peaks exactly where it is needed and fades once the fight moves on. The 2022 event is a straightforward retraction: the original list was "unwisely prescriptive" about science, listing what people should know rather than what they do (ae:ch9_C50, ae:ch9_C52).

P06 — vocabulary as proxy (2006). Stated in KD (2006), MoA (2010) and WKM (2016) and in no book after. Support occurrences continue at a trickle (6, 5, 3 in the last three books) but the claim stops being asserted. Its evidence goes with it: Sticht's San Diego study (S0377) is dropped after 2016.

Evidence dropped, proposition kept — thirteen such events, of which these matter most:

The Hart & Risley case is the one to look at first: it is the single most-cited empirical prop under the achievement-gap proposition, it disappears after 2016, and 2016–2019 is exactly when the replication challenges to it were published. The corpus cannot tell you whether Hirsch dropped it because of that. It can tell you he stopped using it and did not weaken the claim it had been supporting.

A caution on this whole section. "Dropped" here means dropped for this proposition. Four of the thirteen drop events describe a source that is still cited in later books for something else, and the machine-written descriptions of those events over-generalise to "no longer cited". The tables and the site both carry the deterministic correction; the event sentences do not always.

What the data contradicted in the drafted proposition list

Eight of twenty-five drafted first_book values disagree with the mapping. On inspection of the quotes, they are not all the same kind of error.

Real corrections — the drafter guessed late, and the claim is genuinely there earlier:

P21 and P22 matter most: they move the civic and national-identity propositions out of 2010 and into 1987, which weakens any reading on which Hirsch "turned political" late.

A correction in the other direction: P14 is drafted to SWN 1996 but the mapping finds no statement of it before KD 2006 (kd:ch6_C21). The test-invalidity claim is post-NCLB, not pre-.

Probable mis-maps, where the earlier "statement" is thinner than the proposition: P12 is credited to PoC 1977 on the strength of "the lack of direction in our teaching and research… we are beset by conflicting ideologies" (poc:chintroduction_C5, relation narrower) — which is a very long way from Rousseau-through-Dewey-through-Teachers-College. Its earliest same is CL 1987, so the drafted SWN 1996 is still one book late. P05 is credited to PoC 1977 on a broader occurrence about practical arts (poc:ch5_C115); its earliest same is CL 1987, which is what was drafted.

Contradictions of the drafted hard_core flag:

• P16 is drafted hard_core: true. The data does not support it: stated as itself only in 1977. Recommend hard_core: false, or a note that it is a retained assumption rather than a retained assertion.
• P19 is drafted hard_core: true and is stated in seven books, but at collapsing volume, with an explicit partial retraction in 2022. Worth a second look.
• P21, P22, P24, P25 are drafted as auxiliary but are all stated from 1987 onward in eight or nine books. On the structural test they behave like core, not belt. If they are auxiliary in the Lakatosian sense it must be because they are surrenderable, which is a judgement the corpus cannot make.
• P06, drafted auxiliary, is the only proposition of the twenty-five that actually behaves like an abandoned auxiliary. On this evidence the auxiliary/core split is doing much less work than the framing suggests.

What this cannot tell you

• Whether the propositions are true. Nothing here is an assessment.
• Whether an absence is Hirsch's or the extraction's. Books differ in extraction density, and a proposition can look "dropped" because a chapter was thinly extracted. Every event that rests on absence — evidence_dropped most of all — inherits that risk, which is why those events carry lower confidence scores.
• Which critic he is answering. The objections are extracted but overwhelmingly unattributed; the adjudication layer needs named critics acquired from outside the corpus.
• Whether a reframing is development or paraphrase. 39 reframings across 25 propositions, from a model asked to look for change, is exactly the number to be sceptical of. The rule in the prompt was "do not invent development where the text merely repeats"; whether it held is a question for review, not for the pipeline.